AI in Power Electronics Design: Present and Future

APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li AI in Power Electronics Design: Present and Future 1University of Arkansas 2Aalborg University Alan Mantooth1[email protected]u Frede [email protected] Xinze [email protected] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li Speaker Biographies 2/137 Xinze Li was awarded the Ph.D. degree in Electrical and Electronic Engineering by Nanyang Technological University, Singapore, 2023. He gained AI industry experience in computer vision with Singtel (Singapore) and AI research experience in natural language processing. He joined the University of Arkansas, USA, as a postdoctoral research fellow in 2024. His research interests include power converter design automation, light and explainable AI for power electronics with physics-informed systems, and fault prognosis and health management for power semiconductors. Alan Mantooth received the B.S.E.E. and M.S.E.E. degrees from the University of Arkansas, Fayetteville, AR, USA, in 1985 and 1986, respectively, and the Ph.D. degree in electrical engineering from the Georgia Tech, Atlanta, GA, USA, in 1990. He then joined the Analogy, a startup company in Oregon. In 1998, he joined as the Faculty with the Department of Electrical Engineering, University of Arkansas, where he currently serves as a Distinguished Professor. His research interests include analog-and mixed-signal IC design & CAD, semiconductor device modeling, power electronics, power electronic packaging, and cybersecurity. Frede Blaabjerg was with ABB-Scandia, Randers, Denmark, from 1987 to 1988. From 1988 to 1992, he got a Ph.D. degree in Electrical Engineering at Aalborg University in 1995. He became an Assistant Professor in 1992, an Associate Professor in 1996, and a Full Professor of power electronics and drives in 1998. From 2017 he became a Villum Investigator. He is honoris causa at University Politehnica Timisoara (UPT), Romania, and Tallinn Technical University (TTU) in Estonia. His current research interests include power electronics and its applications, such as in wind turbines, PV systems, reliability, harmonics, and adjustable speed drives. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li Seminar Outline –Before Coffee Break 3/137 1. Applications of AI in Power Electronics: An Overview AI in the Life-Cycle Management of Power Converters One-Stop AI Solutions for Power Electronics in Wind Generation Systems 2. AI-Based Power Electronics Design: Present The Promise of AI for PE Design AI in Power Electronics Design 3. New Paradigm of Power Electronics Design Challenges of Existing AI Methods Applied in PE 30-Minute Break Physics-Informed Machine Learning in PE Design PE-GPT: A Generative AI-Based PE Design Paradigm 4. A Future of AI-Native PE Design: Autonomous, Ethical, and Green APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1. Learning Objectives –Section 1 4/137 Roles of Artificial Intelligence (AI) in Power Electronics Explore the application of AI across the life cycle of power electronics systems, including offline modeling and design, as well as online control and maintenance. Fundamentals of Prevailing AI Algorithms Gain insights into widely used AI algorithms through clear explanations of their principles and applications. One-Stop AI Solutions for Wind Generation Systems Discover how various AI algorithms function as one-stop integrated solutions for power electronics, illustrated through case studies focused on AI-driven advancements in wind generation systems. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.1 AI in the Life-Cycle Management of Power Converters 5/137 Annual Number of Publications of AI in Power Electronics (Before 2020) Ref: [1] “In 2024 Alone,Number of AIRelated Publications* in Five IEEE PELS Journals (TPEL, JESTPE, TTE, PEM, OJPELS) Nearly Matches the Total during 2021-2023,with a Staggering 100.7% Annual Growth Rate from 2023 to 2024.” Year No. 2024 271 2023 135 2022 90 2021 52 100.7% Growth *Searched on IEEE Xplore, with all metadata “Artificial Intelligence” OR “AI” OR “ Machine Learning” OR “Deep Learning”. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.1 AI in the Life-Cycle Management of Power Converters 6/137 AI has been Extensively Applied to the Life-Cycle Phases of Power Converters: Design, Control, and Maintenance. Ref: [1] 9.8% 77.8% 12.4% APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.1 AI in the Life-Cycle Management of Power Converters 7/137 Bayesian Network: Quantify Uncertainty Reinforcement Learning: Optimization & Control Ref: [1] Most Widely Applied in Power Electronics Timeline of AI Methods in PE: Expert System and Fuzzy Logic have Limited Developments, Meta-Heuristic Algorithms Show Significant Adoption in Optimization, and Machine Learning Reflects Most Sustained and Diverse Growth. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.1 AI in the Life-Cycle Management of Power Converters 8/137 1. Circuit Parameter Design 2. Control Implementation 3. Feedback Experimental Data to Update Built Models Identify Design Parameters and Objectives Build Circuit Models Optimize Converter Designs Build Control Models Controller Optimization and Design Online Implementation Collect Operation Data Update Models if Outdated Evaluate the Existing Models To fine-tune or retrain models Simulation Data Performance Analysis Circuit Physics Expert Knowledge Hybrid Data-driven and Knowledge-based Offline Surrogate Models • Preprocessing • Feature Extraction • Data-driven Modeling • Fitness / Objective functions Preliminary: Offline Model Building for Circuit or Controller Trained control models Trained circuit models Generic Workflow of AI in Design and Control 3. AI as Optimizers 2. AI as Controllers 1. AI as Surrogate Models AI Functions as Surrogate Models for Converter Behavior Modeling, as Controllers for Control Implementation, and as Optimizers to Search for Optimal Designs. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.1 AI in the Life-Cycle Management of Power Converters 9/137 Generic Workflow of AI in Maintenance Ref: [1] 4. AI as Fault Classifiers 5. AI as RUL Estimators AI Functions as Fault Classifiers for Detection and Diagnosis, RUL Estimators for Predicting Remaining Useful Life, and Decision-Makers for System-Level Predictive Maintenance and Control. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Surrogate Models: Case Study 16/137 Use an Artificial NN to Replace the Thermal Matrix. The Computation Complexity of NN Increases Linearly with the Number of Heat Sources. Ref: [5] Thermal Matrix (Complexity of characterization) Complexity of Computation/Utilization An ANN-Based Thermal Matrix ANN APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Surrogate Models: Case Study 17/137 Ref: [5] Simultaneous Cooling Curves (Single Test) NN can Process MIMO System Directly Heating Current: Experimental Settings for Thermal Measurements and the Accuracy of the NN Model: Comparable Accuracy as Thermal Matrix but Faster Computation. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Controllers: Algorithms 18/137 Fuzzy Logic (FL) “Transform Boolean Logic into Multiple Logical Values to Mimic Human-Like Reasoning.More Robust and Adaptive.” Rule Format: IF … (rule cause), THEN ... (rule consequent) Boolean Logic Fuzzy Logic Is 36-kV medium voltage application? IF: Voltage is within [1 kV, 35 kV] THEN: Yes Membership volt 1 kV 35 kV No (or Yes) volt HV: 0.45 MV: 0.65 LV MV HV Fuzzy sets Multiple values APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Controllers: Algorithms 19/137 … * () () =∫ ∫ y y dy yy dy µ µ … ω 1 ω 2 ω 3 1 11 12 1 =++f ax bx c 2 21 22 2 =++f ax bx c 3 31 32 3 =++f ax bx c 11 2 2 3 3 123 fff f ωω ω ωωω ++ =++ Defuzzify Mamdani FIS Takagi-SugenoKang FIS The Rule Consequent is a Fuzzy Set IF: Then: The Rule Consequent is Numeric. Input fuzzification Expert System, Decision Making Input fuzzification Fuzzy infer Fuzzy infer IF: Then: Aggregate Control System, Real-Time Optimization Applications Fuzzy Inference System (FIS) Fuzzy Inference System (FIS) Includes Mamdani and TSK, Depending on whether the Rule Consequent is Fuzzy or Crisp (Numeric), which is Commonly Applied in Real-Time Control, Optimization, and Decision Making. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Controllers: Case Study 20/137 Ref: [6] Control Objective: MPPT Optimal power curve to achieve MPPT: Fuzzy Controller A Fuzzy Controller for Maximum Power Point Tracking (MPPT) in Wind Energy Systems. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Controllers: Case Study 21/137 Ref: [6] Schematic Diagram of the Fuzzy Controller Conventional Conventional Mamdani Fuzzy Fuzzy Fuzzy Rules Mamdani FIS Achieved Faster Transients when Tracking Maximum Power than Traditional Controllers. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Controllers: Algorithms 22/137 Reinforcement Learning Diagram Action (Control Dynamics) Memory Replay Environment (Power Converters) Reward Function Reinforcement Learning (RL) Optimizes Control Trajectories or Solves Optimization Problems by Reinforcing Positive Actions: Trial-Reward-Learn. Critic (Twin of Environment) Controller APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Controllers: Case Study 23/137 RL-Based MPPT Control Ref: [7] Online Learning & Deploying A Reinforcement-Learning (RL)-Based Controller for MPPT in Wind Energy Systems. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Controllers: Case Study 24/137 The RL-Based MPPT Controller Tracked Optimal Power More Closely, Indicating Faster Transient Response and Reduced Power Dip. Conventional P&O MPPT Ref: [7] RL-Based MPPT Closer to Optimal Power vs APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Optimizers: Algorithms 25/137 S T A R T Initialize Design Objective Function Evaluation Generate New Design Meta-Heuristically Navigate the Search Trajectory Compare, Select, and Update Candidate Pools Stop? Performance Space N Candidate Design Solutions Design Space Effi. Density Param. 1 Param. 2 Evolve towards Generic Workflow of Meta-Heuristic Algorithm (MHA): New Design Candidates are Meta-Heuristically Generated (Like Genetic Mutation, Crossover), Objective Functions Guide the Selection towards Optima. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as Fault Classifiers: Case Study 32/137 Distribution Alignments of Synthetic Datasets Neural Network Accuracy with Real/Synthetic Datasets Real data with 0%-1% noise Real data with 0%-3% noise Wind Speed 5±0.5 m/s Wind Speed 7±0.5 m/s Ref: [9] Synthetic Data from Generative Adversarial Networks Aligned with True Dataset, Increasing FDD Accuracy by > 10% with a Data Size of 5000. Synthetic Data APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as RUL Estimators: Algorithms 33/137 AI for Remaining Useful Life (RUL) Prediction: A Probabilistic Perspective. “Predict the Expected RUL, its Distribution, and Confidence Level.” Predictive Maintenance Probabilistic AI Models can Quantify Uncertainties! Ref: [10] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as RUL Estimators: Case Study 34/137 Conceptual Structure for Condition Monitoring Ref: [10] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as RUL Estimators: Case Study 35/137 Diverse Single Failure Precursors Synthesize Composite Failure Precursor (CFP) Ref: [11] Genetic Programming Genetic Programming Synthesizes Composite Failure Precursors to Improve RUL Prediction Accuracy. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 AI as RUL Estimators: Case Study 36/137 Evolution of RUL and 95% Confidence Interval (CI) Using Composite Failure Precursor (CFP) Performance Comparison Using Single Indicator (Rds,on) and CFP Ref: [11] AI-Synthesized Composite Failure Precursors are More Accurate than Single Precursors, with the 95% Confidence Interval Reliably Covering the Real RUL. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 1.2 Key Takeaways –Section 1 37/137 Applications of AI in Power Electronics AI in the design, control, maintenance of power converters Main functionalities of AI: Surrogate models, controllers, optimizers, fault classifiers, and remaining useful life estimators Fundamentals of Prevailing AI Algorithms Neural network (NN): NN inference and training, model backbone and different NN structures, model head and various learning paradigms Fuzzy inference system (FIS): Mamdani FIS and TSK FIS Reinforcement learning (RL): Trial-reward-learn Meta-heuristic algorithm (MHA): Generic workflow and classifications One-Stop AI Solutions for Wind Generation Systems APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li Seminar Outline –Before Coffee Break 38/137 1. Applications of AI in Power Electronics: An Overview AI in the Life Cycle Management of Power Converters One-Stop AI Solutions for Power Electronics in Wind Generation Systems 2. AI-Based Power Electronics Design: Present The Promise of AI for PE Design AI in Power Electronics Design 3. New Paradigm of Power Electronics Design Challenges of Existing AI Methods Applied in PE 30-Minute Break Physics-Informed Machine Learning in PE Design PE-GPT: A Generative AI-Based PE Design Paradigm 4. A Future of AI-Native PE Design: Autonomous, Ethical, and Green APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li Learning Objectives –Section II 39/137 Fundamental Concepts of Power Electronics Design Explore the principles of model-based design (V-Diagram), design space, performance space, and hierarchical steps in power converter design. The Significance of AI in Power Electronics Design Understand the critical role of AI in advancing power electronics design. “Present” Applications of AI in Power Electronics Design Examine the present-day applications of AI in power electronics design, highlighting its impact and practical implementations. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.1 Basics of Model-Based Design 40/137 Design: Finer Granularity Top-Down Bottom-Up Verification: Coarser Granularity Model-Based Design Flow: The V-Diagram Ref: [12] Model-Based Design is a Prevalent Theme, Featuring a Top-Down Approach to Refine Converter Design from System to Block, Paired with a Bottom-Up Approach for Verification. