Automated Design of 3D Printed Biomedical Products with Use of Knowledge-Based Engineering (KBE)

Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 1 BIOMEDIX Erasmus+ strategic partnership for Higher Education BIOMEDICAL INNOVATIONS THROUGH DIGITAL TRANSFORMATION Chapter 2: Automated Design of 3D Printed Biomedical Products with Use of Knowledge-Based Engineering (KBE) Project Title Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange KA220-HED-AA8A896B Output Development and Publication of an e-Book on Biomedical Innovations through Digital Transformation Module Chapter 2 Automated Design of 3D Printed Biomedical Products with Use of Knowledge-Based Engineering (KBE) Date of Delivery 30.06.2025 Authors Filip GÓRSKI, Magdalena ŻUKOWSKA, Natalia WIERZBICKA, Radosław WICHNIAREK, Dan Sorin COMSA, Emilia SMOLAREK Version V1 This chapter is accompanied with an Augmented Reality application, for Android devices (cellphones and tablets). Use the attached QR code to download it. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 2 Contents 1 Introduction ............................................................................................................................. 3 2 Automated Design of Biomedical Products ............................................................................. 4 2.1 Anatomically Customized Products ............................................................................. 4 2.2 Basic Design Methodologies ....................................................................................... 7 Parametric and non-parametric modelling ............................................................ 7 Modelling of anatomical shapes ............................................................................. 9 Designing prosthetic and orthotic devices ........................................................... 11 Designing implants and operative models............................................................ 13 2.3 KBE Principles ............................................................................................................ 15 2.4 Automated Design Workflow .................................................................................... 21 Intelligent CAD Models ......................................................................................... 21 Automated Design Systems .................................................................................. 24 Transitioning Between Traditional and Automated Workflow ............................ 27 2.5 Finite Element Analysis (FEA) of Automatically Designed Orthopedic Products ...... 29 State of the art in the field of using the finite element method (FEM) for analyzing orthopedic products ......................................................................................... 29 Finite element analysis of a therapeutic orthosis manufactured by 3D-printing . 39 3 Tools and Techniques ............................................................................................................ 48 3.1 CAD Systems with KBE Capabilities ........................................................................... 48 3.2 Design Automation Case Studies............................................................................... 52 AutoMedPrint: A System for Orthopedic Device Design ...................................... 52 Automated Design of Orthoses ............................................................................ 54 Automated Design of Prostheses ......................................................................... 57 3.3 Automation in Additive Manufacturing Workflows .................................................. 61 3.4 Scaling Up: Mass Customization of Biomedical Products ......................................... 65 4 Conclusions ............................................................................................................................ 69 4.1 Challenges and Future Trends ................................................................................... 69 4.2 Summary and Recommendations ............................................................................. 70 References ...................................................................................................................................... 73 Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 3 1 Introduction The accelerating fusion of biomedical engineering and digital technologies has fundamentally reshaped the development of medical products, especially those requiring anatomical customization. Traditional design workflows, while effective in generic applications, increasingly fall short in addressing the needs of personalized healthcare. Emerging demands for precision, reproducibility, and rapid delivery in patient-specific treatment call for a paradigm shift in how medical devices are conceived, designed, and manufactured. This chapter explores how such a shift can be achieved through the integration of Knowledge-Based Engineering (KBE) and additive manufacturing. KBE represents a transformative approach that embeds expert knowledge, design logic, and decision-making processes into intelligent digital systems. By formalizing and automating complex biomedical design workflows, KBE allows for scalable customization—transforming the production of orthotic devices, prosthetics, implants, and surgical aids from artisanal tasks into repeatable, knowledge-driven engineering processes. Combined with the layer-by-layer freedom of 3D printing, this methodology enables efficient creation of anatomically individualized solutions that match the morphological and functional nuances of each patient. This chapter delves into the principles and applications of Knowledge-Based Engineering (KBE) in automating the design of 3D printed biomedical products. It focuses on methodologies, tools, and practical workflows that leverage engineering knowledge to streamline the design process and ensure efficiency, accuracy, and reproducibility. The main objectives of the chapter are to explore various design methodologies in biomedical design, especially using KBE, showcase applications and highlight synergy between KBE, design automation, mass customization and additive manufacturing. By the end of this chapter, readers will gain insight into the principles and practice of biomedical design automation, the technological enablers behind scalable mass customization, and the strategic value of knowledge formalization in next-generation healthcare engineering. In doing so, the chapter underscores a central tenet of modern biomedical innovation: that automation, when grounded in deep domain expertise, can reconcile individuality with industrial efficiency. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 4 2 Automated Design of Biomedical Products 2.1 Anatomically Customized Products In recent years, the field of biomedical product design has undergone a significant transformation, driven by the convergence of digital technologies, patient-specific data acquisition, and advanced manufacturing processes - chief among them, additive manufacturing. At the heart of this transformation lies the concept of anatomical customization - the practice of tailoring medical products to match the unique morphological and functional characteristics of individual patients. These products, often built layer by layer in 3D printing processes, mark a paradigm shift from traditional, one-size-fits-all solutions toward highly individualized medical interventions (Górski 2025). Anatomically customized products, frequently described under the broader umbrellas of customization, personalization, or individualization, are medical devices or tools that incorporate precise anatomical data of a patient into their geometry, structure, or function. Unlike standard prosthetics or implants, which are fabricated in generic sizes and shapes, anatomically tailored solutions are born from the digital representation of a specific human body. This distinction, though seemingly subtle at first, dramatically alters the entire design and production workflow, infusing it with complexity, ethical considerations, and regulatory challenges (Górski 2025). To understand the scope and impact of anatomically customized products, it is crucial to explore their defining principles and contexts of use. At the most fundamental level, anatomical customization leverages advanced imaging and scanning technologies, such as CT, MRI, or 3D surface scanning to capture accurate representations of a patient’s anatomical region of interest. These datasets are then transformed into 3D models using specialized software, serving as the basis for the design of products that must integrate seamlessly with the patient’s body, whether externally (e.g., orthoses) or internally (e.g., joint implants). Such individualized design is often a medical necessity. A joint endoprosthesis, for instance, may require precise matching to a patient’s bone geometry to ensure biomechanical compatibility, reduce the risk of implant loosening, and optimize load distribution. Similarly, craniofacial implants for trauma or oncology patients must adhere to the intricate topography of a patient’s skull to restore symmetry, function, and aesthetic appearance. In these and many other scenarios, anatomical customization ensures that the medical product does not merely serve its purpose, it does so as if it were naturally integrated into the biological system it is supporting (Górski 2025). The benefits of anatomical customization are manifold. Firstly, it significantly enhances comfort and functionality. Devices shaped to an individual’s anatomy are more ergonomic, reducing the likelihood of pressure points, misfits, or mechanical failure. Secondly, it supports better clinical outcomes. Implants designed to fit precisely are associated with lower rates of revision surgery and shorter recovery times. Thirdly, it enables innovations in treatment Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 5 strategies: products that would be impossible or prohibitively expensive to produce in a traditional subtractive manufacturing process become feasible when digital design and additive manufacturing are employed. However, the shift from standardized to individualized product development introduces a host of new challenges. The anatomical data used as the foundation for design must be captured with exceptional precision and clarity. This necessitates high-quality imaging, reliable segmentation, and expert interpretation. Equally, the design process must integrate biomechanical and clinical insights to ensure that the final product is not just an accurate replica of a body part, but a functional substitute or enhancement thereof. The material selection must also account for biocompatibility, sterilizability, and structural integrity, tailored not just to the general use-case, but to the specific anatomical and functional context. From a process standpoint, anatomical customization imposes significant demands on workflow standardization and repeatability. Traditional design workflows are optimized for batch production; the customized route requires workflows that are agile, modular, and often automated to remain cost-effective. Knowledge-Based Engineering (KBE), the subject of this book’s core methodology, plays a critical role here - offering tools to encapsulate expert knowledge and design intent into reusable logic that can be rapidly adapted to a new patient's dataset. This approach addresses one of the core dilemmas of anatomical customization: how to combine uniqueness in shape with efficiency in process. From a regulatory perspective, anatomically customized medical products also inhabit a gray zone. On the one hand, they are products intended for medical use, often classified as Class II or III devices, demanding rigorous safety and efficacy validation. On the other, their individualized nature often places them outside the conventional pipeline of product certification. This can result in a paradox: the product is needed urgently and is likely superior to generic alternatives, yet its route to clinical use is burdened by legal and bureaucratic constraints. This issue is particularly acute in cases involving implants for tumor patients or trauma victims, where timing is critical and no off-the-shelf solution exists. These regulatory hurdles are compounded by issues of data traceability, documentation, and reproducibility. Each custom-designed product must be fully documented - not just in terms of its design rationale and material properties, but in terms of its specific manufacturing steps, quality control checkpoints, and usage conditions. The requirement for traceability becomes even more pronounced in the case of Class III implants, where any adverse event can have significant legal and medical consequences. Despite these challenges, anatomically customized products continue to gain traction, particularly as healthcare systems increasingly recognize the value of patient-specific treatments. Furthermore, the broader trend toward personalized medicine, driven by genomic data and AI-assisted diagnostics, supports the rationale for integrating anatomical customization into the medical product development ecosystem. In many respects, anatomically tailored products represent the material embodiment of the personalized medicine paradigm. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 6 To clarify the nuanced terminology used in this domain, the following distinctions are helpful (Górski 2025). Customization generally refers to deliberate adjustments made with some degree of user or clinician input - such as the choice of a material, mounting mechanism, or external configuration. Individualization, on the other hand, refers to the adaptation of the product’s geometry or functionality to suit a specific patient, often without direct user interaction in the design. It is this concept - especially anatomical individualization - that anchors the methodologies described in this book. Finally, personalization refers to aesthetic or superficial modifications that do not affect the product’s primary function but may improve patient acceptance and comfort, such as adding colors, markings, or decorative elements to devices like orthoses or wheelchairs. Figure 2.1. Customization types of medical products While all three types – customization, individualization, and personalization - may coexist in a single product, the anatomical individualization process is the most technically and clinically demanding. It draws directly from anatomical data, integrates medical expertise, and leverages advanced design methodologies. This convergence of biology, engineering, and computing is emblematic of the shift toward anatomically aware design - a design philosophy that prioritizes not only functionality and aesthetics but also the seamless integration of a product with the biological environment in which it operates. Anatomically customized products offer the potential to revolutionize how medical care is delivered - making it more precise, effective, and patient-centered. However, realizing this potential demands new tools, new thinking, and a robust framework for design automation. It is within this context that Knowledge-Based Engineering provides an indispensable solution, as it allows designers to embed anatomical logic, clinical rules, and expert know-how directly into digital design environments - creating systems that can generate individualized products swiftly, reliably, and safely. The next sections of this chapter will delve into the technical principles, software environments, and methodological structures required to build such systems. But it is the anatomical individuality of each patient that remains the anchor – defining both the problem and the promise of biomedical product design in the era of digital medicine. