Classification-Based Framework for Remaining Useful Life Prediction With Limited Images and Unequal Time Intervals

TL;DR: Image-driven classification-based framework for remaining useful life prediction with limited images and unequal time intervals. The framework augments image streams, extends trajectories, judges working-failure states, and infers remaining useful life based on classification results.

Abstract: Time series of image data on the surfaces or interior structures of units can be used for remaining useful life (RUL) prediction. However, only a few studies have used this type of imaging, and even then those studies have not dealt with situations in which only a limited number of images or images taken at unequal time intervals are available. This article proposes an image-driven classification-based framework (IDCBF) for predicting RUL using a limited number of images. Multiple methods of numerical interpolations are applied to approximate expected degradation paths that are unknown, contributing to the IDCBF’s flexibility. The proposed approach applies and extends the classification of images to the RUL prediction. The IDCBF framework augments the image streams, extends the trajectories of in-field units, judges the working-failure states of units using a classifier, and infers the RUL based on the classification results. Numerical experiments on a simulated image dataset and a case study on a real image dataset demonstrate the effectiveness of the proposed methodology.