When Predictive Maintenance Stops Predicting: How a Wind Farm Operator Used Consulting and Analytics to Fix a Failure Detection Programme That Had Quietly Stopped Working
Executive Snapshot
Client
Situation/Challenge
Objective
Constancy Researchers Solution
Impact
Client Outcome
The Situation / Challenge
A predictive maintenance programme is only as good as its ability to separate genuine early failure signals from routine operating noise, and that distinction drifts over time as a fleet ages, expands, or simply behaves differently than the assumptions baked into the original thresholds. A system that worked well at installation can quietly stop working as conditions change, without anyone noticing until a failure that should have been caught slips through.
The client had rolled out a vibration-based predictive maintenance system across its fleet two years earlier, expecting a meaningful drop in unplanned downtime. The drop never really materialised. The system generated a steady stream of alerts the operations team dutifully investigated, but several gearbox and bearing failures still occurred without adequate advance warning. Nobody had gone back to check whether the original alert thresholds still matched the fleet as it existed two years later.
Without understanding why genuine failures were slipping past a system built specifically to catch them, the client risked continuing to pay for a programme that looked sophisticated on paper but was not delivering the downtime reduction it was built to achieve.
Key Challenges
- No audit comparing the system’s alert history against actual confirmed failures to check its real detection accuracy.
- No review of whether thresholds set at installation still matched the fleet two years later, after newer models had been added.
- A steady stream of alerts consuming operations attention without a clear sense of which ones genuinely mattered.
- Several gearbox and bearing failures occurring without the advance warning the system was specifically built to provide.
- No analytics connecting sensor readings before a confirmed failure to the thresholds that should have flagged them.
- Leadership pressure to justify the programme’s continued cost given its underwhelming impact on downtime.
Predictive maintenance systems are rarely set-and-forget. Alert thresholds calibrated for one turbine model or one point in a fleet’s life can drift out of relevance as the fleet changes, and the only way to know whether a system is still doing its job is to check its alerts against what actually failed.
Constancy Researchers Solution
Constancy Researchers audited the predictive maintenance system’s actual detection performance against confirmed failure history, then used analytics to trace exactly why genuine failures had been slipping past the alert thresholds.
Alert Accuracy & Confirmed Failure Audit
- Conducted an audit comparing two years of alert history against confirmed gearbox and bearing failure records, checking how many failures had been flagged in advance and how far ahead.
- Found that a meaningful share of confirmed failures had received no advance alert at all, while the system generated frequent alerts for issues that were not genuine failures.
Alert Threshold Calibration History Review
- Reviewed how the vibration alert thresholds had originally been set, finding they were calibrated against a small sample of the fleet’s earliest turbine models and never revisited.
- Confirmed that newer turbine models added after installation had meaningfully different baseline vibration signatures, leaving the original thresholds poorly matched to much of the fleet.
Sensor Data Trend Analysis Ahead of Confirmed Failures
- Analysed raw sensor readings in the weeks before each confirmed failure, checking whether early warning signs existed in the data even when no alert had fired.
- Found that readings for several confirmed failures showed a clear rising trend for weeks beforehand, but stayed below the fixed threshold throughout, meaning the data was there but the trigger logic was not.
Recalibration & Trend-Based Alert Design
- Recommended recalibrating alert thresholds separately by turbine model rather than applying one fleet-wide threshold, correcting the mismatch the calibration review had identified.
- Designed a secondary trend-based alert layer flagging a sustained upward trajectory in readings even when the absolute value stayed below the fixed threshold.
Implementation & Performance Validation Plan
- Delivered an implementation plan sequencing threshold recalibration by turbine model alongside rollout of the new trend-based alert layer.
- Built a validation protocol comparing detection performance under the new system against the prior season’s confirmed failures before full reliance on it began.
The work gave the operator a precise explanation for why a system built to predict failures had stopped doing so reliably, along with a specific fix rather than a recommendation to scrap the programme entirely.
Impact
- The audit confirmed the predictive maintenance system was missing a meaningful share of genuine failures despite generating frequent alerts.
- The calibration review traced the gap to thresholds set against early turbine models and never updated as the fleet expanded.
- Sensor trend analysis showed clear early warning signs existed in the data for several missed failures, just below the fixed alert threshold.
- Per-model threshold recalibration corrected the mismatch between alert settings and the current fleet composition.
- The new trend-based alert layer directly addressed failures that rising-but-below-threshold readings had previously let through.
- The validation protocol gave the operations team a way to confirm detection performance improved before fully trusting the new system.
- Two gearbox issues were caught during the following season that the original threshold-only system would have missed.
- The operator avoided abandoning a costly programme that needed recalibration rather than replacement.
Client Outcome
Detection Improvement
Two gearbox issues were caught during the following season that the original system would have missed entirely.
Root Cause Clarity
Outdated, fleet-wide alert thresholds were identified as the actual cause of missed failures, not a fundamental flaw in the sensor technology.
Targeted Recalibration
Alert thresholds were recalibrated by turbine model, correcting a mismatch that had existed since the fleet expanded beyond its original models.
New Detection Layer
A trend-based alert layer now catches sustained risk patterns that fixed thresholds alone had been missing.
Programme Justified
Leadership retained confidence in the predictive maintenance investment once the actual cause of underperformance was identified and fixed.
Validated Confidence
A structured validation protocol confirmed improved detection performance before the operations team fully relied on the recalibrated system.
Operational Efficiency
Operations staff attention is now directed more accurately, with fewer false alerts and more reliable advance warning.
Downtime Risk Reduction
The earlier catch on two gearbox issues avoided the unplanned downtime and repair cost a missed failure would have caused.
Market Positioning
The operator was repositioned as a data-disciplined operator who audits and recalibrates its maintenance technology rather than assuming it works indefinitely.
Case Studies
Learn how our success stories power data-driven growth across industries
Picking the Right Anchor: How a Floating Wind Developer Used Bespoke Research and Strategy Advisory to Settle a Foundation Technology Standoff Before Bidding on a Lease
Executive Snapshot Client Floating Offshore Wind Developer, Scotland & South...
Read MoreWhen Predictive Maintenance Stops Predicting: How a Wind Farm Operator Used Consulting and Analytics to Fix a Failure Detection Programme That Had Quietly Stopped Working
Executive Snapshot Client Onshore Wind Farm Operator, Texas & Iowa...
Read MoreGeared, Direct Drive, or Both: How a Gearbox Manufacturer Used Market Research and Consulting to Decide Where Its Future Actually Lay
Executive Snapshot Client Wind Turbine Gearbox Manufacturer, Germany Situation/Challenge Direct...
Read More