A Proposed Method for Generating Lifetime Failure Data for Manufacturing Equipment: Validation With Bearings

Digitization of manufacturing allows modeling and diagnosis of equipment failure, but training such systems is difficult without observations of failed states. This paper presents a method whereby failure data, from the baseline to the destruction case, is methodically generated to train machine health models. Creating failure classifiers and predictive models of manufacturing equipment suffer from two problems: training-data accuracy and data representation of lifetime machine failure modes of both components and systems. The Purposeful Failure Method aims to generate defect training data at both the component and system level, from initial defect to final destruction. Bearings are a validation case and are monitored from a healthy baseline state to a maximum-damage state by applying artificial damage. Bearings are used for validating the method, due to prior knowledge of failure progression, as well as their prevalent use. The generated data were analyzed to see if they matched expected failure phenomena. Preliminary results with the Purposeful Failure Method show promise for generating failure data for machine health models; continued work on validating the method on other components is justified.