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 7 2.2 Basic Design Methodologies Parametric and non-parametric modelling In the design of anatomically customized biomedical products, Computer-Aided Design (CAD) modelling methods must account for complex, often irregular geometries derived from patient-specific anatomical data. The two primary approaches to 3D modelling are parametric modelling and non-parametric modelling, each with distinct methodologies, advantages, and limitations in terms of geometric control, reproducibility, and suitability for automation. Before discussing the modelling paradigms, it is important to distinguish between basic CAD model types. The most commonly used in biomedical product development include: • solid models – volumetric representations with mass and material properties, typically used in mechanical design (Cucos et al. 2018). • surface models – shell-like structures defined by boundary surfaces without volume, useful for complex freeform geometries, especially in automotive branch (Harries et al. 2019). • mesh models – composed of connected polygons (typically triangles), often used for scanned anatomical data. • wireframe models – skeletal representations consisting of edges and vertices; used primarily for reference and layout. • voxel models – volumetric representations using discrete spatial elements; primarily used in simulations and medical imaging. Each of these model types (Fig. 2.2) can be created using either parametric or nonparametric approaches, although the compatibility and effectiveness of each approach vary by application. Figure 2.2. Model types Parametric modelling is based on the definition of geometry through dimensional parameters, constraints, and a sequential set of operations. The resulting model has a history tree, where each feature is stored and can be edited retroactively. The modelling process is typically composed of: Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 8 • sketch-based features – creation of profiles (confined in 2D or 3D space) followed by operations such as extrusion, revolution, lofting, or sweeping; • no-sketch features: operations like filleting, chamfering, shelling, and patterning that do not require sketches; • transformation features: mirroring, translating, rotating, or scaling features and bodies; • boolean operations: union, subtraction, and intersection of solid bodies (Górski 2025). Parametric modelling is widely adopted in mechanical and medical device engineering due to its ability to support design iteration, reuse, and controlled variability. However, when applied to highly complex or organic anatomical geometries, parametric modelling may introduce challenges. Excessive parameterization can lead to instability, particularly when geometry changes invalidate downstream operations, resulting in failed model regeneration (Górski 2025). Non-parametric modelling (Fig. 2.3) does not rely on a feature history or parameter sets. Instead, it is based on direct geometry manipulation, often in the form of “digital sculpting.” (Alcaide-Marzal et al. 2013). This is typical for mesh-based or surface-based models derived from medical imaging or 3D scanning. Key characteristics include: • no model history or ordered feature tree, • geometry is modified through direct deformation, displacement, or smoothing operations, • local modifications do not propagate across the entire model unless manually applied, • common in tools supporting mesh editing (e.g., Blender, Meshmixer) or direct modelling (e.g., Siemens NX, Autodesk Fusion’s direct mode). Figure 2.3. Non-parametric modelling (digital sculpting) – stages of work Non-parametric methods offer higher flexibility when working with anatomical shapes but are more difficult to standardize or automate. The absence of consistent parameters makes them less suited for workflows requiring reproducibility or batch design generation. Comparison of the two approaches is presented in Table 2.1. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 9 Table 2.1. Comparison of parametric and non-parametric approaches, based on (Górski 2025) Feature Parametric Modelling Non-Parametric Modelling Geometry control High (through constraints and parameters) Local, intuitive but less precise History tracking Full feature history (editable) No global history; local modifications only Suitability for automation High (supports rule-based design, templates) Low (manual, hard to script or replicate) Compatibility with anatomical data Moderate (requires fitting anatomical data to clean geometry) High (can operate directly on scanned data) Performance with complex geometry May become unstable or slow with complexity Efficient, real-time manipulation possible Output consistency High Variable, depends on user input Integration in engineering workflows Excellent (mechanical CAD integration) Limited; often used in early concept or visualizations Modelling of anatomical shapes Modelling anatomical shapes for biomedical product design presents distinct challenges not encountered in standard mechanical engineering. Anatomical geometry is inherently complex, irregular, and often patient-specific. Efficient modelling requires proper strategies for incorporating anatomical data into CAD workflows, maintaining reproducibility, and ensuring compatibility with regulatory standards. The modelling process begins with the acquisition of anatomical data, typically in one of the following forms (Górski 2025): • Anthropometric measurements (from atlases or manual tools), • 2D medical images (X-ray, ultrasound, photographs), • 3D meshes obtained via scanning or segmentation of medical imaging data (CT, MRI), • Extracted geometric features (e.g. curves, planes, landmarks) derived from 3D meshes. Among these, 3D triangle meshes are most commonly used in the design of customized medical products. These meshes can be manipulated directly or used as a reference for generating parametric geometry. Two primary modelling strategies are applied: the meshbased approach and the parametric approach, with intermediate hybrid workflows also in use. The mesh-based approach uses anatomical meshes as the foundation for product design. This method can be subdivided into (Górski 2025): • pure mesh pathway: the entire model, including modifications and devices, is created using mesh-editing tools (e.g., cut, offset, bridge), without transitioning to solids or surfaces. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 16 and scalability. As customization becomes the norm rather than the exception, especially in patient-specific medical applications, the traditional trial-and-error or manually guided design approaches become insufficient. This is where Knowledge-Based Engineering (KBE) emerges as a crucial paradigm, facilitating the embedding of expert knowledge directly into design processes through computational systems (Zawadzki 2018, Górski et al. 2016). In engineering, knowledge is more than raw data or information - it represents an organized structure of relationships, rules, procedures, and heuristics built on experience and validated practice. It is contextual, actionable, and forms the foundation for decision-making across all stages of product development. To distinguish clearly (Fig. 2.9): • Data are raw, unprocessed facts (e.g., scanner output, numerical measurements). • Information is data interpreted within context (e.g., identified dimensions of a residual limb). • Knowledge is the ability to apply that information appropriately, often codified as rules, procedures, or parametric logic (e.g., if residual limb length is below threshold X, use configuration Y). In biomedical product development, knowledge arises from multiple disciplines - engineering, medicine, material science, and manufacturing. Therefore, its effective capture and representation are central to automation and customization. Fig. 2.9. Data, information and knowledge within context of medical product design Knowledge in engineering is diverse in form and function, and its classification is crucial for effective capture, representation, and automation. Within the context of biomedical engineering - especially the design of individualized medical devices - different types of knowledge play distinct roles, from geometric rules embedded in parametric CAD models to clinical decisions made intuitively by experienced specialists. Understanding the nature of this knowledge is a prerequisite for building robust and scalable Knowledge-Based Engineering (KBE) systems. The below considerations are based on authors’ knowledge, as well as on the book (Rzydzik 2013). 1. Declarative Knowledge Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 17 Declarative knowledge refers to static facts and rules about a specific domain. It describes what is, without necessarily indicating how to use that information. In the context of engineering design, declarative knowledge includes: • Anatomical dimensions and tolerances, • Material properties, • Logical constraints (e.g., "screw hole must not intersect nerve canal"), • Clinical conditions and their definitions (e.g., “spasticity involves increased muscle tone”). This form of knowledge is often expressed as parameter tables, design constraints, engineering handbooks, or clinical guidelines. Declarative knowledge is relatively easy to encode in databases or knowledge rules and is central to the construction of CAD templates and design verification routines. 2. Procedural Knowledge Procedural knowledge consists of methods, sequences, and operations - it defines how to do something. In design engineering, it includes: • The steps for generating a 3D model from a mesh, • Manufacturing process plans (e.g., print orientation, support strategy), • Simulation workflows (e.g., setting boundary conditions for FEA), • Post-processing operations or validation steps. Procedural knowledge is essential for automation because it can be formalized into logic trees, process chains, scripts, or CAD macros. In KBE systems, procedural knowledge often controls the execution of design sequences or conditional branching in response to input parameters. 3. Episodic Knowledge Episodic knowledge refers to contextual or time-dependent knowledge - knowledge of events, experiences, or specific cases. It is derived from real-world projects and past occurrences and may include: • A surgeon’s recall of a particularly difficult implant placement, • A record of previous design revisions and associated outcomes, • Case-based reasoning examples (e.g., “in a similar case, model B caused skin irritation”). While episodic knowledge is less structured than declarative or procedural types, it is critical for learning from experience and improving system performance. In engineering contexts, it forms the basis of historical databases, feedback systems, and design reuse strategies. 4. Ontological Knowledge Ontological knowledge concerns the categorization and relationships within a domain. It defines how different entities relate to one another, which is vital for structuring knowledge systematically. Examples in biomedical design include: • Classification of implants by body region (e.g., maxillofacial vs. orthopedic), Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 18 • Relations between anatomical structures (e.g., “the radius is connected to the ulna”), • Grouping of orthotic types (e.g., passive vs. active). Ontologies are used to create semantic networks and knowledge graphs, helping systems “understand” the domain context and improving knowledge retrieval, interoperability, and reasoning. In complex KBE systems, ontological structures support rule hierarchies and decision-making logic. Fig. 2.10. Types of knowledge Complementary to the structural classification (Fig. 2.10), engineering knowledge can also be distinguished by its degree of formality and accessibility - that is, whether it is explicit or tacit (Zawadzki 2018). 1. Explicit Knowledge Explicit knowledge is formalized, codified, and easily communicated. It is accessible in written form and includes: • Engineering drawings and CAD models, • Design rules and checklists, • Medical imaging protocols and segmentation procedures, • Parameterized templates and spreadsheets. This form of knowledge is the easiest to integrate into KBE systems. For example, a rule stating “if socket diameter > 100 mm, use three support ribs” can be programmed directly into a CAD automation script. Explicit knowledge is also shareable and transferable between engineers, clinicians, and systems. 2. Tacit Knowledge Tacit knowledge is experiential, intuitive, and difficult to articulate. It resides in the minds of experts and is acquired through practice rather than formal instruction. In the biomedical context, examples include: Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 19 • A prosthetist’s skill in adjusting a socket by feel, • A surgeon’s experience-based decisions in choosing incision points, • An engineer’s intuition about where stress concentrations may occur. Tacit knowledge is essential but harder to systematize. Capturing it often involves: • Interviews with experts, • Observation of skilled tasks, • Use of collaborative workshops or design thinking sessions. Although challenging to encode, efforts to extract tacit knowledge are vital for transitioning from manual to automated or semi-automated processes, particularly in patientspecific applications. These two classifications - structural (declarative, procedural, episodic, ontological) and functional (explicit vs. tacit) - are not mutually exclusive. Instead, they intersect and complement each other. For instance: • A procedural workflow (explicit knowledge) may have evolved from accumulated episodic experiences (tacit knowledge). • An ontological structure (explicit) may be informed by the expert’s intuitive understanding of anatomical relationships (tacit). In KBE system development, it is crucial to: • Start with explicit, declarative and procedural knowledge (easier to formalize), • Gradually incorporate episodic knowledge (e.g., via databases or case logs), • Use interviews and field studies to externalize and formalize tacit knowledge, • Introduce ontological frameworks for consistency and logic structuring. By combining these knowledge types within the system’s architecture - typically in the form of a rule base, parameter maps, and decision logic - design engineers can achieve automation of complex, customizable workflows such as those found in the development of orthoses, implants, and surgical models. The process of integrating engineering knowledge into design systems involves several key stages – identification, acquisition and representation (Fig. 2.11), according to earlier work (Zawadzki 2018, Górski et al. 2016). 