Research on Long Life Product Prognostics Technology Based on Deep Learning and Statistical Information Fusion

Abstract

The simple use of reliability statistical models to predict the remaining life of products lacks specific information about equipment performance degradation, which may lead to low accuracy in predicting the remaining life of equipment after long-term operation. However, due to the slow or even non changing performance of long-life products in the early stages, using only Deep Learning based state assessment techniques will result in lower accuracy in predicting early remaining life. In order to accurately predict the remaining life of a product throughout its entire lifecycle, this paper proposes a residual life prediction model that integrates reliability and performance information. This method identifies equipment performance degradation indicators in multidimensional time series signals through Deep Learning models, and uses a Discrete Random Damage model to establish the relationship between equipment reliability and operating time. Finally, through Bayesian information fusion technology, the reliability and performance evaluation results are integrated into the remaining life indicator, forming an integrated evaluation method for reliability and performance. Compared with the prediction accuracy of simple ordinary Deep Learning models, this method significantly improves the accuracy of early residual life prediction.