OPTIMISING FORMING PROCESS BEHAVIOR USING ARTIFICIAL INTELLIGENCE

Muhammad Saeed, Ayesha Quddus, Hamas Khan, Hadia Zulfiqar, Racim Radjef, Boris Eisenbart, Matthias Kreimeyer Forming process - used to shape flat sheets of material into three-dimensional components •Stamp Forming - using a stamping press as in Fig 1 •Diaphragm Forming - using a diaphragm or flexible membrane Artificial Intelligence -to increase the simulation accuracy of forming process •Optimisation -to improve simulations and time for computations, while reducing the cost •Defect Detection -to overcome defects by detecting them at early stage of design, using point cloud as shown in Fig 2 AUTHORS OPTIMISING FORMING PROCESS BEHAVIOR USING ARTIFICIAL INTELLIGENCE 01. Motivation and Research Objectives Optimisation - Multiple Linear Regression (MLR) is implemented for curve fitting and Multi-objective Genetic Algorithm (GA) for finding optimas. Pareto plots and Open-source tools like Para View and HDF View are used for visualisations of simulations as seen in Fig 4. 04. Results and Future Outlook •The optimisation tool has been deployed for a beam model, now it is being validated on a Double Dome geometry as displayed in Fig 6. •The scanned points are being removed based on an angle threshold as in Fig 7. Additionally, preprocessing techniques such as Octrees, multithreading, KNNs, and distance threshold are used. 03. Initial Evaluation 02. Introduction •Simulation-driven evaluations help in reducing the cost of forming processes by various Artificial Intelligence based optimisation steps as illustrated in Fig 3. •Several Machine Learning techniques such as Genetic Algorithm are used to predict and classify forming behavior. •Induced forming defects, such as wrinkles, bridging, voids, are optically inspected using point cloud-based system. Fig 1: Stamp Forming Fig 6: Double Dome Visualisation Muhammad Shahrukh Saeed University of Stuttgart / Swinburne University of Technology [email protected] (+49 176 36356049) Fig 5: Cloud Compare Defect Detection -Defects are detected by analysing the surface normal. The normal vectors for the simulation and point cloud are compared and visualised as in Fig 5. Fig 4: Beam Visualisations Fig 3: Process Chain Fig 2: Point Cloud Scanned Fig 7: Point Cloud Visualisation Geometry Parameterisation Genetic Algorithm Analysis A beam model for initial testing Minimise or maximise the parameters to get the optimum results Pareto plots to depict the relationship between different parameters Parametrise the material parameters such as young's modulus and shear modulus Pareto Representation of Trans Displacement Shear Modulus (G) 0.0099997 0.0099998 0.0099998 0.0099999 0.0099999 0.01 0.01 0.0100001 0.0100001 0.0100002 0.0100002 0.0100003 2.7E+11 3.2E+11 3.7E+11 4.2E+11 Translational_DisplacementMagnitude (Maximum) Young's Modulus (E1) Young’s Modulus (E1) Vs Maximum Translational_Displacement-Magnitude Shear Modulus vs Mean Average _Nodal_Velocity Mean Average Nodal Velocity Maximum Translational Displacement E1 E2 B1 G Tape_Tensile 1e-18 Pareto Representation of Displacement