1. Identification Knowledge sources must first be recognized. In the context of biomedical engineering, they may include: • Expert interviews (clinicians, prosthetists, engineers), • Existing product documentation, • Simulation or test results, • Literature and clinical studies, • Observed outcomes from previous projects. 2. Acquisition This phase involves extracting and recording the knowledge using methods such as: • Manual documentation (e.g., decision trees, flowcharts), Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 20 • Case-based reasoning (CBR), • Reverse engineering from existing products, • Data mining and AI techniques for pattern recognition. 3. Representation Knowledge is stored in structured formats within a Knowledge Base, which may include: • Rule-based logic (e.g., IF limb length < X THEN use configuration A), • Semantic networks or ontologies, • Parameterized CAD templates, • Tables of decision criteria or configuration options, • Annotated images or diagrams. Effective knowledge representation enables reusability, validation, and automatic application in KBE systems. Fig. 2.11. Integrating knowledge into engineering systems The application of KBE in biomedical engineering brings measurable benefits, especially when dealing with large numbers of customized products that share a common design logic. Key advantages include: • design cycle time reduction: through automation of repetitive tasks such as fitting, template generation, and documentation, design time per patient can be reduced from days to minutes. • minimization of human error: by relying on validated rules and structured logic, errors related to oversight, inexperience, or miscommunication are significantly decreased. • rapid iteration and updates: changes in design criteria or regulatory constraints can be quickly implemented across all templates and configurations. • efficient documentation generation: bill of materials, technical drawings, and CAM files can be generated directly from the knowledge base, ensuring consistency. For instance, in orthosis or implant design, where certain geometrical rules repeat across cases, a well-developed KBE system can generate a patient-adapted CAD model from a limited input dataset (e.g., scanned geometry and anthropometric attributes), bypassing the need for a fully manual modelling session. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 21 Scalability is a critical concern in the shift from traditional, hand-crafted devices to digitally manufactured, personalized products. KBE systems enable mass customization by combining rule-driven automation with flexible parametrization. In practice, a KBE-based workflow for a prosthetic socket or surgical guide may operate as follows: 1. Input: patient-specific data (e.g., limb scan, age, diagnosis). 2. Processing: automated application of rules and parameters based on clinical and design knowledge. 3. Output: ready-to-manufacture CAD model, technical drawings, and documentation. Because the underlying rules are reusable and generalizable, this process can be repeated for hundreds of unique patients with minimal manual intervention. Such scalability is vital for: • clinics serving large populations, • manufacturers providing customizable devices, • research centers developing libraries of case-adaptable models. KBE is particularly well-suited for anatomically driven design, where geometry is unique but logic and structure are repeated. It transforms the knowledge of experienced professionals into shareable, codified digital assets - ensuring that expertise is not lost, and product quality is consistent across large volumes. Knowledge-Based Engineering serves as the backbone for digital automation in the biomedical product design workflow. It allows for the codification of both structured knowledge (rules, procedures) and previously tacit expertise into intelligent design systems. In the next chapters, this conceptual foundation will be expanded with practical methods and implementation strategies in the context of biomedical engineering, including specific applications to orthotics, prosthetics, surgical guides, and implants. 2.4 Automated Design Workflow Intelligent CAD Models The development of anatomically customized biomedical devices increasingly relies on design workflows that are capable of efficiently handling high variability without compromising precision or repeatability. To meet this demand, modern engineering practice has introduced intelligent CAD models are highly structured, parametric, and logic-driven design artifacts capable of automatic geometry generation based on case-specific input (Skarka et al. 2023, Górski 2025). These models form the core of many Knowledge-Based Engineering (KBE) systems, supporting design automation for orthotic shells, prosthetic sockets, implants, surgical guides, and other patient-adapted solutions. An intelligent CAD model differs fundamentally from a static design. Rather than encoding a single shape or geometry, it represents an entire family of potential designs governed by embedded knowledge structures. These may include dimensional parameters, geometric constraints, logical conditions, and procedural rules. When provided with a new set of input values, such as anatomical measurements or diagnostic classifications, the model regenerates Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 22 a geometry variant that conforms to all necessary constraints and standards. This regeneration is deterministic and repeatable, offering a reliable alternative to manual redesign for each case. From a technical standpoint, the architecture of an intelligent model is built on three interrelated layers. The first is the parametric geometry, where each feature (such as curves, surfaces, or solid bodies) is defined in relation to a set of parameters rather than fixed dimensions. The second is the relational structure, consisting of constraints and dependencies that maintain internal consistency - ensuring, for instance, that a screw hole always lies a fixed distance from an edge, or that the wall thickness scales with the width of an orthotic frame. The third layer is the rule logic, implemented using conditional statements or externally linked control files. This layer enables model behavior to change dynamically based on input - activating or suppressing features, altering feature types, or restructuring the model workflow (Zawadzki 2018, Skarka et al. 2023, Górski 2025). The parameters defining a given variant of a model are usually bound together through the design table – a spreadsheet or hypertext file with a formalized structure, where parameters – and often their mathematical and logical relations – are stored and are being uploaded to the CAD model. Design table can be easily changed from the outside, either manually or automatically, via various systems of configurators and data extraction mechanisms. An example is shown in Fig. 2.12, describing a modular hand prosthesis model, fully presented in a previous publication (Górski et al. 2025). Fig. 2.12. Design table of a modular hand prosthesis, depicting relational structure between parametric geometry of its parts, as described in (Górski et al. 2025) The development of such models follows a structured methodology, which can be generalized into the following steps: Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 23 1. Concept definition and variability analysis. The first stage involves identifying the product family to be automated and analyzing the expected variability across cases. In medical design, this often includes anthropometric ranges, anatomical classifications, and modular options (e.g. size ranges, limb types, fixation methods). This step also involves selecting the reference geometry or design logic to be preserved across variants. 2. Parameter identification. Based on the variability analysis, the key design parameters are selected and formally defined. These can be geometric (lengths, radii, angles), logical (e.g. component presence or absence), or categorical (e.g. type of amputation or bone defect). Each parameter must be mapped to one or more features in the model and assigned limits, units, and dependencies. 3. Model construction in CAD environment. Using a parametric CAD system (e.g. Siemens NX, SolidWorks, Fusion 360), the model is constructed using sketch-based and feature-based operations, ensuring all geometry is fully constrained and linked to the predefined parameters. Care is taken to preserve model stability across the expected parameter range, and regeneration errors are minimized through robust constraint schemes. 4. Embedding logic and conditional behavior. Logic structures are introduced to allow dynamic changes in model topology. This may involve suppressing or enabling features based on dimensional thresholds or categorical choices. For example, an elbow orthosis may include a joint mechanism only if the selected variant covers the elbow area. Such behavior is typically implemented using if-else statements, design tables, or scripting (e.g. iLogic in Inventor, DriveWorks in SolidWorks). 5. External data integration. To support automation, the model is linked to external data sources such as spreadsheets, XML files, or databases. This allows non-CAD users (e.g. clinicians or technicians) to input patient-specific values into a form or data entry interface, which then drives the regeneration of the CAD model automatically, without opening the model file manually. 6. Validation and error handling. The final model is tested across its parameter space to ensure valid regeneration under all reasonable input scenarios. Boundary cases and error conditions are identified and mitigated through constraints, warnings, or fallback behaviors. Documentation is created to define parameter roles, dependencies, and usage protocols. This methodology is particularly effective in the biomedical field, where base product configurations often repeat, but patient anatomy introduces variability. For example, a transradial prosthetic socket may follow the same anatomical logic across cases but differ significantly in diameter, length, and support rib configuration. An intelligent CAD model allows these differences to be captured and generated automatically, improving design speed, consistency, and traceability. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 24 One of the key advantages of intelligent models is their compatibility with automation systems. Once properly structured, they can be operated by external interfaces, scripts, or web-based platforms, allowing models to be generated on demand directly from clinical data or 3D scans. This opens the possibility of integrating intelligent CAD models into semiautomated design pipelines for in-clinic customization or industrial-scale individualized manufacturing. However, several technical limitations must be acknowledged. Intelligent models are sensitive to poor parameter management, and small changes in logic or input can lead to topological errors, especially in complex geometries. Moreover, adapting this methodology to organic shapes - such as facial anatomy or irregular bone surfaces - requires hybrid solutions, where scanned mesh data is merged with parametric substructures. Despite these challenges, the intelligent CAD model remains one of the most effective tools for implementing scalable customization in biomedical design workflows. Automated Design Systems As the demand for individualized biomedical products grows - driven by clinical need, patient expectations, and technological feasibility - manual design workflows quickly become unsustainable. The combination of high variability, medical precision requirements, and timesensitive delivery calls for structured systems capable of consistently generating valid, patientspecific designs. Automated design systems provide this capability, functioning as softwaredriven frameworks that transform anatomical or clinical data into manufacturing-ready digital models with minimal human intervention. An automated design system integrates multiple technical layers: data input interfaces, rule-based decision logic, intelligent CAD modelling, and output generation modules (Zawadzki 2018). These components are orchestrated to execute a design process end-to-end, allowing clinicians, technicians, or automated servers to trigger geometry generation without engaging in traditional CAD modelling. The result is a streamlined, repeatable, and scalable solution for customized product development. At the core of such systems lies the principle of design knowledge formalization. Rules that were once implicitly followed by experienced designers - such as geometric adjustments for comfort, safe distances from anatomical landmarks, or standard placements for support features - must be codified in a form that can be processed algorithmically. Depending on system architecture, this codification may take the form of parameter tables, decision trees, script-based automation, or full-featured KBE logic modules embedded in CAD environments (Zawadzki 2018, Górski et al. 2016). A typical automated design system for a biomedical application begins with patientspecific data acquisition. This can include numerical values (e.g. lengths, diameters, weight), 3D scans (mesh or point cloud), or imaging-derived models (segmented STL files from CT/MRI data). The data is captured using structured forms, guided input interfaces, or direct device Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 25 integration. It is then validated against predefined criteria to ensure it falls within acceptable limits - both for patient safety and for system stability. Following validation, the system processes the data through a set of deterministic rules. These may include geometric calculations, conditional logic, and part configuration selection. For instance, a lower-limb orthosis design system may use leg length to scale the brace shell, foot width to define sole curvature, and a diagnosis code to select among different hinge mechanisms. Each rule is designed to trigger specific actions in the model, and multiple rules may operate simultaneously or hierarchically to handle complex interdependencies. The intelligent CAD model at the heart of the system is configured to accept these inputs, regenerate accordingly, and resolve all embedded features and relationships. Depending on the complexity of the design, the model may be opened and regenerated in a background CAD process or fully embedded in a headless automation framework. In both cases, the user is shielded from the geometric complexity of the design process, interacting only with the input and output stages. The output of the system typically includes one or more of the following: • A fully defined 3D CAD model in a neutral format (e.g. STEP, IGES), • A mesh file ready for additive manufacturing (e.g. STL, 3MF), • Technical drawings or documentation (PDF, DXF), • Associated metadata for tracking, labelling, or post-processing. Some systems also produce post-processing instructions (e.g. for support removal, part labeling), bills of materials, or automatically generated reports summarizing the design logic and patient-specific inputs used. A conceptual of basic operation of an automated design system for design of customized medical devices is shown in Fig. 2.13. Based on data input, rule-based decision logic governs the CAD modelling, using also configuration data input, and generates a mesh model as an output for additive manufacturing. Automated design systems may be implemented in various architectures depending on the workflow requirements. In clinic-based systems, the automation may run locally on a workstation, directly linked to scanning equipment and 3D printers. In industrial settings, systems are often cloud-based or server-controlled, allowing batch processing of multiple cases and integration with ERP or PLM systems. Web-based front-ends enable remote data entry and customer interaction, while backend automation engines execute model generation asynchronously. One of the critical success factors in implementing automated design systems is robustness. Unlike manual workflows, automated systems must operate without supervision, handling a wide range of input cases without errors. This places strong emphasis on input validation, model regeneration testing, and exception handling. If a patient’s anatomical data is outside the expected range, the system must respond appropriately - either by adjusting the model safely, flagging the case for manual review, or rejecting it with a meaningful explanation. