Machine learning for predictive maintenance of distributed systems

Abstract

Evidently, accurate remaining useful life estimation provides numerous benefits in a wide range of real-world applications and industrial settings. Using multi-channel sensing signals, this project investigated a new deep convolutional neural network-based architecture for estimating the remaining useful life of systems. In the proposed deep architecture, an innovative convolutional neural network with 2-dimensional input and 1-dimensional convolution and pooling layer structure is used to better capture notable patterns of sensor inputs at multiple time scales. All detected degrading patterns are methodically integrated in the estimate model and eventually help to estimate the remaining time. The proposed algorithm's performance on the Commercial Modular Aero Propulsion System Simulation (C-MAPSS) dataset is evaluated, and the outcome predictions show that it can achieve high accuracy when compared to some state-of-the-art regression-based models.