Senior design 2026 – electrical engineer team 2637

SDP Final Presentationhttps://uconn.sharepoint.com/sites/SDPTeam46-AITestScripts/_layouts/15/stream.aspx?id=%2Fsites%2FSDPTeam46%2DAITestScripts%2FShared%20Documents%2FGeneral%2FSDP%20Team%20Video%2Emov&referrer=StreamWebApp%2EWeb&referrerScenario=AddressBarCopied%2Eview%2E150d3f12%2D7a9a%2D484a%2D8cd1%2De4e94263e65b

The project requires a NDA however the information made public within the project is not sensitive but contains CUI data

Team number:2637

Faculty advisor: sung-yeul park
ECE Team members names: Halmar Laing Elizabeth Berry
adjacent team : CSE

Project Description
AI-Assisted Automation of Modeling and Simulation Workflows for Power Electronic and Motor Drive Systems

Modern electrical engineering systems particularly in power electronics and motor control rely heavily on high-fidelity modeling and simulation to design, validate, and optimize performance. Tools such as MATLAB and Simulink are widely used to construct plant models, simulate dynamic behavior, and evaluate control strategies. However, despite the availability of these tools, the process of analyzing simulation results remains largely manual, time-consuming, and dependent on expert interpretation. Engineers must repeatedly tune parameters, run simulations, extract relevant signals, compute performance metrics, and interpret system behavior, often across hundreds of experimental scenarios. This workflow creates bottlenecks in productivity and limits scalability, especially when dealing with complex electromechanical systems.

This project addresses these challenges by developing an AI-assisted framework for automating the analysis of modeling and simulation workflows for power electronic converters and Permanent Magnet Synchronous Motor (PMSM) drive systems. The core objective is to integrate structured simulation data with an offline large language model (LLM) to create a system capable of interpreting plant model behavior, identifying relationships between system parameters and performance, and assisting engineers in decision-making processes.

Motivation and Problem Statement

In traditional simulation-based workflows, engineers must manually analyze system responses to changes in plant parameters and control inputs. For example, in a PMSM motor drive system, parameters such as stator resistance, inductance, rotor inertia, and back-EMF constant directly influence system dynamics. Similarly, in power electronic converters such as buck or boost converters, parameters such as switching frequency, inductance, and capacitance determine output voltage ripple, efficiency, and transient response.

While simulation tools provide the raw data needed for analysis, they do not inherently provide insight into why certain behaviors occur or how system performance changes as parameters vary. Engineers must rely on domain knowledge and experience to interpret simulation outputs, which introduces variability and inefficiency. Furthermore, simulation results are often stored in unstructured formats, making it difficult to reuse data for further analysis or automation.

The central problem addressed in this project is:

How can simulation data from plant models be structured, standardized, and leveraged to enable automated, intelligent analysis of system behavior using AI?

Project Objectives

The primary objectives of this project are:

Automate the extraction and structuring of simulation data from MATLAB/Simulink plant models.
Standardize simulation outputs to create consistent, machine-readable datasets suitable for AI training.
Develop a data-cleaning and preprocessing pipeline that ensures data quality, consistency, and physical interpretability.

Integrate an offline large language model (LLM) to interpret simulation data and generate engineering insights.
Enable AI-assisted reasoning about system performance, parameter sensitivity, and dynamic behavior.
Demonstrate the framework on real engineering systems, including power electronic converters and PMSM motor drives.

System Overview

The developed system consists of a multi-stage pipeline that transforms raw simulation models into structured data and ultimately into AI-driven analysis. The workflow can be summarized as follows:

Plant Model Simulation (MATLAB/Simulink)
SLX-to-Text Conversion (Model Metadata Extraction)
Data Export (JSON Format)
Data Cleaning and Standardization
Data Structuring and Labeling
LLM Integration and Training
AI-Based Analysis and Interpretation

Each stage plays a critical role in ensuring that the final system is both accurate and scalable.