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 32 cellular metamaterials. Plesec and Harih proposed the concept of a prosthetic liner with a gyroid-inspired lattice structure, aiming to enhance comfort by better distributing pressure (Plesec & Harih, 2024). The gyroid unit-cell was chosen for its isotropic mechanical properties and efficiency in load distribution. To model this material in FEA, the researchers first performed uniaxial compression tests on sample lattices and used the data to calibrate a custom multilinear material model. This calibrated FE model, when compared against a conventional solid silicone liner, showed the cellular liner could significantly reduce peak contact pressures at the limb-prosthesis interface while tailoring its stiffness to user needs. Notably, a softer lattice liner yielded much lower donning pressures (ideal for sensitive or sedentary users), whereas a stiffer lattice provided better support and stability during highimpact gait, all of which being captured by the FEM simulations. This is a clear example of how bio-inspired material architectures, combined with proper material modeling in FEA, are expanding the design space for prosthetic comfort and performance. Composite materials remain a hot topic. Beyond traditional carbon-fiber, natural fiberreinforced composites (NFRCs) are being investigated for sustainable, lightweight prosthetic components. Castro-Franco and his coworkers (Castro-Franco et al., 2024) note that NFRCs (using fibers like flax, hemp, etc.) have lower cost and environmental impact, and discuss their advantages and drawbacks in prosthetic design. Their review emphasizes that computational biomechanical models used in FE simulations are essential to evaluate the effectiveness of new materials in prosthetic designs. For instance, FEM can simulate how changing the fiber orientation or layering in a prosthetic socket affects its stiffness and the pressure on the residual limb. Overall, the trend is toward multifunctional materials (lightweight, strong, compliant where needed) and using FEM to capture their complex behavior. Whether it is a carbon-fiber foot keel or a 3D-printed nylon elbow orthosis, accurate input data (elastic moduli, yield strengths, damping characteristics, etc.) and advanced material models in FEM are enabling engineers to push the boundaries of design while ensuring safety. 3. Validation techniques (in vitro & in vivo) With the increasing complexity of FE models, rigorous validation has become more important than ever. Over the last five years, researchers have refined both in vitro (lab-based) and in vivo validation methods to ensure FEM predictions match real-world behavior. In vitro validation often involves mechanical testing of prosthetic/orthotic components under controlled conditions, then comparing the results to simulation. For example, to validate an FE model of a prosthetic foot, Balaramakrishnan and his coworkers (Balaramakrishnan et al., 2020) scanned a commercially available foot (Ottobock SACH foot), derived material properties through tensile and compression tests, and then measured the shape of the foot (effective curvature during stance) using a custom test rig. The FE model incorporated the same geometry and a hyperelastic material law, and remarkably, the difference between the simulated and experimental roll-over curvature was only about 7.5%. Such a small error demonstrated the accuracy of the model in replicating the behavior of the Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 33 foot, giving confidence that the FE model could be used to explore design tweaks virtually. In other cases, researchers use standard material testing machines to record force–displacement curves of an orthotic device (e.g., deflecting an AFO or twisting a knee brace) and then check that the FE simulation produces the same stiffness or deformation pattern. Strain gauge measurements on device structures are also common – for example, bonding gauges on a prosthetic pylon or brace rod to record strain under load, and verifying the strain output of the FE model at the same locations. This helps catch any discrepancy in material modeling or boundary conditions in the simulation. In vivo validation techniques have advanced, especially for aspects like interface pressure and fitness which are hard to replicate in bench tests. A notable method is using pressure sensors or pressure mats between the patient’s limb and the device. Recent socket studies placed thin resistive pressure sensors on the residual limb inside a prosthetic socket and measured the pressure distribution while the user stood or walked, then compared it to FE predictions (Shah & Rehman, 2025). In one case, the pressure predicted by the FE model at key points differed by only about 8.5 kPa from the sensor readings, confirming that the simulation closely captured reality (Shah & Rehman, 2025). Motion capture and gait analysis provide another layer of validation: researchers record a subject walking with the prosthesis/orthosis to get ground reaction forces, limb kinematics, and even muscle activation patterns. These data can drive the FE model (as boundary conditions or loading inputs), and the output of the model – such as the deformation of a prosthetic foot or the stress in a brace during gait – can be indirectly validated if the overall behavior (e.g., stride length, energy return) matches experimental data. In the realm of spinal orthoses (like scoliosis braces), imaging and detailed measurements are used for validation. Techniques such as digital image correlation (DIC) or laser-based methods (e.g., Electronic Speckle Pattern Interferometry) have been applied to braces to measure how much they deform or how much load they apply to the body (Guan et al., 2020). Guy and Aubin created a patient-specific FE model of a Boston scoliosis brace and then performed in vivo measurements of the pressure applied by the brace on the torso and the correction achieved in spine curvature (Guy and Aubin, 2023). The preliminary results showed good agreement with FE but also highlighted the need to account for patient breathing and muscle tone in the model for truly accurate predictions. Such mixed validation approaches (combining sensor data, imaging, and motion analysis) are becoming the gold standard to build realistic FE models. It is worth noting that validation is an iterative process. Discrepancies between FEM and experiment lead to model refinements – for example, updating material properties or contact definitions. Over the last five years, the community has also published guidelines and reviews on validation best practices. Al-Fakih and his coworkers (Al-Fakih et al., 2016) reviewed decades of prosthetic socket interface studies and catalogued sensor technologies for measuring stresses in sockets, providing a baseline for current researchers. More recently, standards for reporting FE validation like including error quantification, as done in the Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 34 prosthetic foot roll-over study (Balaramakrishnan et al., 2020) have improved the transparency and reliability of simulation in this field. 4. Clinical applications and case studies FEM advancements are not just theoretical – they are increasingly translating into clinical practice and device development. One major application is in the design of patient-specific devices. Custom prosthetic sockets are a prime example: clinicians now use 3D scans of an amputee’s limb to create a digital model, and FEM is employed to assess the socket fit and predict pressure points before manufacturing. Shah and Rehman (Shah & Rehman, 2025) generated a Total Surface Bearing socket from a CT scan and evaluated its biomechanical behavior under load with FEA, then fabricated it via additive manufacturing. The result was a well-fitted socket that exhibited a uniform stress distribution and closely matched pressure sensor data on the patient, demonstrating how FEM can streamline the fitting process. This approach reduces iterative physical adjustments, potentially cutting down the numerous clinic visits often needed to tweak socket fit in the first year after the amputation (Plesec & Harih, 2024). Another promising area is the use of FEM for optimization and personalization of orthoses. Sophisticated workflows now combine FE simulations with optimization algorithms to achieve tailored device designs. For instance, in scoliosis treatment, Kardash and her coworkers (Kardash et al., 2022) have developed personalized simulation models of a patient’s torso and an automatic design loop to optimize brace geometry. By making the FE model differentiable (so that the optimizer can efficiently see how design changes affect clinical outcomes), that study managed to computationally design braces that improved spinal alignment metrics by an average of 45% in simulation, while maintaining patient comfort within acceptable limits. Such patient-specific optimized braces were generated for multiple subjects, showing the feasibility of a semi-automated design process. While these braces still need to be physically tested on patients (currently the results are based on simulations), this indicates a future where orthotic devices could be algorithmically adapted to each patient’s anatomy and condition. FEM is also accelerating innovation in device concepts. In prosthetic feet, for example, designers are moving beyond static design charts to using FE models as virtual testbeds. Energy-storing prosthetic feet (which flex and return energy during walking) have benefitted from FEM-driven design – allowing engineers to try out new geometries or composite layups and immediately see the impact on deflection and energy return. Balaramakrishnan and his coworkers (Balaramakrishnan et al., 2020) replaced the traditional trial-and-error prototyping with a parametric FE model to design a multi-axial prosthetic foot. By adjusting parameters in simulation (ankle stiffness in various planes), they achieved an optimal balance that was later built and verified experimentally. Similarly, in orthoses, we see 3D-printed wrist splints and ankle braces whose lattice structures were refined via FEA to maximize ventilation and minimize weight, all while ensuring the support function is maintained (Plesec & Harih, 2024). Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 35 These case studies underscore an important point: FEM is enabling rapid innovation by serving as a bridge between concept and clinic. Finally, FEM is increasingly used in clinical decision-making and education. Clinicians can visualize stress distributions or potential problem areas in a patient-specific model, which helps in explaining to patients why a certain orthotic design is chosen (for example, showing where a scoliosis brace will apply pressure on the ribs). Some research teams have even created real-time or near-real-time FE analysis tools so that a clinician could, in principle, adjust design parameters on the computer and immediately see the effect on fitness or alignment. While not yet conventional, this integration of FEM into the clinical workflow is on the horizon as computational tools become faster and more user-friendly. 