Programs Fail at Scale
Seventy percent of predictive maintenance initiatives fail within two years. Data from McKinsey shows 60-80% of PdM projects do not meet ROI targets. Sensors flag anomalies, but technicians chase ghosts. False positives consume 40% of wrench time. Real failures slip through amid the flood. Leadership pushes PdM for 20-30% uptime gains seen in pilots; floor reality delivers 2-5% at best.
Sensor Data Overload Hits First
A single CNC mill with 20 sensors produces 1GB of data per shift. Scale to 50 machines: 50GB daily. MES systems choke. ERP integrations lag by hours. Engineers face 500+ alerts per day. Triage takes 2-4 hours. Response rates drop below 70%. Machines run overdue while teams filter noise. Downtime from ignored true failures averages 8 hours per incident.
Separating Signal from Noise
Filter first. Baseline healthy runs with 72 hours of data per asset under steady load. Compute means, standard deviations, and baselines. Thresholds set at three standard deviations catch 95% of deviations. Apply bandpass filters for vibration, reject DC offsets from mounting. Machine learning helps sparingly: random forests classify anomalies with 85% accuracy on clean data. Domain knowledge trumps algorithms. This approach cuts alerts 70%.
Actual ROI on the Floor
Pilot ROIs hit 300% in controlled tests. Floor reality averages 15-25% over 18 months. Downtime savings avert 10-15 hours per machine yearly. Parts savings reduce premature replacements by 20%. But factor maintenance time: 500 hours triaging at $50/hour adds $25k in costs. OEE rises 3-5 points, not the 15 promised. PdM shines on critical assets like spindles and gearboxes; skip low-value pumps.
Maintenance Still Rules
PdM augments, does not replace, traditional maintenance. Run-to-failure remains appropriate for 70% of assets under $1k repair cost. PdM targets the top 20% by downtime cost. Time-based PM intervals shorten 20% based on PdM trends. Technician upskilling matters: 80% of alerts need human judgment. The hybrid approach of PdM flags, PM prevents, and run-to-failure fills gaps yields 15% MTTR reduction.