Using Long Short Term Memory Based Approaches for Carbon Steel Fatigue Remaining Useful Life Prediction

摘要

In the modern industry, the prediction of fatigue remaining useful life of materials is important for safety improvement and cost reduction. In the era of Internet of Things, large amount of data can be easily collected and analyzed using deep learning based approach for decision making. Deep learning represents a new opportunity for effective prediction of fatigue remaining useful life prediction in facing the challenge of big data. This paper presents a deep learning based approach for material fatigue remaining useful life prediction. First, the relationship between acoustic emission signal and fatigue life is established with a long short term memory (LSTM) model. Then, the convolutional neural network (CNN) models are combined with LSTM to extract features. Finally, based on the carbon steel samples, the model is tested with 1193 groups of carbon steel fatigue test data. As results shown, the prediction results are promising.