5. Combining artificial intelligence (AI) with FEM FEM combined with AI has seen significant advancements in the last years, enabling faster simulations, smarter designs, and more personalized orthotic and prosthetic devices. Below, we analyze state-of-the-art developments across key areas, with reference to recent academic work. a) Surrogate modeling with machine learning Data-driven surrogates are trained on FEM results to emulate detailed simulations at a fraction of the cost. For example, Kriging regression models have been fitted to prosthetic socket FE analyses, providing real-time pressure and strain predictions within milliseconds (Steer et al., 2020a). b) AI in optimization loops AI techniques are increasingly embedded in FEM-driven design optimization. Surrogates (ANNs, Kriging, etc.) act as rapid evaluators in iterative optimization algorithms (like genetic algorithms or particle swarm optimization) to find optimal device designs. This has been applied to topology and shape optimization of prosthetic components. For example, a multiobjective genetic algorithm was combined with a surrogate FEM model to generate personspecific prosthetic socket designs, spanning a spectrum from patellar-tendon-bearing to totalsurface-bearing shapes (Steer et al., 2020b). The AI-assisted optimizer produced a suite of candidate socket geometries that strategically redistribute pressure to improve comfort. Such methods drastically reduce manual trial-and-error in design. c) FEM applications enhanced by AI • Stress analysis In orthotic design, homogenization-based surrogates were used to evaluate lattice deflections much faster than explicit FE, enabling quicker stress-relief tuning for flatfoot insoles (Moeini et al., 2023). • Fatigue life and failure prediction Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 36 AI has improved predictions of long-term performance such as fatigue and stress shielding in prosthetic devices. FEM is traditionally used to evaluate fatigue life by simulating repeated loads (e.g., gait cycles), but this can be time-consuming. Machine learning now helps estimate fatigue performance directly from design parameters and single-run FE results. For example, in the design of semi-porous hip prostheses, researchers calculated safety factors and stresses via FEA for various porosity configurations, then trained ML models to predict those outputs for new designs instantly (Akkad et al., 2023). This approach can identify designs that minimize stress shielding in bone and extend implant life without exhaustive cyclic simulations. • Topology and shape optimization AI accelerates topology optimization and design refinement for custom prosthetic components. Traditional topology optimization iteratively calls FEM solvers to find an optimal material layout, but AI can guide or replace some of these iterations. Recent studies use AIassisted design to achieve lightweight yet strong prostheses. One approach combined patient gait data, multi-body dynamics, FEM, and machine learning to optimize the shape of a hip implant for each patient (Milone et al., 2024). The algorithm adapts the prosthesis geometry to the patient’s weight, height, and walking pattern, improving fatigue strength and reducing the stress peaks by up to 40% compared to the original design. In general, neural networks have been trained to predict optimal material distributions or to refine designs after a few initial FEM runs, dramatically speeding up the design cycle. While AI does not fully replace physics-based topology optimization, it serves as a powerful helper – suggesting near-optimal layouts or providing an excellent starting point for final FEA fine-tuning (Kulkarni et al., 2024). • Material behavior and constitutive modeling FEM accuracy depends on good material models for biological tissues and device materials. AI is used to develop more accurate material behavior predictions. For example, AI has been employed to predict effective properties of new 3D-printed orthotic materials (e.g. lattice infills) without exhaustive physical testing. These advances help engineers evaluate novel materials and composites, accelerating the adoption of high-performance materials in prostheses design (Milone et al., 2024). • Patient-specific modeling A major advantage of AI-enhanced FEM is the ability to handle patient variability efficiently. Statistical shape models and AI allow the creation of patient-specific FE models on the fly. Recent work combined principal component analysis (PCA) of limb shapes with a Kriging surrogate to account for anatomical variability in socket design (Steer et al., 2020a). The result was a surrogate that could take a new amputee’s residual limb shape and socket design parameters as input and instantly predict interface pressures and tissue strains, something that would normally require a lengthy FE setup for each patient. • Clinical applications of AI-FEM integration Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 37 A multi-objective AI optimizer generated a suite of “best fit” socket shapes for an amputee, letting the clinician choose an optimal starting design from data-driven options (Steer et al., 2020b). This reduces the iterative guesswork in achieving a comfortable fit. These AI-assisted design platforms act as decision support tools, helping clinicians and engineers explore a large design space quickly and base their choices on quantitative predictions rather than trial-anderror. The result is a faster design process and devices that are tailored to the patient’s needs from the outset. One of the most promising clinical benefits of AI-enhanced FEM is real-time or near-real-time simulation, which opens the door to interactive fitting and tuning of devices. Surrogate models can run nearly instantaneously, enabling real-time what-if analyses during patient consultations. For instance, Steer and his coworkers achieved socket interface pressure predictions in 1.6 milliseconds using a surrogate, essentially providing instant feedback on design adjustments (Steer et al., 2020a). This capability means a prosthetist could virtually adjust an orthosis (strap position, stiffness, etc.) and immediately see the predicted effect on pressure distribution or alignment, even during an appointment. In the future, realtime FEM predictions could be combined with motion capture or gait analysis: as a patient moves, an AI model could continuously estimate stresses or identify instabilities, allowing onthe-fly tuning of smart prosthetic components. Early steps toward this vision are seen in mechatronic twin frameworks, where a physical prosthesis tester (robotic platform) and a virtual FEM model run in parallel (Chen et al., 2022). Such setups have been used to train deep learning algorithms to recognize dynamic load patterns in sockets. Ultimately, this could lead to closed-loop systems where sensor data from a patient’s device is fed into an AI model (informed by FEM), which then suggests real-time adjustments to improve comfort or performance. Beyond design optimization, AI+FEM are being leveraged to predict clinical outcomes and guide patient care. Machine learning algorithms can mine simulation results and patient data to identify patterns that correlate with success or complications. For example, large datasets of prosthetic use (e.g. sensor readings from smart prostheses) can be cross-analyzed with FEM-based stress calculations to predict issues like socket pain or risk of skin breakdown. One interdisciplinary review highlighted the development of ML models that analyze prosthetic sensor data alongside other parameters to find trends in patient outcomes (Kulkarni et al., 2024). In practice, this means AI might detect that a certain pattern of pressure distribution (predicted by FEM and confirmed by sensors) often precedes a skin sore, thereby alerting clinicians to intervene earlier. AI-driven predictive models are also being explored for rehabilitation: using simulations of different alignment or component choices, algorithms could forecast a patient’s gait stability or energy cost and recommend the optimal configuration for that individual. This kind of personalized outcome prediction represents a step toward evidence-based, customized prosthetic care. As Kulkarni and his coworkers emphasized, by identifying patterns in big data from prosthetic usage, AI can enable truly personalized treatment plans tailored to each patient’s lifestyle and physiology (Kulkarni et al., 2024). The integration of AI and FEM strongly aligns with the goals of personalized medicine. Every patient’s anatomy and needs are unique, and AI-enhanced FEM makes it Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 38 feasible to account for those differences in device design and tuning. Patient-specific FE can be generated rapidly with AI assistance, as discussed. Clinically, this leads to custom orthoses/prostheses that are optimized for the individual rather than a one-size-fits-all approach. In the context of orthoses, for instance, an AI-FEM pipeline can take a 3D scan of a patient’s foot and automatically output a 3D-printable orthotic insole design that redistributes pressure according to that patient’s arch geometry and gait load profile. Such workflows are beginning to appear in research prototypes. The use of cloud computing further helps in personalization: large databases of prior patients’ FE models and outcomes can be stored, and AI can leverage this collective knowledge to design a new patient’s device with similar successful features (an analogy to case-based reasoning). Overall, AI is enabling a shift from manual customization to model-driven personalization, where data and simulations ensure each device is the best possible fit for that one patient. • Validation techniques and hybrid modeling A common validation step is comparing AI surrogate predictions with traditional FE results or experimental measurements. Many studies report the error margins of their AI models to establish credibility. For example, the surrogate transtibial socket model developed by Steer and his coworkers (Steer et al., 2020a) was validated against full FE solutions, showing pressure errors < 4 kPa and strain errors < 3%, which was deemed acceptable for clinical use. Ultimately, AI-enhanced FEM models must be proven in real-world scenarios. Researchers are increasingly designing protocols to compare AI-FEM predictions with real device performance on patients. For example, an instrumented prosthetic socket (with embedded pressure sensors) can provide ground truth data on load distribution, which is then compared to the model’s predictions for the same patient walking. Karamousadakis and his coworkers (Karamousadakis et al., 2021) developed a sensor-based system to monitor transfemoral socket pressures and used an FE model as a reference for validation. Validation efforts also focus on the reliability and safety of AI in healthcare contexts. One recognized concern is that AI models are only as good as their training data, and may exhibit bias or errors in scenarios not well represented in that data (Kulkarni et al., 2024). Researchers are thus careful to train on diverse datasets (e.g., multiple anatomies, load cases) and to quantify uncertainty in predictions. Some recent FEM-AI studies incorporate Bayesian methods to estimate confidence intervals for a predicted stress or life span, alerting users if a prediction is extrapolating beyond the known domain. Moreover, the community emphasizes that AI should assist, not override, engineering judgment – any critical design predicted by AI often undergoes a final FE verification or a safety factor buffer. By transparently reporting model accuracy and uncertainties, and by validating against both simulations and experiments, engineers aim to build trust in these AI-augmented tools. The goal is a validated hybrid framework where AI provides speed and insight, while FEM and physical testing ensure accuracy and safety. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 39 FEM is now commonly used in orthotic and prosthetic innovation. Computational modeling is now more powerful and patient-specific, materials from composites to metamaterials are being rigorously simulated, and validation efforts give confidence in these virtual models. FEM is actively driving clinical solutions – from improved socket fits to optimized scoliosis braces – and gains even more capability through AI integration. These advancements ultimately converge on a common goal: to create prostheses and orthoses that are lighter, stronger, more comfortable, and perfectly adapted to each user, all achieved through the synergy of simulation and experimentation. Finite element analysis of a therapeutic orthosis manufactured by 3D-printing The presentation below is focused on evaluating the strength characteristics of a therapeutic orthosis by simulating a three-point bending test (Łukaszewski et al., 2020). The principle of the test is shown in Figure 2.14. As one may notice, after being placed on two support blocks, the orthosis is loaded by a downward vertical force acting on the red surface patch. This load gradually increases from 0 (zero) to a maximum value depending on the material used for 3D-printing. The contact between the orthosis and the support blocks takes place along perfectly matching surfaces. Fig. 2.14. Principle of the three-point bending test simulated for evaluating the strength characteristics of the therapeutic orthosis The following assumptions have been made when preparing the finite element model of the three-point bending test with the SOLIDWORKS Simulation module of the SOLIDWORKS computer-aided design software package: a) The orthosis is made from PLA or PA12 exhibiting an isotropic linear elastic behavior. Table 2.4 lists the physical and mechanical properties of PLA and PA12 that are relevant for the Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 40 finite element model of the three-point bending test. These parameters have been stored in the custom library BIOMEDIX FEA Materials.sldmat provided to the students together with the 3D model shown in Figure 2.14. b) The support blocks are perfectly rigid bodies. c) The orthosis is allowed to slide along its contact surfaces with the support blocks. The frictional component of this contact interaction is neglected. Table 2.4. Physical and mechanical properties of PLA (Farah et al., 2016) and PA12 Mass density ρ [kg/m3] Elastic modulus E [MPa] Poisson’s ratio ν [-] Yield strength Y [MPa] PLA 1252 3500 0.36 59 PA12 1010 1800 0.45 48 The following steps have been performed to prepare the finite element model of the three-point bending test: a) Defining the support blocks as perfectly rigid bodies b) Associating the PLA/PA12 material to the orthosis by accessing the BIOMEDIX FEA Materials.sldmat library (see the example in Figure 2.15) c) Specifying the contact interaction between the support blocks and the orthosis: frictionless sliding contact (Fig. 2.16) Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 41 Fig. 2.15. Associating the PLA material to the orthosis by accessing the BIOMEDIX FEA Materials.sldmat library Fig. 2.16. Specifying the contact interaction between the support blocks and the orthosis (frictionless sliding contact) d) Enforcing a full locking kinematic constraint on the bottom faces of the support blocks (Fig. 2.17) Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 48 3 Tools and Techniques 3.1 CAD Systems with KBE Capabilities The design of anatomically customized biomedical products increasingly requires more than basic 3D modelling tools. Efficient, scalable production of individualized designs - such as orthoses, prostheses, implants, or surgical models - demands systems capable not only of parametric modelling but also of embedding design knowledge directly into the modelling process. These are referred to as KBE-capable CAD systems, meaning they support KnowledgeBased Engineering functionality through intelligent modelling, rule-based automation, and external data integration. KBE-capable CAD platforms combine traditional solid and surface modelling tools with logic-driven model generation, scripting interfaces, and parametrization frameworks. This enables the creation of intelligent CAD models that can adjust geometry based on patientspecific data or pre-set clinical rules, forming the core of automated design systems. In this section, several commonly used KBE-capable CAD platforms are briefly presented and compared with regard to their capabilities, limitations, and suitability for implementation in biomedical design automation workflows. Several professional CAD systems provide mature environments for both mechanical modelling and knowledge-based automation. The most prominent in this space include Autodesk Inventor, Fusion 360, SolidWorks, and CATIA V5. Each of these systems supports intelligent model development, but they differ in terms of automation capabilities, modelling paradigms, accessibility, and target user base. The following analysis is based on earlier work (Górski 2025, Zawadzki 2018). Autodesk Inventor Inventor is a robust mechanical design software widely used in product development. It offers strong support for parametric and feature-based modelling, with integrated modules for part and assembly creation, simulation, and documentation. Its primary advantage in KBE workflows is the iLogic system (Fig. 3.1), a scripting environment that allows users to embed conditional logic, parameter relationships, and rule-based behaviors within CAD models. iLogic also enables linking to external data sources (e.g., Excel files), allowing model regeneration based on patient-specific data sets. For more complex automation, Visual Basic for Applications (VBA) can be used for scripting advanced routines. Inventor supports hybrid solidsurface workflows, making it suitable for developing components that need anatomical adaptation with structured features such as flanges or mounts. