SMI Predictive Maintenance — Feasibility Report & Architecture

Coursework for the Artificial Intelligence (CSI_6_ARI) module: a feasibility study for a predictive-maintenance decision-support system at Seiryu Mobility Industries (SMI), targeting rare but safety-critical failures in an auxiliary braking and suspension-stability subsystem. Synthesised six peer-reviewed papers on ensemble and hybrid failure-detection methods (GBDT, XGBoost, stacking) against SMI’s constraints — severe class imbalance, distribution shift over a vehicle’s lifetime, and an asymmetric cost function where a missed failure vastly outweighs a false alarm. Specified a 10-module end-to-end architecture spanning vehicle edge, backend platform, AI inference and operations, with a decision-support agent that turns calibrated risk scores into ranked, explainable alerts for human review.