Predictive Maintenance in Python: Rolling Features, Anomaly Alerts and Failures You Can't See

Tapti Valley Spinning Mills runs 28 motors around the clock outside Surat, and every breakdown is a surprise at the worst hour. Its maintenance head wants one thing: tell him the day before which motor is going to fail, so his crew can swap it in a planned slot.You get your own copy of the mill's data: six weeks of two-hour sensor readings with the breakdown log, and the next 16 days of readings without it. You build the early warning yourself, lesson by lesson. Then you give the same job to your AI assistant (ChatGPT, Claude, Gemini or Copilot) and check what it gives back. Bartman runs your script in a sealed sandbox and scores your risk scores and alerts against the breakdowns that really happened. Neither you, your AI nor the sandbox ever sees them.<ul><li>Clean real sensor data: rows out of order, repeated uploads, a Fahrenheit gateway, and blanks that are not zeros.</li><li>Build the label from a breakdown log: will this motor fail in the next 24 hours? Compare times, not rows, and leave out what you can't know yet.</li><li>Compute rolling features per motor, in time order, from the past only, against each motor's own baseline, and catch the centred window that reads the future.</li><li>Write a rule-based anomaly alert and measure it the way a maintenance crew would: precision, recall and failures caught.</li><li>Fit logistic regression and gradient boosting with folds that hold out whole motors, choose an alert threshold on out-of-fold predictions, and draw a trend chart for the maintenance office.</li></ul>Every learner gets different data, so most answers (and every table and model score) differ from anyone else's. Each check explains what went wrong without giving the answer away, and the classic mistakes (windows that cross motors, a centred window, blanks filled with zero, a baseline from the test weeks) are recognised and named.

Tools: Python, pandas, scikit-learn, Time Series, Rolling Windows, Anomaly Detection, Logistic Regression, Gradient Boosting, Precision and Recall, matplotlib, AI Assistants for Python