Siamese Inception Time Network for Remaining Useful Life Estimation

Authors : Uğur CEYLAN, Yakup GENÇ

Pages : 165-175

View : 16 | Download : 7

Publication Date : 2021-12-30

Article Type : Research Paper

Abstract :Predictive maintenance tries to reduce cost in engine maintenance in power plants, aircraft, and factories by predicting when maintenance is needed insert ignore into journalissuearticles values(or by estimating the remaining useful life);. Recently, with significant advances in deep learning and the availability of high volumes of data extracted from manufacturing processes, data-driven methods for predicting remaining useful life insert ignore into journalissuearticles values(RUL); have received a lot of attention. A major problem with data-based prognosis is that it is costly and requires large amounts of training data that could be difficult to obtain in large numbers of failure cases insert ignore into journalissuearticles values(often impossible and expensive to obtain in the real world);. This paper proposes a learning-based method for fault diagnosis requiring fewer failure data. To this end, we use a Siamese network architecture based on a specific deep Convolutional Neural Network insert ignore into journalissuearticles values(CNN); called InceptionTime. The Siamese part of the network allows repeated use of the existing data to establish a similarity metric for two separate time windows. The turbofan engines C-MAPSS dataset supplied by NASA is used to verify the proposed model. Experiments are conducted using various sizes of the turbofan engines data to compare the performance on small datasets between the proposed model and a base-line model. The results demonstrate that our model can be used in fault diagnosis and provide satisfying prediction results with fewer data with comparable performances against state-of-art methods for RUL prediction.
Keywords : C MAPSS, deep learning, prognostics and health management, remaining useful life, siamese networks