Teaching Predictive Maintenance Through Aircraft Engine Degradation Modeling: A Simulation-and-Machine-Learning Case Study for Engineering Education

Abstract

Introduction: The rapid integration of artificial intelligence into industrial maintenance practices has created a pressing need for engineering curricula that prepare students to work with data-driven prognostic systems. Predictive maintenance, particularly remaining useful life (RUL) estimation for aircraft engines, is a useful teaching domain because it integrates simulation, machine learning, engineering interpretation, and decision-making under uncertainty.

Objective: This study designs, implements, and evaluates a project-based learning module that uses aircraft-engine degradation modeling to teach predictive maintenance concepts to undergraduate engineering students.

Method: A structured case study was developed around the NASA C-MAPSS turbofan-engine degradation dataset and a Python-based simulation environment. Thirty-eight mechanical and aerospace engineering students participated in a twelve-week implementation. Learning outcomes were assessed through pre- and post-module concept inventories, project deliverables, and a student-experience survey. Quantitative performance was also described in relation to a historical cohort that received conventional instruction on the same topic.

Results: Students in the simulation-and-machine-learning cohort demonstrated statistically significant gains in conceptual understanding (p < 0.001, Cohen’s d = 1.42) and completed projects that connected RUL predictions with maintenance recommendations. Survey responses indicated strong perceived relevance and engagement, with 87% of participants reporting that the aircraft-engine case made abstract machine-learning concepts tangible.

Conclusion: Aircraft-engine degradation modeling offers a potentially scalable framework for teaching predictive maintenance in engineering education. The results are promising, but the historical comparison and single-institution setting limit causal interpretation and generalizability. The module is therefore best understood as a replicable instructional design supported by preliminary evaluation evidence.