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.1 Basics of Model-Based Design 41/137 Ref: [13] Model-Based Power Electronics System Design, Starting from System Functionality and Progressing through System Architecture, Converter Design, and Component Design. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.1 AI is a Key Enabler for PE Design 4.0 48/137 PE Design 4.0  Autonomous Design Augmented Design Assisted Design (SOTA) Human-in-Loop for Design Human AI Human AI Human AI Semi-Autonomous Flow Suggest Design Details Based on Prev. Designs Limited User Interaction Mainly Meta-Heuristic Optimization Make Decisions Select/Validate/ Iterate Design Generate Full Designs Visualize Designs Def. Workflows/Rules/Models Human Duties AI Duties Ref: [14] AI is a Key towards Fully Autonomous PE Design, Independently Generating Full Design Details. PE Design 3.0 PE Design 2.0 Manual Design PE Design 1.0 APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in Power Electronics Design: An Overview 49/137 System-Level Circuit Topology Power Semiconductors, Modules, Gate Drivers System Architecture Circuit Parameter Design Modulation and Control Magnetics Capacitors Busbar and PCB EMI Filters Thermal Management Component -Level Converter -Level APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in PE System Architecture Modeling 50/137 Case 1. AI-Based Emulation of More Electric Aircraft (MEA) PE Systems, Using Feedforward NNs and Recurrent NNs to Model IGBT Switches, Converters, PMSM Current, Torque, and Speed. Transient Behaviors (during Switching) Steady-State (when Oscillation Settles) Converter Behavioral Model Current Torque, Speed Ref: [16] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in PE System Architecture Modeling 51/137 Real-Time AI-Based Emulation Platform Mean Absolute Accuracy of Each NN Waveforms Emulated by Neural Networks Generator 3ph Voltage PMSM Torque Rectifier DC Voltage Converter DC Voltage Ref: [16] A More Electric Aircraft (MEA) PE System was Simulated on FPGA Boards. Both Low-Frequency Dynamics and Switching-Frequency Behaviors were Emulated in High Modeling Accuracy. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in Converter Topology Derivation 52/137 Four-Port-1 … Four-Port-2 Six-Port-1 Six-Port-2 Eight-Port-1 Synthesize Synthesized Circuit Configuration Synthetic Topology Pool Ref: [17] RL Diagram Case 2. Reinforcement Learning (RL) Derived New Multiport DC-DC Converter Topologies for Lower Current and Voltage Stresses with Minimal Added Components. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in Converter Topology Derivation 53/137 Low Current Stress Hardware Waveforms of the Synthetic Topology with Low Current Stress Synthetic Topology Pool Hardware Prototype of the Synthesized Topology Output Voltage Waveforms Decoupling Control Waveforms Ref: [17] Hardware Validation of a Synthesized Low Current Stress Four-Port DC-DC Converter. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in Modulation Optimization 54/137 Particle Swarm Optimization Interfacing Data-Driven Models Modulation StrategyMulti-Level DAB Ref: [18] Case 3. Residual-Like Ensemble Learning Models Used Hybrid Simulation and Experimental Data for Accurate Efficiency Modeling and Modulation Optimization of a Dual-Active-Bridge (DAB) Converter. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in Modulation Optimization 55/137 97.4% 96.4% 95.4% η 98.4% 200 400 600 800 1000 1200 1400 1600 1800 2000 P (W) Highest η of D 2 EA: 98.451% Proposed D 2 EA Proposed D 2 EA Modeling with simulation data only Modeling with experimental data only Peak η: 98.208% Peak η: 98.236% Experimental Results and Optimality Validation Hardware of the Multi-Level DAB A Multi-Level DAB Converter Model Accuracy P = 600 W P = 1600 W Optimality Check Efficiency Ref: [18] Efficiency Modeling Accuracy Higher than 99.9%, Improved by ~1% Compared to Model Using Only Simulation Data. Modulation Optimality was Validated, Achieving a Highest Efficiency of 98.45%. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in Power Module Design: PowerSynth2 56/137 D2 D1 DC+DCDC-Out D2 D1 D4D3 DC-D2 Out DC+D1 Via Via DC+ DCOUT D2 D1 Case 4. PowerSynth2: A 3-D Power Module Layout Generation Tool PowerSynth2 Architecture 2D Half-Bridge Power Module 2.5D Full-Bridge Power Module 3D Half-Bridge Power Module https://e3da.csce.uark.edu /release/PowerSynth/ Ref: [19]-[21] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in Power Module Design: PowerSynth2 57/137 Simulated Annealing NSGA-II, MOPSO, etc. Design Flow of PowerSynth2 for Module Layout Generation Layout Design Considering CM Gain and Loop Inductance AI as Layout Optimizers Ref: [19]-[21] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 AI in PCB Routing 64/137 Hotspot Temperature 2℃↓ Thermal Modeling Error (℃) An Automatically Routed PCB and its Thermal Distribution Ref: [25] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 2.2 Key Takeaways –Section 2 65/137 Model-Based Design in Power Electronics Top-down design: System -> Converter -> Component -> Material Design space (design variables) and performance space (performance metrics) Why Power Electronics Design Needs AI Surging applications of power converters Conflicting and evolving performance demands for diverse applications AI in Power Electronics Design Case studies of AI in PE design: System architecture modelling, converter topology derivation, modulation optimization, power module design, magnetic design, PCB routing Shortage of talents APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li Seminar Outline –Before Coffee Break 66/137 1. Applications of AI in Power Electronics: An Overview AI in the Life Cycle Management of Power Converters One-Stop AI Solutions for Power Electronics in Wind Generation Systems 2. AI-Based Power Electronics Design: Present The Promise of AI for PE Design AI in Power Electronics Design 3. New Paradigm of Power Electronics Design Challenges of Existing AI Methods Applied in PE 30-Minute Break Physics-Informed Machine Learning in PE Design PE-GPT: A Generative AI-Based PE Design Paradigm 4. A Future of AI-Native PE Design: Autonomous, Ethical, and Green APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.1 Challenges of Existing AI Methods in PE 67/137 Challenge 1: Power Electronics Requires Data-Light & Computation-Light AI. Characteristics of Power Electronics Compared to Other Fields APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.1 Challenges of Existing AI Methods in PE 68/137 Challenge 1: Power Electronics Requires Data-Light & Computation-Light AI. Case: Efficiency Modeling of a Three-Level DAB Case: Startup Control of a CLLC-DAB Secondary side V 1 + - Q 1 Q 3 Q 4 Q 2 V 2 + - L k n:1 I 2 i L + - v p S 1 S 3 C 1 S 2 + - v s Primary side I 1 D 1 D 2 S 4 S 5 S 7 S 6 S 8 D 3 D 4 C 1 C 2 Simulation (12,500) Experiments (1000) 1250 (10%) 2500 (20%) 400 (40%) 200 (20%) 400 (40%) To learn XGBoost models To select hyperparameters of XGBoost models To validate generalization accuracy on unseen data TRAINING SET TRAINING SET TEST SET TEST SET VALIDATION SET VALIDATION SET Massive Train &Test Data Computation Model Accuracy Achievable Data-Driven Capacities Fall Short when Addressing More Complex Function! Conflicting Ref: [18], [26] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.1 Challenges of Existing AI Methods in PE 69/137 Challenge 2: Explainable AI is Essential to Enhance Deployment Confidence. Input-Output Analysis: Decision Traceback Analysis: AI should be Explainable in Mathematics and Power Electronics Domains! Param. 