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 49 Fig. 3.1. Autodesk Inventor and iLogic scripting system – modular prosthesis Autodesk Fusion 360 Fusion 360 is a lightweight, cloud-based CAD/CAM platform designed for integrated product development. It supports parametric modelling, assembly design, FEA simulation, and generative design. Although its scripting capabilities (through Python and JavaScript) are more limited than Inventor’s iLogic, it remains highly suitable for automated workflows, especially for smaller-scale applications. Fusion’s open architecture and native cloud connectivity make it a strong candidate for distributed or remote design workflows, and its free licensing options for educational or startup users further enhance accessibility. Its strength lies in its seamless user experience and full integration of design-to-manufacturing tools. SolidWorks SolidWorks, developed by Dassault Systèmes, is one of the most widely used CAD systems in engineering. It features a comprehensive parametric modelling environment, powerful assembly tools, and a broad ecosystem of plugins. For KBE applications, SolidWorks offers built-in automation through equations, design tables, and macros, as well as scripting through VBA. It supports configuration management, enabling multiple design variants to be managed within a single file. While SolidWorks does not offer a dedicated KBE module comparable to CATIA’s Knowledgeware, its environment is mature enough to support intelligent modelling when combined with scripting and modular design logic. CATIA V5 CATIA is a highly advanced CAD/CAE system built for complex engineering, particularly in automotive, aerospace, and large-scale industrial design. It offers full-spectrum modelling tools - from A-class surfacing and freeform design to mechanical simulation and PLM integration. Its Knowledgeware module allows users to create parametric templates, embed Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 50 logic-based rules, build product configurators, and automate design tasks using its own scripting language. CATIA handles hybrid modelling quite well and can work directly with imported meshes coming from 3D scans or medical imaging, making it ideal for anatomical modelling at scale. However, the software is very expensive and not easy to learn. Typically it is used by large enterprises and research institutions, rather than small healthcare providers, SMEs or independent specialists. To facilitate selection for practical implementation, a comparative summary is presented in the table 3.1, emphasizing KBE-relevant features. Tab. 3.1. Comparison of most popular KBE-capable CAD systems, based on (Górski 2025) Feature / System Inventor SolidWorks CATIA V5 Fusion 360 Parametric modelling Yes (fully supported) Yes (fully supported) Yes (advanced, scalable) Yes Surface modelling Moderate (hybrid-friendly) Moderate Advanced (GSD module) Moderate Mesh handling Limited Limited Good (with DSE module) Basic Rule-based modelling iLogic, VBA scripting Equations, macros Knowledgeware (native) Basic scripts (Python) External data linking Excel, OLE, APIs Excel, macros External tables, XML, APIs CSV, API Simulation integration Basic FEA Robust FEA suite Advanced multiphysics Integrated FEA/CFD Usability Medium (engineering focus) High (intuitive) Complex (steep learning) High (beginnerfriendly) Licensing cost High High Very high Low/free (for startups) Best suited for SMEs, engineering teams General design firms Large enterprises, R&D Startups, educational use Based on the context developed throughout this chapter, where anatomically customized products are generated using intelligent CAD models and embedded design logic, a few observations can be made regarding system selection: 1. For large-scale implementations in hospitals, research labs, or medical device manufacturers dealing with high variability and complexity, CATIA V5 remains unmatched in capability. Its Knowledgeware module supports complete KBE workflows, while its surfacing tools handle organic geometry with precision. However, its complexity and licensing cost restrict its adoption to institutions with significant technical and financial resources. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 51 2. For engineering-focused environments that require solid automation functionality and are already familiar with Autodesk ecosystems, Inventor offers a good balance of performance and accessibility. Its iLogic scripting allows for modular and configurable models, well-suited for generating prosthetic components, surgical guides, and mechanical interfaces for implants. It also integrates well with external spreadsheets and forms, making it practical for clinical design automation pipelines. 3. For smaller teams, startups, or educational environments, Fusion 360 presents a compelling option. Its unified platform covers design, simulation, and manufacturing, with cloud collaboration features ideal for distributed work. While its scripting and automation capabilities are limited compared to Inventor or SolidWorks, it offers sufficient flexibility for simple individualized products such as passive orthoses or surgical templates. Additionally, it supports 3D printing workflows natively, reducing the friction between design and fabrication. 4. For general-purpose design with strong simulation and documentation features, SolidWorks is a viable alternative to Inventor, particularly if simulation accuracy or design verification are critical. It may be preferred by teams already embedded in the Dassault Systèmes ecosystem or familiar with SolidWorks’ interface and plugin ecosystem. 5. In environments where non-parametric mesh editing is essential (e.g. maxillofacial reconstruction, tumor modelling), none of the above systems is sufficient alone. A hybrid workflow using specialized mesh processing tools (e.g. Blender, Meshmixer) combined with one of the parametric CAD systems is often the best approach. CATIA has some mesh capabilities (Digitized Shape Editor), but for smaller operations, a Fusion-Blender or Inventor-Meshmixer pipeline is often more realistic. The choice of CAD system for KBE applications in biomedical design should be guided by: • technical requirements – level of model complexity, frequency of case variation, need for simulation or analysis, • scope of automation - whether full automation is needed or partial is enough, • scalability - volume of cases per week/month and levels of variability, • user profile – clinicians and other medical (non-tech) personnel biomedical engineers, CAD technicians, or hybrid roles. • economic constraints - licensing costs and models (permanent, subscription, tokens), maintenance, training availability. In most academic or SME contexts, the combination of Fusion 360 (for general modelling and automation) and Inventor (for intelligent model development and integration into design systems) provides an optimal balance between functionality and cost. For larger-scale systems where funding and expertise are available, CATIA V5 (or Siemens NX) remains the gold standard for implementing deeply integrated KBE systems capable of handling anatomical variability at industrial scale. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 52 3.2 Design Automation Case Studies AutoMedPrint: A System for Orthopedic Device Design The AutoMedPrint system, described in numerous previous publications (Górski 2025, Górski et al. 2024a, Górski et al. 2020) represents a complete and modular implementation of an automated design and manufacturing framework for anatomically customized orthopedic and prosthetic devices. It was developed as a demonstrator of a fully functional digital workflow that integrates 3D scanning, data processing, rule-based CAD model generation, and additive manufacturing preparation - all without requiring active participation of an engineer during the routine operation. The system is centered on the principle that complex design logic, expert decision pathways, and model variability can be captured and embedded within intelligent CAD structures, accessible through an intuitive user interface and operated by medical or technical staff. The description in this chapter is based on the aforementioned earlier published works, mostly (Górski 2025, Górski et al. 2024a). At the core of AutoMedPrint lies a set of intelligent, rule-driven CAD models developed in Autodesk Inventor using the iLogic environment. These models represent modular products such as cosmetic and mechanical upper limb prostheses, as well as wrist-hand orthoses and ankle-foot orthoses (AFOs). Each model is designed to regenerate dynamically based on anthropometric data extracted from 3D scans. This enables the rapid creation of individualized device geometries that conform to clinical and functional needs without redesign from scratch. The main ideas are presented in Fig. 3.2. The process begins with non-contact anthropometric measurement, using a structuredlight 3D scanner mounted on a mobile or fixed rig. A supporting structure stabilizes the patient’s limb, while a software module (AutoMedPrint Operator Panel) coordinates scan acquisition, limb positioning, and data capture. The scanned mesh is then processed through a chain of macro-enabled software tools (notably MeshLab), which perform cleanup, orientation correction, and cross-sectional analysis. Point cloud data, diameters, and other measurements are extracted and stored in a structured format for further processing. A visual application called the Limb Calibrator - developed in Unity 3D - allows the user to manually define anatomical reference planes (e.g., wrist, end of limb), adjusting cross-section positioning as required. This ensures consistency in measurement interpretation and model alignment. The measured values are written to external files (typically Excel spreadsheets), which serve as input for the automated design table. This table transforms the raw anatomical data into a complete configuration dataset suitable for regenerating the intelligent CAD model. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 53 Fig. 3.2. Work of AutoMedPrint system based on (Górski et al. 2024) The actual model regeneration is launched via Excel macros and batch scripts. The system reads the configuration file, opens the parametric CAD model, and applies all necessary adjustments - including geometry scaling, feature activation/suppression, and module selection. All of this occurs without user interaction with the CAD interface itself. Upon completion, the CAD system exports a manufacturing-ready file (e.g., STL) to the fabrication workstation. The final stage of the pipeline involves the rapid manufacturing module, where the digital model is sliced using predefined settings and sent to an extrusion-based 3D printer. The printing process is semi-automated, with post-processing and assembly performed by trained technical staff. The overall process – from scanning to production-ready design – can be completed in under 15 minutes in typical cases, making AutoMedPrint a viable solution for same-day device fabrication under certain conditions. The system's modular architecture is a key enabler of its adaptability. Four major functional modules are identified: 1) 3D scanning and measurement; 2) automatic design; 3) interactive user interface; and 4) additive manufacturing. Each module can be deployed individually or in combination depending on the operational context. For example, in a clinic lacking in-house fabrication capabilities, design data may be sent to a centralized production site using a distributed manufacturing model. Alternatively, scanning and configuration may be completed locally while design generation is performed remotely on a cloud-hosted CAD server - allowing institutions to share access to high-cost licenses. In terms of software, AutoMedPrint relies on a combination of commercial and open source tools (e.g., Autodesk Inventor, Excel, MeshLab, slicer software) and proprietary interfaces built using Unity and Visual Studio. The integration is achieved through external scripting, macro automation, and batch execution, which together create a seamless pipeline. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 54 From a usability standpoint, the system minimizes operator burden. Patients can interact with the system via a touch interface or VR headset to customize aesthetic features of the product, while technical and medical staff assist with scanning and post-processing. No specialized CAD expertise is required during the operational phase; all design logic is embedded in the system and updated as part of a continuous improvement process based on user feedback, clinical outcomes, and test-case evaluations. AutoMedPrint provides a comprehensive case study in the development of a real-world, fully functional automated design system tailored for biomedical applications. It exemplifies how structured knowledge representation, modular software development, and intelligent modelling can be combined to deliver a scalable solution for individualized patient care. Automated Design of Orthoses The automated design of orthoses using the AutoMedPrint system streamlines the creation of personalized medical devices by integrating 3D scanning, digital modeling, and additive manufacturing. This enhances precision, reduces costs, and ensures a better fit for patients. To realize the design of orthoses in an automated manner, the modular structure of AutoMedPrint is utilized, consisting of (Górski et al. 2024a, Górski et al. 2024c): • 3D scanning & design station – operator-controlled scanner and software for data acquisition, • user interface station – a touchscreen, VR headset, and interactive software for customization, • rapid manufacturing dtation – 3D printing and post-processing tools for production. A functional prototype was iteratively improved, with further modifications needed for mass production. A basic, foundational product contained in the AutoMedPrint system is wrist hand orthosis, developed since the initial versions of the system (Górski et al. 2020). The orthosis is customized on the basis of a non-contact measurement of geometry of patient’s hand and forearm (or mirror image of the other limb, when the actual limb is damaged and e.g. wrapped in plaster cast). The measurement is done by optical 3D scanning, usually at the workplace developed as a part of the AutoMedPrint system, developed at Poznan University of Technology. After measurement, data is processed from raw scans to reconstructed, smooth limb model (Fig. 3.3). Out of this model, sets of points are extracted to feed the intelligent CAD model. