A Tensor-based t-SVD-LSTM Remaining Useful Life Prediction Model for Industrial Intelligence

Abstract

The Industrial Internet of Things (IIoT) integrates smart sensors and actuators for the widespread digitization and enhancement of industrial and manufacturing processes. Smart equipment is used to improve the industrial intelligence and make industrial production more flexible, safer and more efficient. For complex equipment, product life-cycle management (PLM) including remaining useful life (RUL) is one of the essential issues for industrial intelligence. In this paper, a tensor-based remaining useful life prediction model is proposed to facilitate the life-cycle management, which combines features from time domain and frequency domain. For the characteristics of continuous generation of industrial data streaming, tensor singular value decomposition (t-SVD) is combined with long shortterm memory network (LSTM) method to predict the RUL of devices from high-order and high-noise time series data. Finally, experiments are carried out on three different data sets including the battery charge and discharge data set, the bearing acceleration life cycle data set, and the turbofan data set to measure the performance of the proposed model.