1 Effi. ηTrends of efficiency ηw.r.t. inputs Abnormal i<0 Normal v>x Normal Fault Ref: [27] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.1 Challenges of Existing AI Methods in PE 70/137 The inductor helps to smooth out the ripple in the output current by resisting sudden changes in current flow. Lower ripple percentages typically require larger inductors. AI: For the task to design LC filter for the buck converter which has minimum current ripple, why do you recommend this inductor value? User: “Leverage PE-GPT to Provide PE Insights into a Designed Buck Converter.” PE-GPT by F. Lin et al. Challenge 2: Explainable AI is Desirable to Enhance Deployment Confidence. 2. Power Electronics-Wise Explainable 1. Mathematically Explainable Training: Does it converge? Lyapunov stable? Trajectory analyzed? Infer: Are ML outputs 𝑓𝑓,∇𝑓𝑓 bounded? Controller: Is it stable? Asymptotically stable? Regret reduces to 0? Lipschitz continuity? Lipschitz ∇continuity? Β-smoothness? lim 𝑇𝑇𝑅𝑅(𝑇𝑇) 𝑇𝑇→0? 𝑅𝑅(𝑇𝑇) 𝑇𝑇=Ο� 1𝑇𝑇 Ref: [28]-[29] F. Lin et al., “PE-GPT: A New Paradigm for Power Electronics Design,” IEEE Trans. Ind. Electron., pp. 1–14, 2024. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.1 Challenges of Existing AI Methods in PE 71/137 ØTraditional data-driven is fixed in operational settings, threatening model feasibility when settings change. ØBuilding separate surrogate models for all settings is dataintensive and computationally demanding. ØHandling multiple settings need Nt=m×n×zmodels (imaging only 3 variables), increasing complexity significantly. ØTopology, circuit parameters, control strategies, performance metrics, operation specifications, etc., could all change. #Models Required for Various Operational Settings Challenge 3: Existing AI Lacks Flexibility for Diverse Scenarios. #Modulation Strategy (MS) #Operation Specifications (OS)#Performance Metrics (PM) MS 3 MS 1 MS 2 Optimization Space Ref: [30] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.1 Next Generation of AI for Power Electronics 72/137 NEXT GENERATION OF AI FOR POWER ELECTRONICS: The Ultimate Frontier of AI for PE is: Physics-Informed AI. AI + PHYSICS Ref: [31] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li Seminar Outline –After Coffee Break 73/137 1. Applications of AI in Power Electronics: An Overview 2. AI-Based Power Electronics Design: Present 3. New Paradigm of Power Electronics Design Challenges of Existing AI Methods Applied in PE Physics-Informed Machine Learning in PE Design Physics-in-Loss PIML: Concept and Case Study Physics-in-Architecture PIML: Basics, its Advantages, and Code Demo PE-GPT: A Generative AI-Based PE Design Paradigm 4. A Future of AI-Native PE Design: Autonomous, Ethical, and Green APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PIML for Multi-Physics Modeling: Case Study 80/137 Thermal Dynamics: Diffusion, Advection-Diffusion Fluid Dynamics: NavierStokes Equations hfins lfins tfins x y z GeometryCoordinates Tsolid TFluid PSolid Physics to Embed Optimizers Generate Structure Temperature / Pressure Fields Optimized Structure Ref: [33] Physics-informed neural network APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PIML for Multi-Physics Modeling: Case Study 81/137 Modeling Accuracy of PINN Models for Multi-Physics Simulation 0 1000 2000 3000 4000 5000 OpenFOAM FEM PINN ~17 × Faster 4 Nvidia V100 GPUs 20 CPU Processors Hours Ref: [33] Promising to Replace FEM! APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 Physics-In-Architecture Neural Network (PANN) in PE Design 82/137 u[tk] u[tk+1]y[tk+1] x[tk+1] Physics-in-Architecture Neural Network (PANN) + ho ( u[tk+1]; θ ) h ( u[tk]; θ )Δ tk+1 States at tk+1 go ( u[tk+1]; θ ) 1+g ( u[tk]; θ )Δ tk+1 delay + States at tk ( ) ( ) () (); () (); () dxt g ut xt h ut ut dt θθ = + ( ) ( ) () (); () (); () oo yt g ut xt h ut ut θθ = + Physics-in-Architecture Neural Network (PANN), (Submitted on 21 Jun 2023) Mamba State Space Models (Submitted on 1 Dec 2023) All Roads Lead to Destination, but PANN is a Shortcut for PE https://github.com/XinzeLee/PANN Circuit Dynamics Ref: [28]-[30], [34]-[36] Concept of Physics-in-Architecture Neural Network (PANN): APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 Physics-In-Architecture Neural Network (PANN) in PE Design 83/137 Ref: [28]-[30], [34]-[36] Fundamentals of PANN; Data-Light and Lightweight, Flexibility across Operating Conditions [34] X. Li et al., "Temporal Modeling for Power Converters With Physics-in-Architecture Recurrent Neural Network," in IEEE Trans. on Ind. Electron., vol. 71, no. 11, pp. 14111-14123, Nov. 2024. PANN for DAB Modulation Optimization; Data-Light, and Explainable in Power Electronics Contexts [35] X. Li, F. Lin, X. Zhang, H. Ma and F. Blaabjerg, "Data-Light Physics-Informed Modeling for the Modulation Optimization of a Dual-Active-Bridge Converter," in IEEE Trans. on Power Electron., vol. 39, no. 7, pp. 8770-8785, July 2024. PE-GPT: a Generative AI-Based PE Design Paradigm; PANN for Power Converter Design [28] F. Lin et al., "PE-GPT: A New Paradigm for Power Electronics Design," in IEEE Trans. Ind. Electron., pp. 1–14, 2024. PANN’s Flexibility across Diverse Performance Metrics [30] F. Lin, X. Li, X. Zhang and H. Ma, "STAR: One-Stop Optimization for Dual-Active-Bridge Converter With Robustness to Operational Diversity," in IEEE J. Emerg. Sel. Topics Power Electron., vol. 12, no. 3, pp. 2758-2773, June 2024. [36] X. Li et al., "A Generic Modeling Approach for Dual-Active-Bridge Converter Family via Topology Transferrable Networks," in IEEE Trans. on Ind. Electron., doi: 10.1109/TIE.2024.3406858. PANN’s Flexibility across Circuit Parameters and Topological Variants APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 Physics-In-Architecture Neural Network (PANN) in PE Design 84/137 Explainable Physics-Crafted Architecture •Intrinsic dynamics Power Electronics Explainable •Switching behaviors •Commutation Light Data-Light •Reduce data by 3 orders of magnitudes •Countable in one hand Lightweight •Deployable on edge •Down to kB level Flexible Training-Free to Outof-Domain Scenarios •Operating conditions •Modulation strategies •Performance metrics •Circuit parameters •Topology variants Advantages of PANN: Explainable, Light, and Flexible Ref: [28]-[30], [34]-[36] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 Physics-In-Architecture Neural Network (PANN) in PE Design 85/137 ( ) ( ) ( ) ( ) () (); () (); () () (); () (); () oo xt g ut xt h ut ut yt g ut xt h ut ut θθ θθ = +  = +    ( ) ( ) ( ) ( ) ( ) 11 1 1 11 11 [ ] 1 []; [] []; [] [] []; [] []; [] k k k k k kk k ok k ok k xt g ut t xt h ut ut t yt g ut xt h ut ut θθ θθ ∆∆ ++ + + ++ ++ =++  = +   ( ) 1 11 ; ...; ; k k kk k k xt xt xt ut t t