Six steps to predict when things will go wrong (before they do!) with IBM Predictive Maintenance and Quality
MARTIN KEEN 1200007VU3 MKEEN@US.IBM.COM | | Tags:  pmq predictive big-data maintenance martin_keen
2 Comments | 6,992 Visits
Predictive maintenance involves predicting when a component will fail or require service, before that failure happens. It goes well beyond traditional approaches, such as reactive maintenance (replace components after they fail) or scheduled maintenance (replace components on a pre-determined schedule). Instead a predictive maintenance solution uses historical data to predict when a failure is likely to occur, so you can take action and avoid costly downtime.
1. Loading master data
Before you can start making predictions you'll need to load some master data into the analytic data store (the DB2 database central to the Predictive Maintenance and Quality solution). Master data includes a list of all the devices to be monitored by the solution.
The second thing the analytic data store needs is to receive events from monitored devices. In the power transformer scenario, the device being monitored (the transformer) might provide events with several different observations such as temperature and current load measurements.
Events are aggregated into key performance indicators (KPIs) and profiles using measurement types and profile variables. A measurement type defines how to interpret a particular device reading (so a reading of “107” is understood to be a temperature reading and not something else). Profile variables designate a specific profile calculation that should be performed on the incoming data (for example to calculate the average temperature of the transformer and its current load).
Scoring is where the magic starts to happen. Predictive models are created in IBM SPSS Modeler. These predictive models use historical data to determine the probability of certain future outcomes. For example, a model could be created based on historical data regarding transformer temperature, current load, and occurrences of failure. The score that is returned can be thought of as an estimate of the likelihood that the transformer will fail within a designated period of time, based on the most recent readings.
With scores calculated, it's time to start making decisions. With SPSS Decision Management, rules can be authored, tested, optimized, and deployed. For example the recommended action that results from the predictive score may be to perform a detailed on-site inspection to look for early signs of trouble. When the predictive score shows a particularly high probability of failure, the action may be to transfer the load to another device and shut down the transformer for a component-level inspection and possible repair.
The communication of recommended actions (such as to perform an on-site inspection) can be accomplished by the creation of work orders in IBM Maximo. The accumulated KPIs and current profile values (such as the average temperature of the transformer) can be viewed in IBM Cognos Business Intelligence reports.
Martin Keen is an IBM Redbooks Project Leader. He works with technical experts to create books, guides, blogs, and videos. Follow Martin on Twitter at @MartinRTP.