Self-directed project
Predictive Maintenance: Industrial ML for Fault Detection
Condition-Monitoring Engineer · AI / ML · August 2026
A ProoV case study · educational project, not employment
Built a compact bearing-fault detector end to end: plotted a faulty vibration snapshot, used time-domain features including kurtosis, trained a shallow decision tree on training bearings only, and evaluated it on a held-out set with catch-rate and false-alarm rate. Also produced a maintenance memo defending a 0.50 alarm threshold with explicit cost trade-offs and a plain-language explanation for operations staff.
Graded against
- Feature engineering (signal to honest numbers)25%
- Held-out evaluation (honest measurement)25%
- Cost-based threshold decision + memo25%
- Compact model + embedded-budget justification15%
- Plain-language explanation10%
Passed · pass mark 60/100
What stood out6
- Used a depth-capped decision tree and explicitly stated depth 1 and 3 nodes against the embedded budget
- Reported held-out catch rate and false-alarm rate rather than evaluating on training data
- Wrote a maintenance memo that ties the threshold choice to concrete catch-rate and false-alarm numbers
- Leakage-Free Model Training
- Honest Held-Out Evaluation
- Cost-Based Operational Decision-Making
The work I submitted18 tasks
10 tasks · 6.4k characters · Python · 8 written answers · 428 words
- Initial Prediction62 characters
- Load Plot Signal CodePython · 1.9k characters
- Load Plot Signal205 characters
- Example Rms WorkedPython · 698 characters
- Time Features143 characters
- Freq FeaturesPython · 1.1k characters
- Train Classifier363 characters
- Evaluation Heldout288 characters
- Threshold Decision334 characters
- Decision Memo1.4k characters
- Initial Prediction9 words
- Load Plot Signal Code163 words
- Load Plot Signal15 words
- Example Rms Worked72 words
- Time Features10 words
- Freq Features104 words
- Train Classifier31 words
- Evaluation Heldout24 words
Summarised on purpose — the submitted code and writing stay private so this page cannot be reused as an answer key. The full submission sits behind the verified certificate.