hardCase StudyScenarioSeniorStaffLead
Most predictive maintenance projects never reach production. Why, and what would you do differently?
6 min readUpdated 2026-08-15
Answer
Short answer
They fail on labels rather than models: failures are recorded as repair dates in free-text work orders, the positive class is nearly empty, feature windows leak the outcome, and no baseline was established, so an encouraging first result cannot be trusted or beaten.
Was this answer helpful?
Likely follow-ups
- You have 40,000 hours of sensor data and eleven recorded failures. What can you honestly attempt?
- How would you split this dataset, and what does a random split get wrong?
- The plant already has vibration alarms. What does your model have to beat, and by how much?
- Six months in, precision has halved with no code change. What do you check first?
Related questions
- You have fine-tuned a model for a task. How do you prove it helped?hardAlso on baselines and data-leakage5 min
- Your predictive maintenance system is live and the maintenance team has stopped acting on its alerts. How do you get that back?hardAlso on predictive-maintenance6 min
- A churn model scores 0.94 AUC offline and barely beats the baseline in production. One of its strongest features is populated by a process that runs after the customer has already churned. Explain what happened and how you would fix it.mediumAlso on data-leakage5 min
- One transaction in two thousand is fraudulent and you have been asked to build the detector. How do you approach it?hardAlso on class-imbalance5 min