Plant Models and Simulation Environment

The project utilizes a variety of plant models implemented in MATLAB/Simulink, focusing on two primary domains:

1. Power Electronics Systems
Buck converters
Boost converters
Input voltage disturbance (ride-through) studies

These models are used to analyze:

Voltage regulation
Efficiency
Ripple characteristics
Transient response
2. Motor Drive Systems
PMSM Field-Oriented Control (FOC) models
Speed control and load disturbance scenarios
Parameter estimation studies

These models provide insights into:
Speed tracking performance
Torque production
Current behavior
Stability and control dynamics

Simulation scenarios include parameter sweeps, step responses, and disturbance injections, allowing for comprehensive evaluation of system behavior under varying conditions.

SLX-to-Text Conversion

A key innovation in this project is the development of a method to convert Simulink .slx files into structured, human-readable text representations. Since .slx files are binary and cannot be directly interpreted by language models, a MATLAB script (describe_simulink_model) is used to extract key metadata from each model.

The extracted information includes:

Model name and file path
Solver type and simulation settings
Top-level input and output ports
Block inventory and system structure

This information is stored in a “model card” format, which serves as a bridge between the simulation environment and the AI system. By converting plant models into structured text, the system enables the LLM to reason about model architecture without requiring direct access to Simulink.

Data Extraction and JSON Structuring

Simulation outputs are exported into structured JSON files, which contain both system parameters and time-series data. Each JSON file corresponds to a simulation run and includes:

Plant parameters (e.g., Rs, Ld), J, Ke
Simulation configuration (e.g., duration, sampling time)
Logged signals (e.g., speed, torque, current)
Derived metrics (e.g., ripple, efficiency)

This structured format enables consistent storage and retrieval of simulation data, forming the foundation for subsequent data processing and AI training.

Data Cleaning and Standardization

Raw simulation data is not immediately suitable for AI training. A data-cleaning pipeline is implemented to ensure that all datasets are consistent, accurate, and physically meaningful. The cleaning process includes:

Unit normalization (e.g., converting all speeds to rpm or rad/s)
Signal standardization (consistent naming conventions)
Removal of redundant or irrelevant data
Handling missing or corrupted values
Validation of parameter ranges

This step is critical because inconsistencies in the data can lead to incorrect or unreliable AI outputs. By enforcing strict data standards, the system ensures that the LLM learns from high-quality, meaningful information.

Data Linkage and Physical Interpretation

A major contribution of this project is the explicit linkage between simulation data and physical system behavior. Each parameter and signal is mapped to its physical meaning and impact on system performance.

For example:

Stator resistance influences copper losses and efficiency
Rotor inertia affects acceleration and settling time
Inductance impacts current dynamics and torque ripple

By embedding this domain knowledge into the dataset, the system enables the LLM to move beyond pattern recognition and toward true engineering reasoning.

LLM Integration and Training

The cleaned and structured datasets are used to train an offline large language model (Mistral 7B). The LLM is not used as a generic chatbot, but rather as a specialized tool for interpreting engineering data.

Training data consists of:

Simulation inputs (parameters and configurations)
Output metrics (KPIs)
Natural language prompts describing experiments
Expected analytical responses

The model is trained to:

Interpret parameter changes
Analyze system responses
Explain cause-and-effect relationships
Provide engineering insights

Importantly, the system operates in an offline and secure environment, ensuring compatibility with industry requirements for sensitive data handling.

Applications and Impact

The developed framework has significant implications for engineering workflows:

Reduced analysis time through automation
Improved consistency in interpreting simulation results
Enhanced scalability for large parameter studies
Support for less-experienced engineers through AI assistance

Potential applications include:

Power system design optimization
Motor control tuning
Fault analysis and diagnostics
Rapid prototyping and validation

Conclusion
This project demonstrates how artificial intelligence can be integrated with traditional engineering tools to enhance modeling and simulation workflows. By transforming raw simulation data into structured, interpretable datasets and leveraging a domain-aware LLM, the system enables automated analysis of complex electromechanical systems.

The result is a scalable, secure, and intelligent framework that bridges the gap between simulation and engineering insight, paving the way for future advancements in AI-assisted design and analysis in electrical engineering.