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 55 Figure 3.3. Data processing of 3D scans for the wrist hand orthosis model (Górski 2025, Górski et al. 2020) The product was originally designed in the Autodesk Inventor CAD system (Figure 3.4), as an intelligent model – its design can be changed freely by supplying it with various data from 3D scanning, leading to automated re-design. Figure 3.4 Design of orthosis in Autodesk Inventor (AutoMedPrint system materials) The orthosis consists of basic parts (Fig. 3.5): - bottom part (in contact with palm), Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 56 - top part (in contact with back of the hand), - optionally – the bottom and top part could be transversally divided if orthosis Figure 3.5. Wrist hand orthosis – parts (Górski et al. 2024a) The orthosis design table contains points extracted from the limb reconstruction. It is an Excel spreadsheet. The point extraction is realized by AutoMedPrint system algorithms, using Excel and MeshLab software. Integral element is the aforementioned Limb Calibrator application (Fig. 3.2). The process of data extraction is automatic. As a result, new Excel spreadsheed is generated. Then the user (e.g. a student) opens the Inventor model and updates the design table. After updating, check for errors and possible improvements, the model must be saved to external file for further use. It is usually done in two ways: - whole orthosis is saved in STP file, for the database - individual parts (solid bodies) are saved as OBJ or STL files for 3D printing. Case study example As an example of a patient, a case of a 26-year old man was selected, with an injury to his right wrist, caused by bite of a dog resulting in bone crush. A full process was undergone and recorded for him (3D scanning shown in Figure 3.6, finished with obtaining a complete functional orthosis (Figure 3.7). The scanning process involved capturing both limbs separately. The left (healthy) arm was scanned in a correct anatomical position using a David SLS-3 scanner. The right (affected) arm was manually scanned with an EinScan Pro 3D scanner, positioned for patient comfort. bottom part top part openwork basic assembly features Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 57 Figure 3.6 Patient with injured hand 3D scanning using AutoMedPrint system at PUT Using the AutoMedPrint system, automated macros extracted key data, incorporating 3 mm offset for lining and 4 mm standard thickness for durability. A CAD-based wrist-hand orthosis model was created on the basis of extracted parameters. The design was digitally aligned with the scan to ensure a precise fit and avoid structural conflicts. Figure 3.7. Automatically designed model of the orthosis The whole design process took approximately 30 minutes after the data was extracted from the orthosis. The longest time is actually the automatic regeneration of the model by Autodesk Inventor. Also, to ensure lack of extra iterations, the orthosis was manually tested against the scan of the affected hand for collisions, using MeshLab software, which took additional 15 minutes before accepting the design and putting it into production. Automated Design of Prostheses The automatic prosthesis design procedure described in the following section pertains to a solution developed within the AutoMedPrint project. It involves a modular mechanical prosthesis specifically designed for cycling, described in earlier work (Górski et al. 2024b, Górski 2025). Personalization of hand prostheses is a particularly crucial aspect of orthopedic product manufacturing. This is due to the improved fit of the prosthesis to the patient’s stump, Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 64 Another critical component of an automated workflow is the orchestration engine that links all the stages together (Ben-Nun et al. 2020). This layer manages the flow of data from CAD conversion through slicing and into the generation of machine instructions. It monitors input directories, triggers processing scripts, manages logging and error reporting, and interfaces with external systems such as product lifecycle management (PLM) platforms. In many cases, the orchestration logic is implemented using general-purpose programming environments such as Python or Node.js, often coupled with workflow automation tools that allow for modular and scalable integration. This orchestration is essential to ensure that all processing steps occur in the correct sequence, that errors are caught early, and that all outputs are properly stored, tracked, and made available to downstream systems. The final output of the workflow is the numerical control code that instructs the additive manufacturing machine how to build the part. This code, which may take the form of G-code, X3G, or other proprietary formats depending on the machine, must be transmitted to the printer in a reliable and traceable manner. Automation at this stage ensures that the correct file is associated with the correct job, reducing the risk of operator error. In production environments, this often involves direct network transfer to machine, sometimes through secure APIs or cloud-based interfaces (Baumann et al. 2017). Alternatively, systems may prepare USB drives or other physical media with the correct file structure and naming conventions, ready for manual transfer where required. In both cases, the goal is to minimize or eliminate manual handling of digital files, thereby preserving the integrity of the process. Several industrial sectors have already adopted end-to-end automation in their additive manufacturing workflows. In the medical device industry, for example, the production of patient-specific orthoses, implants, and surgical guides has been revolutionized by automation systems that accept anatomical data, process it into printable geometry, and generate NC code without human intervention (Górski et al. 2022). These systems often operate in compliance with stringent regulatory standards and include robust validation steps to ensure clinical safety. In aerospace and automotive manufacturing, automated workflows are employed to accelerate the production of fixtures, jigs, and prototype components, integrating directly with enterprise data systems to retrieve part specifications and manufacturing requirements. In consumer-facing applications, online customization platforms allow customers to personalize products via web interfaces, with all subsequent processing steps — from geometry regeneration to slicing and job queuing — handled automatically in the cloud. Despite these successes, implementing automation in additive manufacturing workflows is not without challenges. One of the primary obstacles is the lack of standardization across software tools, machine interfaces, and data formats (Xiao et al. 2018). While some initiatives, such as the 3MF Consortium, aim to establish unified formats that preserve both geometry Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 65 and metadata, widespread adoption remains limited. As a result, many automation pipelines require customized adapters or translators to interface with specific slicers, CAD tools, or printers. Another challenge is ensuring robustness in the face of input variability. Unlike traditional manufacturing, where inputs are highly standardized, additive workflows often deal with unique geometries and customer-defined parameters. Automated systems must therefore include comprehensive error handling, input validation, and exception management routines to prevent unexpected failures. Additionally, managing complexity in parameter selection and manufacturing strategy definition can be difficult, particularly when attempting to scale automation across diverse product families or materials. Successful deployment of automation in additive manufacturing also depends on integration with upstream and downstream processes. A fully digital thread requires that design inputs, production parameters, quality control data, and logistics information flow seamlessly between systems. Automation workflows must be capable of interfacing with ERP or PLM platforms to retrieve or update relevant information, schedule jobs, and track status. Modular architecture is often preferred, allowing individual components of the automation stack to be updated, replaced, or scaled independently. This not only facilitates system maintenance but also supports innovation and customization without risking the stability of the entire pipeline. However, a significant challenge in maintaining such systems arises from the rapid pace of development in additive manufacturing technologies. New machines are frequently introduced, often implementing different protocols or standards for data transmission, machine control, and manufacturing job management. As a result, a system once fully functional may require repeated adaptation to remain compatible with evolving hardware ecosystems. Furthermore, uncertainty regarding the long-term availability or support for existing equipment raises concerns about whether worn or obsolete machines can be replaced with equivalents that will function reliably within the established automation framework. These factors introduce a layer of complexity that demands forward-looking planning, flexibility in system design, and the capacity to rapidly integrate new standards without disrupting the continuity of automated workflows. 3.4 Scaling Up: Mass Customization of Biomedical Products The increasing demand for personalized care, alongside advances in digital technologies, has created fertile ground for the development of scalable, customizable medical solutions. Among the most transformative approaches emerging in this context is mass customization, a manufacturing strategy that merges the efficiency of mass production with the specificity of individual tailoring. While initially applied in consumer sectors like automotive and electronics, mass customization is now making significant inroads into the biomedical domain, particularly Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 66 in the production of anatomically customized products such as orthoses, prostheses, and implants (Górski 2025). Mass customization (MC) challenges the binary of craft-based individualization and highvolume standardization. It allows for the creation of highly individualized products with lead times and unit costs approaching those of mass-produced goods. This is particularly relevant in biomedical applications, where anatomical variability across patients makes customization not just desirable but clinically necessary. Within the spectrum of production methods—ranging from piece production to smallseries and mass production—mass customization occupies a unique position. It leverages the scalability and logistics of mass production while enabling the generation of case-specific variants, made possible by automation, digital manufacturing, and integrated design workflows. For biomedical products, this means producing tailored solutions at scale, with consistency and repeatability, while retaining the clinical specificity required for individual treatment. The analysis below is based on previous work (Górski 2025, Górski et al. 2024a, Górski et al. 2025, Zawadzki 2018). The feasibility of mass customization in medicine has been significantly advanced by the convergence of several key technologies: • 3D Scanning and imagingmodern structured light and laser scanning systems enable accurate and fast acquisition of external body geometries. For internal structures, CT and MRI data provide detailed 3D models; • Computer-Aided Design (CAD) they allow the transformation of anatomical data into individualized product geometries. Intelligent CAD models with embedded design logic enable consistent adaptation to varying patient anatomies; • Additive Manufacturing and 3D printing technologies, especially those adapted for medical-grade materials (e.g., biocompatible polymers, titanium alloys), allow ondemand fabrication of products with complex geometries, optimized for each patient; • Automated Design Systems (ADS) – as demonstrated in the AutoMedPrint framework, ADS platforms integrate patient data acquisition, rule-based CAD model regeneration, and manufacturing preparation. This reduces the need for manual CAD intervention and shortens the design cycle dramatically; • Knowledge-Based Engineering (KBE) – by formalizing expert knowledge and decision logic, KBE helps automating repetitive design tasks, managing configuration complexity, and reducing human error. In biomedical contexts, this translates into reliable, high-speed generation of anatomically adapted device models. These enablers collectively support the transformation from traditional one-off production to a scalable, semi-automated customization pipeline. