CNN-Informer-Based Remaining Useful Life Prediction for Electrical Devices

Accurate Remaining Useful Life (RUL) prediction plays an important role in the health management and predictive maintenance of electrical systems. Advanced AI technologies, such as Convolution Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), have been heavily involved into RUL prediction methods. However, the existing RUL prediction models, still do not fully consider the sequence information, or suffering the problem of long-term dependence. A RUL prediction model combining the advantages of CNN and Informer is proposed in this paper. In this model, CNN is used to reduce the dimension and denoise the original sensor data and transform it into a time series that is easy to be accepted by Informer. Then, Informer extracts the life-related sequence information contained in the time series based on the attention mechanism, and relies on the sparsity matrix to simplify the calculation of attention. Finally, the full connection layer maps the output of Informer into a lifetime vector. Comprehensive experiments have been conducted using two popular public datasets, and the comparison results show that the proposed method over-performs the existing data-driven-based methods.