Faults Prediction and Diagnoses of Shield Machine Based on LSTM
Zhonghai Sun, Zheng Hengyu, Shi Buhai · 2019
Because the environment where shield machine is running is full of complexity and risk, it's possible for some serious faults to occur. And if operators could not find and solve the problems timely, it may lead to accidents. This paper applies long short-term network to predict and diagnose some common faults including the formation of mud cake, the wear of cutters, the blockage of slurry pipe and the subsidence of land by time series sensor data from shield machine. The simulation verifies the feasibility of our method. In this paper, a LSTM network is improved and we demonstrate that compared with LSTM model and GRU(Gated Recurrent Unit) model, the loss function of the improved model converge faster in the same epoch and achieve higher accuracy. The results show that our method to predict and diagnose the faults is satisfying.