φθ ∆∆ + ++      = +      Generic State-Space Equations of Power Converters Continuous Discrete, Recurrent Numerical Methods Discretize PANN is a Form of Neural Partial Differential Equations (NeuralPDEs) Ref: [28]-[30], [34]-[36] Process to Build PANN Architecture APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 Physics-In-Architecture Neural Network (PANN) in PE Design 86/137 Current states x[t k ]+ Inputs u[t k ], … Parameters θ Next states x[t k+1 ] PANN + Δt k+1 Neural Euler core ( ) ( );g ut θ θx[t k ] u[t k ] + ( ) ( );h ut θ + φ Increment function x[t k+1 ] Trainable Converter Parameters, Data-Driven Physically Explainable, Direct Assignment Switching behaviors Expert systems Custom Recurrent Neural Architecture of PANN out PANN has Custom Recurrent Structures to Embed Discretized State-Space Equations Ref: [28]-[30], [34]-[36] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PANN for Time-Domain Modeling: Case Study 87/137 + - s5s7 s8 s6 vo + - Lkn:1 + - v s FB1 s1s3 s4 s2 + - v p i L vi RL + - FB2 + - L k i L v p n·v s R L vs[tk+1] vp[tk+1] iL[tk]iL[tk+1] + - + - Lk n Δtk+1 k Lk LRt + +∆ 1 1 Equivalent Circuit of Inductor Discretize via Euler Algorithm Simplify A PANN Model for DAB Specs States x(t)iL Inputs u(t)vp, vs Circuit θRL, Lk, n Case Study: PANN for DAB Converters () () () () L k LL p s di t L R i t v t nv t dt +=− ( ) 1 1 11 1 [] [] [] [] []k Lk Lk pk sk LLk k t it it vt nvt Rit L ∆+ + ++ + −= − − ( ) ( ) 11 1 1 1 [] [] [] [] k p k s k kL k Lk k Lk t v t nv t L i t it L Rt ∆ ∆ ++ + + + −+ =+ Ref: [28]-[30], [34]-[36] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PANN for Time-Domain Modeling: Case Study 88/137 Expert System vp vs iL tktk+1 tk+2 Inference Breakdown (Code Demo*) "DAB-inference and training.ipynb" . . . . . . Animated Inference Breakdown XinzeLee/PANN Ref: [28]-[30], [34]-[36] PANN Inference: Predict the Next State APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PANN for Time-Domain Modeling: Case Study 89/137 vs vps pri(t)vi ssec(t)vo D0, D1, D2 vi, vo ssec(t) spri(t)1 00 -1 0 1 0 -1-1 Loop 1 Loop 2 iL(t) s1 s4 s5 s2 s3 s6 s7 s8 s1 s4 s5 s2 s3 s6 s7 s8 D1_cycle, D2_cycle iL iL + - vp + - vs Commutation Mode: vp=vi, vs=vo, iL>0 Commutation Mode: vp=0, vs=vo, iL<0 Expert System Switching Behaviors Commutation Loop Analysis PANN Pushes the Definition of AI Explainability to a NEW Height: PE Explainable. PANN Models Reveal Power Electronics Circuit Insights Ref: [28]-[30], [34]-[36] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PANN for Time-Domain Modeling: Case Study 96/137 PANN’s Flexibility across Topologies (Code Demo*, "DAB-Topology Transfer.ipynb") Piecewise Linear Sinusoidal Source: Non-resonant DAB with Single L,Lk= 63 μH Target: Resonant Lr= 63 μH, fr= 1.1·fs (Capacitive) Target: Resonant Lr= 63 μH, fr= 0.9·fs (Inductive) Target: Resonant Lr= 63 μH, fr= 1.0·fs (Resistive) Ref: [28]-[30], [34]-[36] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PANN for Time-Domain Modeling: Case Study 97/137 7 4 1-6 -4 -2 0 2 CASE III : Boost CASE II : Unit-gain CASE I : Buck T-sne feature 1 T-sne feature 2 Val.: Robustness to operational diversity Train Test CASE II : Unit-gainCASE I : Buck CASE III : Boost NUMBER OF DATA SAMPLES Train: Test: Val: 0 20 40 60 80 100 108, 100% 108, 100% 93, 86.1% 10, 9.3% 5, 4.6% 0 0 0 0 * * Intentionally Biased Data Partition 3 2 1 0 4 LSTM CNN TST PANN Figure Legends Mean Squared Loss L(θ) Good OOD Generalization to Case I and III Train-Unit Gain: Test-Unit Gain: NO OOD Capability Test-Unit Gain: Val.-Buck: Val.-Boost: Data-light Statistics of Condition Transferability Statistics of Modulation Transferability R2=99.68% R2=99.75% Statistical Study of PANN’s Flexibility Ref: [28]-[30], [34]-[36] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 PANN for Time-Domain Modeling: Case Study 98/137 Stats. of Performance Transferability Multi-PortSeries LCSingle L 6% Train to infer circuit parameters θ Direct assignment of θ 5% 4% 3% 2% 1% 0% Mean average percentage error (MAPE) 2.27% 3.52% iLiLvCiLiL1iL3 iL2 4.67% 3.13% 4.08% 5.3% 3.19% Multi-Level 95% 100% LSTM CNN TST PANN Accuracy of i pp , n ZVS , and n ZCS modeling High performance modeling accuracy i pp accuracy: n ZVS accuracy: n ZCS accuracy: 90% 85% 80% 75% Topology Transfer –From Lto LC -6 -2 2 6 i L * (t) i L (t) Time: 10 μs/div A -80 -40 0 40 V vC *(t) vC(t) Time: 10 μs/div Topology Transfer –From DAB to TAB Stats. of Topology Transferability -3 -1 1 3 i L1 * (t) i L1 (t) of PANN Time: 10 μs/div A -6 -2 2 6 iL2 *(t) iL2(t) of PANN A -5 0 5 10 i L3 * (t) i L3 (t) of PANN A -10 Statistical Study of PANN’s Flexibility Ref: [28]-[30], [34]-[36] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.2 Key Takeaways –Section 3: PIML Parts 99/137 Challenges of Existing AI Methods in Power Electronics Data and computation intensive, unexplainable, inflexible Physics-Informed Machine Learning (PIML) Concept: Steer learning towards physically consistent solutions Functionalities: Solve PDEs (forward), identify physical parameters (inverse) Classifications: Physics-in-loss, physics-in-architecture, physics-in-initialization Physics-in-Architecture Neural Network (PANN) Concept: Physics-crafted recurrent neurons to integrate discretized state-space equations Functionalities: Time-domain modeling (inference), identify circuit parameters (training) Advantages: Explainable in mathematics and PE, data-light (2 data samples are enough) and lightweight (~ 2 kB), flexible with zero-shot (training-free) generalization capabilities APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li Seminar Outline –After Coffee Break 100/137 1. Applications of AI in Power Electronics: An Overview 2. AI-Based Power Electronics Design: Present 3. New Paradigm of Power Electronics Design Challenges of Existing AI Methods Applied in PE Physics-Informed Machine Learning in PE Design PE-GPT: A Generative AI-Based PE Design Paradigm Fundamentals of PE-GPT: Challenges, RAG, Workflows (Multi-Modality) Code Demo of PE-GPT: Let’s Build the PE-GPT (Simplified) 4. A Future of AI-Native PE Design: Autonomous, Ethical, and Green APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 History of Natural Language Processing (NLP) 101/137 Predict the Next Word Becomes the Main Task of NLP Syntactic and rulebased structures Lexical taxonomy Probabilistic and statistical Sequence modeling Attention is All You Need “After ChatGPT (GPT-3.5), Large Language Models Mark New Era of Productivity.” Ref: [37] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Main Task of Large Language Models 102/137 ChatGPT Response: DAB converters are bidirectional DC-DC power converters that are widely used in high-efficiency energy transfer applications. Recurrently Predict the Next Word … ChatGPT Model Architecture DAB converters are bidirectional DC-DC power converters … DAB converters DAB converters are DAB converters are bidirectional DAB converters are bidirectional DC-DC DAB converters are bidirectional DC-DC power Other tokens Other tokens Other tokens Other tokens Other tokens 𝑝𝑝𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡+1 