What CALEB demonstrated

Built an end-to-end rolling-bearing fault detector from vibration data: engineered RMS, kurtosis, crest factor, and envelope-spectrum defect-band energy; trained a compact depth-1 decision tree on training bearings only; and evaluated it on held-out bearings with reported catch-rate and false-alarm rate. The final memo justified a 0.42 alarm threshold using the student’s own cost numbers and correctly localized the fault to the outer race via the BPFO peak.

• Leakage-Free Feature Engineering
• Honest Held-Out Evaluation
• Physics-Based Fault Localization
• Cost-Based Operational Decision-Making