ORIGINAL COMPANY-CONTEXT PRACTICE
Distinguish sensor drift from a process change
Original practice scenario: An energy-operations simulator reports rising pressure readings after a sensor replacement. How would you investigate before training an anomaly detector?
Written for independent practice using company context; this is not a verified question from a company interview.
A REASONED APPROACH
Model answer.
Separate instrumentation changes from changes in the measured process. A replacement can alter calibration, reporting frequency or units, so an apparent anomaly may describe the measurement system.
Work through it
- Build a timeline of installation, calibration, maintenance and operating conditions; preserve raw readings with device identity, units and event timestamps.
- Compare overlapping reference sensors and comparable operating regimes, examining missingness and sampling intervals before aggregating trends.
- Evaluate detectors on held-out time periods with known simulated incidents and maintenance transitions; report alert delay and false alerts separately.
Tradeoffs to discuss
- Normalizing every device independently can conceal meaningful absolute limits; pooled models may overreact to harmless calibration differences.
Common pitfalls
- Do not label all post-replacement changes as equipment faults.
- Keep maintenance information available at prediction time rather than leaking later inspection outcomes.