What Dinan demonstrated

Built an end-to-end rolling-bearing fault detector from vibration data: engineered RMS, kurtosis, crest factor, and envelope-spectrum defect-band energy; localized the fault to the outer race via BPFO; trained a depth-capped decision tree on training bearings only; and evaluated it on held-out bearings with reported catch-rate and false-alarm rate. Also justified an alarm threshold using the candidate’s own cost numbers in a maintenance memo.

• Leakage-Free Feature Engineering
• Honest Model Evaluation
• Physics-Based Fault Localization
• Cost-Based Decision Making