Predictive Maintenance is all about predicting failures, to critical assets, in a timely fashion to allow for repair/replacement without too much or no upset(s) to production/processes.
The business case for running a Predictive Maintenance Program (PMP) is clear to most clients/companies. For many we find it is often unclear as to how to run an effective PMP and what techniques to embrace.
This post is too short to dive into the finesses of asset, part selection and using available data and engineering resources such as FMEAs and FMECAs to create models that predict failure and degradation … but long enough to focus on one of the important aspects in PMP: use of event asset data to enhance predictive models and detect early failures.
Use Asset Event Data!
Many assets provide streams of event data often barely analyzed but stored. We have found on a number of occasions that these streams contain valuable (early) warnings for failures. The warnings are often not events by themselves but combinations of multiple events in time. This is exactly where Event Stream Processing (ESP) and Complex Event Processing (CEP) are of great value.
There are multiple techniques and algorithms available to mine sequential event patterns (so called episodes) from event history. UReason’s experience is that when you combine episode mining results with failure data you get a powerful data cocktail on which you can unleash your data science skills:
Combing event pattern detection with pure failure data driven models allows you to provide early warning of incipient failures. UReason has successfully applied described techniques for customers in the Chemical, Petrochemical and Transport industries.
Want to Learn More?
UReason has developed proven event pattern mining algorithms that allow you to mine sequential event patterns in event databases. Next to this we provide the platform within which you can combine found/validated event patterns, train & validate models and link these to live data and event streams.
If you have developed models in Python, R or other language integration facilities exist to integrate these and combine it with UReason’s platform capabilities.