A Data-Driven Diagnosis and Prognosis method for Machinery Tools Based on EMD and Dual-Task Deep Neural Networks

Health state recognition and remaining useful life prediction are two major challenging tasks in the area of diagnosis and prognosis of rotating machinery tools. The performance of data-driven-based methods mainly depends on the quantity and quality of input features. Therefore, to improve the performance, this article proposes a new framework based on Empirical Mode Decomposition (EMD) and dual-task deep convolutional neural networks. The EMD method has been used to extract energy-based discriminative input features. A dual-task 1-D deep convolutional neural network has been designed to predict the remaining useful life and health state of machinery tools. Moreover, the proposed approach has the ability to deal with non-stationary signals. A case study through the PRONOSTIA degradation dataset provided by FEMTO-ST Institute has been performed to validate the proposed methodology. The obtained best root mean square error and mean absolute error are 0.02 and 0.012, respectively, for the measure of remaining useful life prediction performance. The health state recognition accuracy is as high as 96.5%. The experimental results and comparative study shows the effectiveness of the proposed framework.