Power Converter Design Workflows with Multimodal AI and Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models (LLMs) lack expertise in power electronics, struggle with unstructured data, and are isolated from power converter design workflows, leading to inefficiencies and suboptimal results in power converter design.
Innovation Solution
Integrate multimodal large language models (MLLMs) with power converter design workflows, utilizing structured design data, user queries, and real-time data to perform similarity matching, extract parameters, and determine design workflows, incorporating surrogate models for optimization and simulation validation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing LLMs are used for power converter design, then general-purpose language processing capability is provided, but domain expertise in power electronics is lacking
Solution Approach 1:
The patent fine-tunes the LLM by training it on domain-specific power electronics datasets, changing the model's parameter weights to adapt to power converter design tasks. This transforms a general-purpose LLM into a domain-expert model capable of providing accurate technical recommendations for power electronics.
Solution Approach 2:
The patent introduces a retrieval-augmented generation (RAG) system that acts as an intermediary between the LLM and power electronics knowledge bases. The RAG system retrieves relevant domain-specific information and provides it to the LLM, enabling the model to access up-to-date power electronics knowledge without retraining.
2Adaptability or versatility
If existing LLMs are used for power converter design, then language processing capability is provided, but ability to process unstructured data such as time-series waveforms and frequency spectra is limited
Solution Approach 1:
The patent merges the LLM with specialized analysis tools including Fourier transform capabilities for frequency spectrum analysis and time-series processing functions. This combination enables the system to handle unstructured power electronics data such as operational waveforms and spectral information with high precision.
Solution Approach 2:
The patent creates a composite AI system that integrates the LLM with domain-specific analysis models and simulation tools. This composite architecture combines the language understanding strengths of the LLM with the analytical precision of specialized algorithms for processing waveforms, spectra, and other unstructured power electronics data.
3Ease of operation
If existing LLMs are used for power converter design, then independent language model functionality is provided, but integration with simulation tools such as MATLAB, PLECS, and Ansys is lacking
Solution Approach 1:
The patent enhances the LLM with multi-functionality by integrating it with simulation tools (MATLAB, PLECS, Ansys), component selection databases, and performance optimization algorithms. This universal system can perform language processing, circuit simulation, component selection, and optimization within a single unified workflow, eliminating the need for separate tools.
Solution Approach 2:
The patent implements feedback loops where simulation results from integrated tools are automatically fed back to the LLM for analysis and optimization. This closed-loop system allows the LLM to learn from simulation outcomes and iteratively improve design parameters, enhancing both workflow integration and design productivity.
Data Source
AI summary
A system and a method for generating an optimized power converter design based on one or more multimodal large language models (MLLMs) are disclosed. The system obtains structured design data, one or more user queries, one or more design parameters, and real-time power converter design data. The system performs a similarity matching between one or more embeddings associated with the structured design data, and the one or more user queries and the one or more design parameters to compute an utmost similarity score. The system retrieves one or more similar document chunks. The system processes the one or more user queries and the one or more design parameters to extract one or more parameters. The system determines one or more design workflows based on the one or more parameters. The system executes at least one design workflow. The system visualizes design outcomes, simulation results, and prediction results to a user.