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡,𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡−1,…,𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡1 Sample from: APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Can LLMs Help PE Design Address New Challenges?103/137 New Materials and Interdiscipline More PE Applications and Increased System Complexity Digitalization and Perform. Trade-offs AI assisted modeling, optimization Can LLMs Independently Generate Full Designs? PE Design 4.0 - PE Autonomous Design PE Design 3.0 -Augmented Design Computer assisted computation & simulation PE Design 2.0 -Assisted Design Manual computation, analysis, and design Manual Design Modulation Design of DAB Converters as an Example Secondary Full Bridge q1q3 q4 q2 V2 + - Ln:1 + - vs Primary Full Bridge s3 s4 V1 + - + - vp s1 s2 iL RL 5DoF Strategy q4q3 q1q2 s4s3 s1s2 D0, D2, φ2 D1, φ1 PE Design 1.0 APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Capabilities of LLM for PE Design 104/137 Capabilities of LLM Chatbot and Q&A Understand Context and Summarize Retrieval Augmented Generation Fine-tune Model for Vertical Industries Function Calling and Routing Web Search Engine Code Generation Multi-Modality Planning and Organizing PE Design Tasks ØDesign Assistant, Trouble Shooting ØPE Reasoning, Design Overview and Logs ØIntegrate PE Knowledge Base, Design Reuse ØBespoke PE-Oriented LLM ØInvoke Design Rules and Perform Design Steps, Simulation, PE Data Processing ØCrawl PE-Related Materials (e.g., Datasheet) ØGenerate Control Code ØInterpret Circuit Schematic, Waveforms, Frequency Spectrum, Multi-Physics Models ØProject Management, Risk Analysis APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Off-the-Shelf LLMs Fall Short in PE Design 105/137 Challenge 1 -Limited PE Expertise Trained on General Corpora, LLMs Lack Domain Expertise. Challenge 2 -Incapability to Process PE-Specific Unstructured Data PE-Specific Data like Time-Series Waveforms / Frequency Spectrum, Multi-Physics FEA GPT 4, tested on 29th Sep 2024: It fails to specify a phase shift strategy due to limited expertise. (e.g. SPS, EPS, TPS, etc.) vs iL vp vp: [200 V / div] vs: [200 V / div] iL: [5 A / div] Time: [5 μs / div] V 60 V, 300 W, ( , , φ, φ) (0.7 , 0.88, 57%, 56.8%) s2 ZCS s1, s2, s3, s4 ZVS q1, q2, q3, q4 ZVS q3 ZCS q4 ZCS q1 ZCS Time-Series Waveforms Frequency Domain Challenge 3 –Isolated from the Existing Design Workflows / Tools Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Experimental Validation 112/137 Efficiency η 100% 50 100 300 500 700 P L /W 91% 97% Current stress i pp /A 30 24 18 12 6 SPS-η TPS-i pp SPS-i pp 900 85% 79% 73% 5DoF-i pp 5DoF-η 97.53% 4.9% 1.8% TPS-η Ø Optimal i pp ZVS 5DoF TPS 46666888888 888888 88888 50 100 200 300 400 500 600 700 800 900 1000 P L /W ZCS 5DoF TPS 0 0 002220000 22000000243 Ø Extended ZVS & ZCS ranges V 2 = 160 V v s i L v p v p : [200 V / div] v s : [200 V / div] i L : [5 A / div] Time: [5 μs / div] V 2 = 160 V, P L = 300 W, (D 1 , D 2 , φ 1 , φ 2 ) = (0.71, 0.88, 57%, 56.8%) s 2 ZCS s 1 , s 2 , s 3 , s 4 ZVS q 1 , q 2 , q 3 , q 4 ZVS q 3 ZCS q 4 ZCS q 1 ZCS v s i L v p V 2 = 160 V, P L = 600 W, (D 1 , D 2 , φ 1 , φ 2 ) = (0.78, 1, 50%, 51%) s 2 , s 1 ZCS s 1 , s 2 , s 3 , s 4 ZVS q 1 , q 2 , q 3 , q 4 ZVS v s i L v p V 2 = 160 V, P L = 1 kW, (D 1 , D 2 , φ 1 , φ 2 ) = (0.88, 1, 50%, 50%) s 1 , s 2 , s 3 , s 4 ZVS q 1 , q 2 , q 3 , q 4 ZVS i L : [10 A / div] Efficiency, Current Stress (CS), and Soft Switching of CSOptimized TPS and 5-DoF Modulation “Improves Efficiency, ZVS, and ZCS under Light Load Conditions.” Waveforms under Voltage Step-Down Scenarios Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Code Structure of PE-GPT 113/137 XinzeLee/PE-GPT main.py core/gui/design_stages.py core/llm/llm.py core/model_zoo/pann_net.py Important Code Blocks Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Let’s Build PE-GPT! 114/137 PE–GPT “Simplified Version of PE-GPT for APEC2025 tutorial, to Demonstrate RAG and FunctionTool.” FunctionTool Function calling and routing; Define tools to handle power electronics design tasks. RAG Integrate external knowledge bases to augment power electronics domain expertise. by F. Lin, X. Li et al. Ref: [28] Agentic RAG APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 LLM Orchestration for PE-GPT 115/137 External Knowledgebase Orchestration Framework (LlamaIndex, LangChain) Available LLMs RAG User Query Synthesized Response Pipeline Manager Load & Chunk Retrieved Documents/ Invoked Functions FunctionTool Vectorize Define Parse & Route & Invoke Simulate Validate the optimized D1and D2 in simulation Optimize Optimize D1and D2 for EPS of DAB with specified Vin, Vout, PL Embed Modulation of DAB q4q3 q1q2 s4s3 s1s2 DAB Modulation Knowledge Multi-Modal: Process PE Data Types (Waveforms) Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Implementing RAG with LlamaIndex 116/137 1. Chunking 2. Embedding 3. Retriever 4. Query 0. Loading RAG Steps Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Case Study: RAG 117/137 1. PE–GPT GPT-4.0 User Query VS Query 1 1. Query 2 User: I am looking for wide ZVS range with easy implementation. What modulation should I use for DAB converters? User: In DAB converters, what are the controllable parameters in the 5-DoF strategy? Please elaborate “Lack of knowledge for modulation design and 5-DoF modulation.” “Customized for DAB modulation design.” Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Case Study: RAG 118/137 “Keeping ‘AI for good’ as the mission, PEGPT aim to revolutionize the paradigm for diverse power electronics design tasks, and democratizing bespoke large language models for diverse applications in power electronics.” Let’s add the contents below into the knowledgebase, and test whether RAG captures the new information. Save new knowledge about PE-GPT into the knowledgebase. PE–GPT Query 3 User: What’s the initiative of PE-GPT? PE–GPT without new knowledge with new knowledge Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 llamaxindex FunctionTool: Function Calling and Routing 119/137 1. Define Each Task 2. Define Tools 4. Query Use FunctionTool to Achieve Multi-Modality and Interface Existing Design Tools 3. Aggregate Tools & Define Agent (1) Task Description: Clear & Separable (2) Task Executable: Multi-Modal, Math Define Tools with FunctionTool Class Optimize Simulate Query Context-Aware Tool Router Etc. Route & Invoke Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Case Study: llamaxindex Functiontool 120/137 Query 1 User: Please help me design EPS modulation for DAB converters. My Vi is 200 V, output voltage is 140 V, load power is 100 W.Query 2 User: Can you simulate the designed modulation for DAB converters? Invoke optimization for wide ZVS and minimal current stress. Parse & Route i pp Ø Current stress Ø Soft switching v p v s ZCS ZVS i L i L PE-GPT Metaheuristic Algorithm ZVS, ZCS, i pp Generate Candidates Objective Values i L (t k )i L (t k+1 ) + L v s (t k+1 ) v p (t k+1 )- +- n Δt k+1 Lk LR t + +∆ 1 1 Recurrently Used as Next v p (t k+1 ) v s (t k+1 ) i L (t k+1 ) i L (t k ) v p (t k ) v s (t k ) i L (t k ) i L (t k-1 ) Knowledge PANN for time-domain modeling Performance Evaluation D 0 , D 1 , D 2 , φ 1 , φ 2 Mod. Strategy; Spec. (V 1 , V 2 , P L ) Evolving Multi-Modality: Process Waveforms via PANN Optimization: Optimize both Current Stress and ZVS Invoke simulation for optimized inner phase shift. Route Waveforms from PANN Waveforms in Plecs -200 0 200 -100 0 100 -4 -2 0 2 4 Simulation Repository is Invoked via Text Query Ref: [28] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 3.3 Key Takeaways –Section 3: PE-GPT Parts 121/137 Large Language Models (LLM) and PE Design Challenges of general-purpose LLM: Limited PE expertise, incapable to process PE-specific data (like waveforms), isolated from existing design tools PE-GPT: A Generative AI-Based PE Design Paradigm PE-GPT framework: LLM agents, retrieval augmented generation (RAG), model zoo, simulation, and model fine-tuning RAG for knowledge embedding: Chunking -> Embedding -> Retrieval -> Query Multi-modality of PE-GPT: Time-domain modeling and converter model fine-tuning Workflows of PE-GPT: Design with model zoo, simulation repository, model fine-tuning Code Demo: Build Simplified PE-GPT Implement RAG with LlamaIndex (an LLM orchestration framework) Function calling and routing with LlamaIndex FunctionTool APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 4.2 Future 2: Paths towards Ethical AI in PE 128/137 Inference Stability Analysis Ref: [29] X. Li, F. Lin, H. A. Mantooth, and J. J. Rodríguez-Andina, “Explainable Physics-in-Architecture Neural Networks for Power Electronics: From a Lipschitz Continuity Perspective,” 2025, arXiv. 𝒙𝒙 𝑡𝑡𝑘𝑘+1 =1−𝐴𝐴Δ𝑡𝑡 −1 1−𝐴𝐴Δ𝑡𝑡 −1𝐵𝐵Δ𝑡𝑡 𝒙𝒙 𝑡𝑡𝑘𝑘 𝒖𝒖 𝑡𝑡𝑘𝑘+1 Params. 𝜽𝜽are Defined in Matrices A and B PANN: Formulations of PANN Definitions of Lipschitz Continuity To Mathematically Explain AI: From a Lipschitz Continuity Perspective 𝑓𝑓 𝒙𝒙1−𝑓𝑓 𝒙𝒙2≤𝐿𝐿1𝒙𝒙1−𝒙𝒙2,∀𝒙𝒙1,𝒙𝒙2∈𝑅𝑅𝑛𝑛 Validation of Inference Stability Validation of Training Convergence ‖∇𝜽𝜽𝑓𝑓(𝜽𝜽)‖≤𝐿𝐿1𝜽𝜽𝑦𝑦𝑦𝑦𝑡𝑡𝑦𝑦𝑦𝑦 � ⎯ ⎯ � ‖𝑓𝑓(𝜽𝜽𝟏𝟏)−𝑓𝑓(𝜽𝜽𝟐𝟐)‖≤𝐿𝐿1𝜽𝜽‖𝜽𝜽𝟏𝟏−𝜽𝜽𝟐𝟐‖ ‖∇ 𝑧𝑧 𝑓𝑓(𝒛𝒛)‖≤𝐿𝐿 1𝑧𝑧𝑦𝑦𝑦𝑦𝑡𝑡𝑦𝑦𝑦𝑦 � ⎯ ⎯ � ‖𝑓𝑓(𝒛𝒛 𝟏𝟏 )−𝑓𝑓(𝒛𝒛 𝟐𝟐 )‖≤𝐿𝐿 1𝑧𝑧 ‖𝒛𝒛 𝟏𝟏 −𝒛𝒛 𝟐𝟐 ‖ 0 Bounded 20 40 60 80 100 1 0.9995 0.999 Lipschitz Const. of Jacobian L 1z Theoretical L1z = 0.999996 MC Evaluation �∇ 𝒛𝒛𝑦𝑦𝑑𝑑𝑑𝑑 𝒙𝒙 � 𝑦𝑦𝑑𝑑𝑑𝑑 (𝒛𝒛 𝑦𝑦𝑑𝑑𝑑𝑑 )� Proof: Jacobian Matrix Norm is Bounded. Proof: Gradient Matrix Norm is Bounded. PANN has Bounded Jacobian Matrix Norm 𝜵𝜵𝒛𝒛𝒇𝒇(𝒛𝒛) APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 4.2 Future 2: Paths towards Ethical AI in PE 129/137 Regret of PANN ~ 𝜪𝜪(𝑻𝑻) 𝐥𝐥𝐥𝐥𝐥𝐥 𝑻𝑻→∞𝑹𝑹(𝑻𝑻) 𝑻𝑻=𝟎𝟎 lim 𝑇𝑇𝑅𝑅 𝑇𝑇 𝑇𝑇=� 𝑡𝑡=1 𝑇𝑇𝑓𝑓𝑡𝑡𝜽𝜽𝒕𝒕−𝑓𝑓𝑡𝑡𝜽𝜽∗ 𝑇𝑇=𝟎𝟎 Prove Training Convergence: R(T): Regret T: Epochs Algorithm Experiments to Validate the Training Convergence of PANN 𝑓𝑓𝑡𝑡𝜽𝜽𝒕𝒕: Loss Function at time t Regret R(T)Average Regret R(T)/T T Ref: [29] Training Convergence Analysis 𝑓𝑓 𝜽𝜽 = 0.5 � 𝒙𝒙−𝒙𝒙∗ 𝑇𝑇 � 𝒙𝒙−𝒙𝒙∗ ∇𝜽𝜽𝑓𝑓 𝜽𝜽 =𝜕𝜕𝑓𝑓 𝜕𝜕� 𝒙𝒙𝜕𝜕� 𝒙𝒙 𝜕𝜕𝑊𝑊𝜕𝜕𝑊𝑊 𝜕𝜕𝜽𝜽 =� 𝒙𝒙−𝒙𝒙∗ 𝑇𝑇 𝒛𝒛𝜕𝜕𝑊𝑊 𝜕𝜕𝜽𝜽 ∇𝜽𝜽𝑓𝑓 𝜽𝜽 ≤ 𝑊𝑊−𝑊𝑊∗� 𝒛𝒛 2�𝜕𝜕𝑊𝑊 𝜕𝜕𝜽𝜽 =𝐿𝐿1𝜽𝜽 To Mathematically Explain AI: From a Lipschitz Continuity Perspective X. Li, F. Lin, H. A. Mantooth, and J. J. Rodríguez-Andina, “Explainable Physics-in-Architecture Neural Networks for Power Electronics: From a Lipschitz Continuity Perspective,” 2025, arXiv. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 4.3 Future 3: Green AI with Energy Awareness Consumption vs Performance Gains Consumption Evaluation of AI Systems Q: Is Neural Architecture Search (NAS) Green? Light model Select Train Evaluate Refine AI Development Process 𝑃𝑃𝐴𝐴𝐴𝐴 =𝜔𝜔𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑛𝑛𝑃𝑃𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑛𝑛 +𝜔𝜔𝑒𝑒𝑒𝑒𝑡𝑡𝑙𝑙𝑃𝑃𝑒𝑒𝑒𝑒𝑡𝑡𝑙𝑙 +𝜔𝜔𝑡𝑡𝑒𝑒𝑟𝑟𝑡𝑡𝑛𝑛𝑒𝑒𝑃𝑃𝑡𝑡𝑒𝑒𝑟𝑟𝑡𝑡𝑛𝑛𝑒𝑒 +𝜔𝜔𝑑𝑑𝑒𝑒𝑜𝑜𝑙𝑙𝑜𝑜𝑑𝑑𝑃𝑃𝑑𝑑𝑒𝑒𝑜𝑜𝑙𝑙𝑜𝑜𝑑𝑑 Deploy “Power Consumption of Developing AI should be Considered.” Performance Gains Consumption Ref: [39] 130/137 Energy Breakdown of a PE-GPT Design Case with a GPT-4o-mini APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 4.3 Future 3: Paths towards Green AI in PE 131/137 Only Use AI when We Truly Need it Optimize Energy Efficiency with PE Minimize the Computation of AI Establish Regulations and Foster Collaboration Reconsider whether AI is the best tool to address specific PE tasks. Energy consumption can be reduced through algorithms, architectures, and hardware. Advanced power electronics can help reduce energy consumption in return. Establish regulations to promote awareness and encourage opensource collaboration. Image Credit to Open AI GPT-4o: Trade Off Between AI Gains and Energy Consumption. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 5. References and Further Readings 132/137 [1] S. Zhao, F. Blaabjerg,and H. Wang, “An Overview of Artificial Intelligence Applications for Power Electronics,” IEEE Trans.Power Electron.,vol.36,no. 4, pp.4633–4658,Apr.2021,doi: 10.1109/TPEL.2020.3024914. [2] F. Blaabjerg,M.Chen, and L. Huang, “Power electronics in wind generation systems,” Nat Rev Electr Eng,vol. 1, no.4,pp.234–250,Mar.2024,doi:10.1038/s44287-024-00032-x. [3] K. Hornik, M. Stinchcombe, and H. White, “Multilayer feedforward networks are universal approximators,” Neural Networks,vol. 2, no.5,pp.359–366,Jan.1989,doi:10.1016/0893-6080(89)90020-8. [4] Asimov Institute, “The Neural Network Zoo.” [Online].Available:https://www.asimovinstitute.org/neural-network-zoo/ [5] Y. Zhang, H. Wang, Z. Wang, F. Blaabjerg,and M. Saeedifard,“Mission profile-based system-level reliability prediction method for modular multilevel converters,” IEEE Trans.on Power Electron., vol.35, no.7,pp.6916-6930,Jul.2020,doi:10.1109/TPEL.2019.2957826. [6] J. Chen, J. Chen, and C. Gong, “New Overall Power Control Strategy for Variable-Speed Fixed-Pitch Wind Turbines Within the Whole Wind Velocity Range,” IEEE Trans.Ind.Electron.,vol.60,no.7,pp. 2652–2660,Jul.2013,doi:10.1109/TIE.2012.2196901. [7] C. Wei, Z. Zhang, W. Qiao,and L. Qu, “An Adaptive Network-Based Reinforcement Learning Method for MPPT Control of PMSG Wind Energy Conversion Systems,” IEEE Trans.Power Electron.,vol.31, no.11,pp.7837–7848,Nov.2016,doi:10.1109/TPEL.2016.2514370. [8] F. Lin, X. Zhang, and X. Li, “Design Methodology for Symmetric CLLC Resonant DC Transformer Considering Voltage Conversion Ratio, System Stability, and Efficiency,” IEEE Trans.Power Electron.,vol. 36,no.9,pp.10157–10170,Sep.2021,doi:10.1109/TPEL.2021.3059852. [9] J. Liu, F. Qu, X. Hong, and H. Zhang, “A Small-Sample Wind Turbine Fault Detection Method With Synthetic Fault Data Using Generative Adversarial Nets,” IEEE Trans.Ind.Inf.,vol.15,no.7,pp.3877– 3888,Jul.2019,doi:10.1109/TII.2018.2885365. [10] S. Zhao and H. Wang, “Enabling Data-Driven Condition Monitoring of Power Electronic Systems With Artificial Intelligence:Concepts, Tools, and Developments,” IEEE Power Electron.Mag.,vol. 8, no. 1, pp.18–27,Mar.2021,doi:10.1109/MPEL.2020.3047718. [11] S. Zhao, S. Chen, F. Yan g, E. Ugur, B. Akin, and H. Wang, “A Composite Failure Precursor for Condition Monitoring and Remaining Useful Life Prediction of Discrete Power Devices,” IEEE Trans.Ind.Inf., vol.17,no. 1, pp.688–698,Jan.2021,doi:10.1109/TII.2020.2991454. [12] P. Wilson and H. A. Mantooth,Model-based engineering for complex electronic systems.Newnes,2013. [13] H. A. Mantooth and M. Vasić,“Design Automation in Power Electronics,”IEEE Design Automation for Power Electronics Workshop,Feb.2019,[Online].Available:https://resourcecenter.ieeepels.org/conferences/proceedings/pelspro0021 [14] J. W. Kolar, “Power Electronics 4.0.” Expert Discussion on “Design Automation and Next Generation Measurement Technologies in Power Electronics,” Jul.2019.