Application of Long Short-Term Memory Neural Network to Crack Propagation Prognostics

Condition-based maintenance (CBM) is a predictive maintenance strategy that monitors the machinery states and provides optimum sets of maintenance decisions. Diagnostics and prognostics are considered to be the main aspects of CBM which are used for assessment of the monitored states. Diagnostics focuses on the detection, isolation and identification of faults while prognostics determines whether the faults or failures are forthcoming or how soon they will occur. The importance of precise prediction on the potential problems of an asset have made prognostics the topic of much recent scholarly research. Crack propagation in mechanical systems is considered as one of the main sources of mechanical failure that can bring about catastrophic consequences. Hence, obtaining a precise model for the crack propagation is crucial from the maintenance point of view. The current paper takes advantage of long short-term memory (LSTM) neural networks' ability in forecasting the evaluation of the sequential date in predicting crack growth. The presented approach is applied to the Virkler crack growth dataset. The effectiveness of the proposed method is demonstrated by post-processing the outputs of the LSTM neural network.