Physics Didactics in the Era of Artificial Intelligence: A Machine Learning Framework for Conceptual Modeling in Mechanics

Abstract

Teaching introductory mechanics in higher education remains challenging because students often struggle to develop genuine conceptual understanding beyond procedural problem-solving. The abstraction of mathematical formalism frequently overshadows the physical intuition necessary for engineering practice. This paper proposes a pedagogical framework that integrates computational modeling and machine learning to support conceptual learning in physics. The approach centers on a nonlinear damped pendulum a canonical system rich in dynamical behavior yet analytically intractable to shift student focus from symbolic derivation to physical exploration. Students construct Python-based computational models, generate numerical datasets across parameter regimes, and train a Multilayer Perceptron to predict the next state dynamics from simulated trajectories. This inversion of the traditional learning sequence positions the Multilayer Perceptron as a pattern discovery tool that students must interpret physically. The instructional design is grounded in Cognitive Load Theory, Constructionism, and Didactic Transposition, each addressing a distinct dimension of the learning process: managing cognitive demands, learning through model-building, and reconciling formal physics with executable representations. We present a proof-of-concept instructional design and propose an experimental evaluation protocol combining pre-test and post-test conceptual assessments, cognitive load measurements, and qualitative analysis of student computational artifacts. Expected outcomes include reduced extraneous cognitive load, improved conceptual mapping between simulation parameters and physical behavior, and stronger epistemic connections between academic physics and engineering practice. By framing artificial intelligence as a cognitive amplifier rather than a replacement for scientific reasoning, this work advances a human in the loop paradigm for physics instruction and offers a replicable model for integrating data driven methods into science, technology, engineering, and mathematics curricula.

Keywords: Physics Didactics; Machine Learning; Computational Modeling; Conceptual Learning; Cognitive Load; Python Programming; Mechanics Education; Artificial Intelligence.

Available Online at: https://www.ijsrisjournal.com/index.php/ojsfiles/article/view/764