Predictive maintenance failures are often a query architecture problem, not a data collection problem. Three SQL patterns power effective predictive maintenance: threshold triggers (sustained limit crossings to suppress false positives), trend detection (rate-of-change comparisons across time windows), and cross-asset correlation (finding sister assets showing matching failure patterns). CrateDB is presented as the data layer that enables these queries to run against live sensor data without export steps or pre-aggregation, supporting sub-second latency at 1 million ingested values per second. The post includes concrete SQL examples for each pattern and references ABB's Ability Genix platform as a production-scale deployment.
Table of contents
The three signals that precede equipment failureWhy the database layer determines how early you can actThreshold trigger: querying for sustained limit crossingsTrend detection: rate of change, not just absolute valueCross-asset correlation: prioritizing what to inspect nextCrateDB as the data layer, not the ML modelAt production scale: ABB Ability GenixBuilding the predictive maintenance data layer194 Impressions