AI-Driven Adaptive Model Simplification Framework for Teaching Reform in Mechanical Engineering

Abstract

We propose an AI-driven adaptive model simplification framework to reform mechanical engineering education by dynamically tailoring the complexity of simulations to individual learning needs. Traditional teaching methods often rely on static simplifications that either overwhelm students with excessive detail or oversimplify critical physical behaviors, hence limiting effective learning. The proposed framework integrates symbolic regression and unsupervised clustering to automatically reduce high-fidelity mechanical models while preserving their essential dynamics, thereby enabling students to interact with tractable yet accurate representations. A symbolic regression engine identifies dominant terms in governing equations, while spectral clustering groups similar subsystems to further reduce dimensionality. Moreover, real-time feedback loops adjust the level of abstraction based on student performance, measured through a recurrent neural network that predicts comprehension levels from interaction data. The framework replaces instructor-defined simplifications with a data-driven approach, ensuring personalized learning experiences without manual intervention. Implemented using state-of-the-art tools such as PySINDy and scikit-learn, the system demonstrates how AI can bridge the gap between theoretical complexity and pedagogical accessibility. This work contributes a novel paradigm for mechanical engineering education, where adaptive model simplification fosters deeper understanding by aligning simulation fidelity with student proficiency. The results highlight the potential of AI to transform traditional curricula into dynamic, student-centered learning environments.