[Online].Available:https://www.amspublications.ee.ethz.ch/uploads/tx_ethpublications/workshop_publications/7_ECPE_Design_Automation_Expert_Discussion_JWK_as_published_070819.pdf [15]Boston Consulting Group, “The US Needs More Engineers.What’s the Solution?” 2023.[Online].Available:https://www.bcg.com/publications/2023/addressing-the-engineering-talent-shortage [16] S. Zhang, T. Liang, and V. Dinavahi,“Machine Learning Building Blocks for Real-Time Emulation of Advanced Transport Power Systems,” IEEE Open J. Power Electron.,vol. 1, pp.488–498,2020,doi: 10.1109/OJPEL.2020.3039117. [17] M. Dong, R. Liang, J. Yang, C. Xu, D. Song, and J. Wan, “Topology Derivation of Multiport DC–DC Converters Based on Reinforcement Learning,” IEEE Trans.Power Electron.,vol.38,no. 4, pp.5055– 5064,Apr.2023,doi:10.1109/TPEL.2023.3235053. [18] X. Li et al.,“Data-Driven Modeling With Experimental Augmentation for the Modulation Strategy of the Dual-Active-Bridge Converter,” IEEE Trans.Ind.Electron.,vol.71,no. 3, pp.2626–2637,Mar. 2024,doi:10.1109/TIE.2023.3265027. [19] T. M. Evans et al., “PowerSynth: A Power Module Layout Generation Tool,” IEEE Trans.Power Electron.,vol.34,no.6,pp.5063–5078,Jun.2019,doi:10.1109/TPEL.2018.2870346. [20] I. Al Razi, Q. Le, T. M. Evans, H. A. Mantooth,and Y. Peng, “PowerSynth 2: Physical Design Automation for High-Density 3-D Multichip Power Modules,” IEEE Trans.Power Electron.,vol.38,no. 4, pp. 4698–4713,Apr.2023,doi:10.1109/TPEL.2022.3227300. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 5. References and Further Readings 133/137 [21] Q. Le, I. A. Razi, T. M. Evans, S. Mukherjee, Y. Peng, and H. A. Mantooth,“Fast and Accurate Parasitic Extraction in Multichip Power Module Design Automation Considering Eddy-Current Losses,” IEEE J. Emerg.Sel.Topics Power Electron.,vol.11,no.6,pp.5613–5625,Dec.2023,doi:10.1109/JESTPE.2022.3175150. [22] T. Guillod, P. Papamanolis,and J. W. Kolar, “Artificial Neural Network (ANN) Based Fast and Accurate Inductor Modeling and Design,” IEEE Open J. Power Electron.,vol. 1, pp.284–299,2020,doi: 10.1109/OJPEL.2020.3012777. [23] H. Li et al.,“How MagNet:Machine Learning Framework for Modeling Power Magnetic Material Characteristics,” IEEE Trans.Power Electron.,vol.38,no.12,pp.15829–15853,Dec.2023,doi: 10.1109/TPEL.2023.3309232. [24] M. Chen et al., “MagNet Challenge for Data-Driven Power Magnetics Modeling,” IEEE Open J. Power Electron.,pp. 1–16,2024,doi:10.1109/OJPEL.2024.3469916. [25] T. Chen, S. Xiong, H. He, and B. Yu, “TRouter:Thermal-Driven PCB Routing via Nonlocal Crisscross Attention Networks,” IEEE Trans.Comput.-Aided Des.Integr.Circuits Syst.,vol.42,no.10,pp.3388– 3401,Oct.2023,doi:10.1109/TCAD.2023.3243544. [26] Z. Xiao, X. Li, and Y. Tang, “A Lightweight Artificial Neural Network Start-Up Controller for CLLC Resonant Converters,” IEEE Trans.Power Electron.,vol.39,no.11,pp.14775–14786,Nov.2024,doi: 10.1109/TPEL.2024.3436847. [27] G. Leopold, “Opening Up Black Boxes With Explainable AI.” [Online].Available:https://www.datanami.com/2018/05/30/opening-up-black-boxes-with-explainable-ai/ [28] F. Lin et al.,“PE-GPT: A New Paradigm for Power Electronics Design,” IEEE Trans.Ind.Electron.,pp. 1–14,2024,doi:10.1109/TIE.2024.3454408. [29] X. Li, F. Lin, H. A. Mantooth,and J. J. Rodríguez-Andina, “Explainable Physics-in-Architecture Neural Networks for Power Electronics:From aLipschitz Continuity Perspective,” 2025,arXiv. [30] F. Lin, X. Li, X. Zhang, and H. Ma, “STAR:One-Stop Optimization for Dual-Active-Bridge Converter With Robustness to Operational Diversity,” IEEE J. Emerg.Sel.Topics Power Electron.,vol.12,no. 3, pp.2758–2773,Jun.2024,doi:10.1109/JESTPE.2024.3392684. [31] X. Li, “Next Generation of AI for Power Electronics:Explainable, Light, and Flexible,” Nov.2024,doi:10.5281/ZENODO.14036281. [32] G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,” Nat Rev Phys,vol. 3, no. 6, pp.422–440,May 2021,doi:10.1038/s42254-021-00314-5. [33]Modulus Contributors, “NVIDIA Modulus:An open-source framework for physics-based deep learning in science and engineering.” Feb.2023.[Online].Available:https://github.com/NVIDIA/modulus [34] X. Li et al.,“Temporal Modeling for Power Converters With Physics-in-Architecture Recurrent Neural Network,” IEEE Trans.Ind.Electron.,vol.71,no.11,pp.14111–14123,Nov.2024,doi: 10.1109/TIE.2024.3352119. [35] X. Li, F. Lin, X. Zhang, H. Ma, and F. Blaabjerg,“Data-Light Physics-Informed Modeling for the Modulation Optimization of aDual-Active-Bridge Converter,” IEEE Trans.Power Electron.,vol.39,no. 7, pp.8770–8785,Jul.2024,doi:10.1109/TPEL.2024.3378184. [36] X. Li et al.,“A Generic Modeling Approach for Dual-Active-Bridge Converter Family via Topology Transferrable Networks,” IEEE Trans.Ind.Electron.,pp. 1–13,2024,doi:10.1109/TIE.2024.3406858. [37] M. Appé,“The Evolution of NLP.” Oct.2023.[Online].Available:https://blog.dataiku.com/nlp-metamorphosis [38] X. Zhou et al., “ElecBench: a Power Dispatch Evaluation Benchmark for Large Language Models,” 2024,arXiv.doi:10.48550/ARXIV.2407.05365. [39] Y. Li, M. Mughees, Y. Chen, and Y. R. Li, “The Unseen AI Disruptions for Power Grids: LLM-Induced Transients,” 2024, arXiv. doi: 10.48550/ARXIV.2409.11416. APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 5. References and Further Readings 134/137 Handouts: Eat this Fish and Master AI for PE Design More Code Tutorials Here APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li 6. Conclusions 135/137 Overview of AI in the Lifecycle of Power Converters AI has been applied in the design, control, and maintenance of power converters, serving as surrogate models, controllers, optimizers, fault classifiers, and RUL estimators. AI in Power Electronics Design: What’s the Present AI is omnipresent in PE design: including system-level, converter-level, and component-level designs. Next Generation of AI in PE Design: Physics-Informed and Generative AI-Empowered PIML, which supports the data-driven capacity of AI by integrating PE physics, is leveraged to solve PDEs and identify parameters, paving the ways for the next-generation PE-wise explainability and light and flexible AI. PE-GPT marks new PE design paradigm with multi-modal generative AI directing interactivity at a high level. Future of AI-Native PE Design: Autonomous, Ethical, and Green Multi-agent generative AI to empower autonomous PE design. Ethical use of AI for security, fairness, and transparency in PE design. Build green AI with energy awareness. Thank You for Attending! Q&A 1University of Arkansas 2Aalborg University Alan Mantooth1[email protected]u Frede [email protected] Xinze [email protected] APEC2025 S.05 · AI in PE Design: Present and Future ·Alan Mantooth, Frede Blaabjerg, Xinze Li