The objective of this study is to compare the accuracy of Pinnacle’s automated, data-driven model with the results of current industry practices regarding the prediction of degradation rates for assets and components.
- Pinnacle has developed an automated, data-driven model that uses machine learning (ML) techniques to predict degradation rates for assets and components. When calculating degradation rates, the ML model considers information about the asset including its operating temperature, process stream data, system type, and other asset attribute data.
- Rather than employing rule-based models, such as API 581, the ML model learns how to predict degradation rates by being exposed to a large amount of data. It uses this data to naturally learn how different variables influence the overall degradation rate, and will continue to improve over time as it is exposed to additional, higher-quality data.
- Pinnacle conducted an analysis that compared the accuracy of the degradation rates predicted by Pinnacle’s ML model to the degradation rates predicted by a human subject matter expert using the current industry standards. Pinnacle’s ML model performed significantly better than standard industry practice in overall accuracy.
- The results of the analysis exemplify the exciting possibilities for how “Big Data” can be used to solve real-world reliability challenges faster and more accurately than current industry practices.