Leading indicators are the sensors, measuring points, etc. whose time-series readings are most correlated with an asset’s failure. These leading indicators are usually instrumented on the target asset, but sometimes they belong to other assets in a complex system and influence the target equipment downstream. Identifying specific indicators can help build more effective condition-based maintenance (CBM) rules and predictive models, since a change in the leading indicators can predict an asset’s impending failure.
Customers planning to launch CBM programs usually have to rely on the asset’s manufacturers to provide the leading indicators (e.g., temperature, pressure, flow rate) and the high/low thresholds (e.g., temperature greater than 80 degrees) and build alert rules based on those conditions. Alternatively, they can build these conditions from their own experience. There is no reliable way these methods can use data science or machine learning to create and execute these condition rules. There is also no way to account for indicators pertaining to an external factor (e.g., a certification level, an output quality level, or sensors from an upstream asset).
Determining a leading indicator for a failure event entails understanding which features are important to classify the characteristics of failure. This requires an interpretation of the model used to create the failure predictions (classification). The solution lies in using automated machine learning to determine, for an asset model and by its failure modes, the most significant leading indicators and the relevant conditions that are associated with failures. Then the solution can export these conditions to an alarm system that executes the model with minimal user interaction. These indicators can be sensors on the target equipment, from calculated metrics, or from sensors from upstream equipment in a complex system.
Explore how the SAP Predictive Maintenance and Services solution, part of SAP Intelligent Asset Management, offers functionality to automatically initiate machine learning, display the most relevant indicators for the assets and models, and enable this data to be used to set up new rules based on real data.