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 67 Mass customization is particularly well-suited to a range of biomedical products where anatomical fit and patient-specific function are essential: • orthoses and prosthetic sockets: these products are inherently patient-specific due to variability in residual limb shape, movement range, and patient condition. Mass customization enables digital workflows where scanning, model adaptation, and fabrication occur in rapid succession, producing clinically effective and comfortable devices. • customized implants: cranial plates, spinal cages, and joint replacements benefit from individualized design to improve integration and reduce complication rates. Automated design workflows reduce time-to-surgery and enable hospitals or manufacturers to offer customized implants as a service. • surgical guides and preoperative models: custom models used for planning or guiding surgery can be generated from imaging data, offering greater precision and reducing operative time. With automated design processes, such tools can be produced rapidly and at a sustainable cost. • wearable rehabilitation devices: devices such as wrist-hand orthoses or dynamic splints can be customized to patient anatomy and therapy goals. Mass customization allows these to be fabricated in small clinics using cloud-based tools and desktop additive manufacturing. Traditionally, many customized medical devices, especially orthotic and prosthetic products, have been manufactured using manual techniques. These include hand-moulding thermoplastics or laminating around plaster models derived from negative moulds. While functional, these processes are labor-intensive, difficult to scale, and highly dependent on technician skill. The digital shift introduces repeatability, documentation, and opportunities for automation. Modern workflows, such as those implemented in AutoMedPrint, begin with 3D scanning, followed by digital measurement extraction, intelligent CAD model generation, and additive manufacturing. Each stage replaces a manual step, bringing traceability and reducing variability. Still, scalability demands more than digital tools, as it requires automation of decisionmaking and integration across systems. Only through KBE and robust design automation can large volumes of customized designs be handled effectively without increasing the design team’s workload proportionally. Despite its advantages, implementing mass customization in medical device production is not without challenges: Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 68 • creating a system capable of generating reliable, individualized designs requires extensive domain knowledge and collaborative input from clinicians, engineers, and regulatory experts. • each patient presents a unique case, which introduces high variability in design inputs. Automated systems must be robust against errors, incomplete data, or borderline anatomical cases. • biomedical devices must meet stringent safety and documentation standards (more on that in chapter 7 of this e-book). Custom products still require traceability, consistent performance, and appropriate certification—even when made on-demand. • transitioning to mass customization often redefines professional roles. Orthotic technicians, for instance, must adapt from manual fabrication to operating digital scanners and verifying CAD-generated results. • while additive manufacturing and automation reduce labor, initial setup costs, hardware, software, and development can be significant. However, in many cases, long-term ROI justifies the investment, especially for midto high-volume customization. The trajectory of modern healthcare increasingly favors personalized medicine, where treatments, interventions, and devices are tailored to individual patients. Mass customization offers the technological backbone to deliver on this promise efficiently. With maturing digital infrastructure, greater integration of AI, and enhanced imaging techniques, we can expect increasing uptake of these systems across hospital networks and medical device manufacturers. Crucially, mass customization must remain focused on value creation, improving patient outcomes, reducing lead times, and enabling clinicians to access better tools with minimal complexity. Intelligent automation, especially when grounded in clinically validated design logic, allows healthcare to offer customization as standard practice rather than an exception. In conclusion, mass customization represents a scalable pathway to delivering patientspecific biomedical products in a controlled, cost-effective, and repeatable manner. As enabling technologies continue to evolve, and the gap between manual and digital processes narrows, biomedical engineering stands at the threshold of a transformation where personalization and scalability can coexist, bringing precision medicine into the realm of routine care. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 69 4 Conclusions 4.1 Challenges and Future Trends The adoption of Knowledge-Based Engineering (KBE) methodologies in biomedical product design offers tremendous potential, yet its implementation in clinical and engineering settings remains complex. Several technical, organizational, and regulatory barriers must be addressed for KBE systems to reach their full transformative capacity. At the same time, rapid technological developments, particularly in artificial intelligence, cloud computing, and collaborative platforms, are shaping the next generation of KBE-enabled biomedical design. This section outlines key challenges and presents emerging trends that will define the evolution of automated, anatomically aware design workflows. One of the most critical challenges in deploying KBE in biomedical contexts lies in the acquisition and formalization of expert knowledge. Much of the relevant expertise – whether clinical, biomechanical, or manufacturing-related - remains tacit, residing in the routines and intuition of experienced professionals. Transforming this implicit knowledge into formalized design logic requires structured interviews, reverse engineering, long-term observation, and interdisciplinary collaboration. Moreover, the knowledge must be contextualized and validated to ensure its accuracy and relevance across diverse cases. Effective representation of this knowledge is equally important. Declarative rules, parametric templates, ontological maps, and decision trees must be implemented within systems that remain interpretable, modifiable, and scalable. In medical applications, the risk of embedding outdated or overly rigid logic poses a safety concern. Therefore, KBE implementations must be coupled with continuous feedback loops and robust validation mechanisms, particularly when used in high-risk scenarios such as implant design or surgical planning. Another challenge is interoperability. While many CAD platforms now support parametric modelling and scripting capabilities, integration with additive manufacturing (AM) workflows, patient data systems, and simulation tools is far from seamless. Differences in file formats, model fidelity, metadata standards, and automation interfaces hinder the creation of fully automated design pipelines. Custom interfaces or middleware solutions are often required to bridge gaps between data sources (e.g., DICOM or mesh files), CAD environments (e.g., SolidWorks, Fusion 360, NX), and AM preparation tools (e.g., slicers, post-processing scripts). Without standardization, these bridges are fragile and difficult to scale. Additionally, regulatory frameworks for medical devices remain oriented toward static, repeatable products, not dynamically generated, patient-specific solutions. Even though KBE Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 70 systems aim to increase consistency and traceability, the individualized nature of their outputs places them in a grey area with respect to certification and liability. Future regulatory models must evolve to assess not only the product but also the design system itself, validating the logic, traceability, and reliability of automated design platforms. Despite these challenges, several technological trends are poised to reinforce and expand the capabilities of KBE in biomedical engineering. Most prominently, the integration of artificial intelligence into KBE workflows offers new pathways for data-driven optimization, predictive modelling, and real-time personalization. Machine learning models, trained on extensive simulation or clinical datasets, can serve as surrogate evaluators in optimization loops, accelerating the generation of anatomically adapted devices. Additionally, AI can assist in the extraction of anatomical features from scans, recognition of pathological patterns, or identification of optimal material configurations—all tasks that enhance and complement rule-based KBE systems. Another major development is the move toward cloud-based and collaborative platforms for biomedical design. These platforms allow distributed access to intelligent design tools, enabling clinicians, engineers, and manufacturers to interact with patient data and design logic in real time. Such systems not only reduce the latency between data acquisition and product generation but also support version control, auditing, and multi-user workflows—features essential for regulated medical environments. Integration with hospital information systems (HIS), PACS/RIS systems, and additive manufacturing services allows for seamless transitions between diagnosis, design, and production. Moreover, advances in web-based configurators, parametric geometry APIs, and digital twins are making it possible to build KBE systems that are accessible not only to engineers but also to clinical users. Surgeons, prosthetists, or rehabilitation specialists can interact with simplified interfaces—inputting patient measurements, selecting treatment scenarios, and previewing device options—while the backend system dynamically applies expert logic to generate optimized designs. Looking forward, the convergence of KBE with AI, cloud infrastructure, and digital health ecosystems points toward a new paradigm of biomedical product development: one where automation does not reduce human involvement but rather augments it, capturing expert reasoning, ensuring consistency, and enabling rapid, patient-specific care at scale. 4.2 Summary and Recommendations The growing demand for anatomically customized biomedical devices has prompted a rethinking of traditional design paradigms in favor of automated, knowledge-driven approaches. This chapter has outlined how Knowledge-Based Engineering (KBE), when combined with modern CAD systems and additive manufacturing, offers a scalable and reliable Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 71 pathway to deliver individualized solutions in orthotics, prosthetics, implants, and surgical planning. Beginning with the rationale and principles of anatomical customization, the chapter introduced key modelling strategies—parametric and non-parametric—and demonstrated how they influence the design of patient-specific devices. Intelligent CAD models and KBE logic structures were shown to form the backbone of automation, enabling rapid regeneration of geometries based on clinical and anatomical inputs. Through real-world examples and case studies, it was illustrated how design automation systems can transform the engineering process from a time-consuming, expert-driven task into a streamlined, rule-based pipeline. The incorporation of finite element analysis (FEA) was also explored, highlighting how simulation and AI can be used to evaluate and optimize device performance before fabrication. From this foundation, several critical insights and recommendations can be drawn, based on earlier work (Górski 2025, Górski et al. 2024a): 1. Formalization of domain knowledge is essential. Successful implementation of KBE in biomedical design depends on the structured capture and encoding of clinical, anatomical, and engineering expertise. Institutions developing such systems should invest in interdisciplinary knowledge acquisition strategies and treat knowledge itself as a critical engineering asset. 2. Robust modelling infrastructure enables scalability. Intelligent CAD models must be designed with parameter stability, regeneration robustness, and logic transparency in mind. Reusability and modularity are key attributes for building scalable automation workflows that can adapt to evolving clinical needs. 3. Simulation and AI extend the value of automation. The integration of FEA and AI not only strengthens design validation but also unlocks opportunities for real-time personalization, predictive adjustments, and feedback-informed optimization. These technologies should be embedded into early stages of system architecture rather than added as afterthoughts. 4. Clinical workflows must be considered from the start. To be adopted in practice, automated design systems must align with existing clinical processes, user competencies, and regulatory expectations. Web-based interfaces, guided inputs, and automatic documentation generation are critical enablers of clinical integration. 5. Continuous validation and human oversight remain indispensable. While automation enhances efficiency and consistency, the need for expert supervision – particularly in rare, specific cases – remains. Hybrid workflows that combine automated processing with expert review strike an optimal balance between speed and safety. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 72 Ultimately, the transition toward automated biomedical design is not merely a technical upgrade. It is a shift in how knowledge, technology, and clinical care converge to serve individual patients more effectively. By embedding medical logic into digital tools, KBE systems help ensure that customization becomes not a barrier, but a default standard in the next generation of patient-centered care. Biomedical Innovations through Digital Transformation of Additive Technologies and Knowledge Exchange - ERASMUS KA 220 BIOMEDIX This project has been funded with support from the European Commission. This publication reflects the views only of the authors, and the Commission cannot be held responsible for any use which may be made of the information contained therein. Page | 73 References Akkad K., Mehboob H., Alyamani R., & Tarlochan F. (2023). A Machine-Learning-Based Approach for Predicting Mechanical Performance of Semi-Porous Hip Stems. Journal of Functional Biomaterials, 14(3), 156. Alcaide-Marzal, J., Diego-Más, J. A., Asensio-Cuesta, S., & Piqueras-Fiszman, B. (2013). An exploratory study on the use of digital sculpting in conceptual product design. Design Studies, 34(2), 264-284. Al-Fakih E.A., Abu Osman N.A., Mahmad Adikan F.R. (2016). Techniques for Interface Stress Measurements within Prosthetic Sockets of Transtibial Amputees: A Review of the Past 50 Years of Research. Sensors, 16(7), 1119. Balaramakrishnan T.M., Natarajan S., & Srinivasan, S. (2020). Roll-over shape of a prosthetic foot: a finite element evaluation and experimental validation. Med. Biol. Eng. Comput., 58(10), 2259. Baumann, F. W., Kopp, O., & Roller, D. (2017). Abstract API for 3D printing hardware and software resources. The International Journal of Advanced Manufacturing Technology, 92(1), 1519-1535. Ben-Nun, T., Gamblin, T., Hollman, D. S., Krishnan, H., & Newburn, C. J. (2020, November). Workflows are the new applications: Challenges in performance, portability, and productivity. In 2020 IEEE/ACM International Workshop on Performance, Portability and Productivity in HPC (P3HPC) (pp. 57-69). IEEE. Bonnet X., Pillet H., Fodé P., Lavaste F., & Skalli W. (2012). Finite element modelling of an energystoring prosthetic foot during the stance phase of transtibial amputee gait. Proc. Inst. Mech. Eng. Part H, 226(1), 70. Castro-Franco A.D., Siqueiros-Hernández M., García-Angel V., Mendoza-Muñoz I., Vargas-Osuna L.E., & Magaña-Almaguer H.D. (2024). A Review of Natural Fiber-Reinforced Composites for Lower-Limb Prosthetic Designs. Polymers, 16(9), 1293. Chen D., Su P., Ottikkutti S., Vartholomeos P., Tahmasebi K.N., & Karamousadakis M. (2022). Analyzing Dynamic Operational Conditions of Limb Prosthetic Sockets with a Mechatronics-Twin Framework. Applied Sciences, 12(3):986. Cucos, M. M., Pista, I. M., & Ripanu, M. I. (2018). Product engineering design enhancing by parameterizing the 3D solid model. In MATEC Web of Conferences (Vol. 178, p. 05011). EDP Sciences. David Müzel S., Bonhin E.P., Guimarães N.M., & Guidi E.S. (2020). Application of the Finite Element Method in the Analysis of Composite Materials: A Review. Polymers, 12(4), 818. Górski F., 2025, Computer Aided Design of 3D Printable Anatomically Shaped Medical Devices: Methodologies and Applications, CRC Press, Taylor & Francis. Górski, F., Denysenko, Y., Kuczko, W., & Żukowska, M. (2024b). Automated Design and Virtual Fitting of 3D Printed Bicycle Prostheses for Children. International Conference Innovation in Engineering, 242– 253. Springer. Górski, F., Denysenko, Y., Kuczko, W., Żukowska, M., Wichniarek, R., Zawadzki, P., & Rybarczyk, J. (2024a). Individualized 3D Printed Orthopaedic and Prosthetic Devices Using AutoMedPrint Technology--Methodologies and Examples. Advances in Science and Technology. Research Journal, 18(6).