Fault Diagnosis And Prediction Of Lubricant Based On Principal Component Analysis And Long Short-term Memory Network

With the development of the times,the demand for efficient and reliable machine performance in China's industry is higher than ever before.The traditional periodic maintenance also encounters tremendous challenges in meeting long-term operation,preventing failure and reducing maintenance costs.In the decades of rapid development of sensor technology,preventive maintenance(PM)and condition-based maintenance(CBM)have been proposed and tried as alternatives to traditional methods.As an important part of condition monitoring technology,oil monitoring technology has significant advantages in revealing the wear,lubrication and friction conditions of friction pairs.However,the current lubricating oil data analysis technology is still in the stage of diagnostic analysis based on single index and simple mathematical statistics,and more data-driven diagnostic and prediction algorithms are still being explored.At the same time,the related oil data analysis software still has the problem of single function,which needs to be improved urgently.In this paper,the concept of control limit based on principal component analysis is introduced in the field of oil data analysis,including Q statistics,T~2 statistics and combined statistics.Through the test on oil monitoring data of steam turbine in power plant,the fault alarm rates of the three methods are compared,and the accuracy and efficiency of combined statistics in oil data fault diagnosis are verified.The feedforward neural network(FNN)is introduced,and the recurrent neural network(RNN)is deduced step by step,the long short-term memory network(LSTM)is introduced naturally.The reason why LSTM can remember data characteristics is explained in detail.The unit structure and multi-parameter optimization algorithm are described in detail.Because the training of deep neural network takes a long time when the data dimension is large,Random Forest Regression(RFR)algorithm is used to sort and filter the oil index.Then a LSTM framework for combined statistics prediction is designed according to the screening index,and a LSTM model with high prediction accuracy is obtained by piecewise mesh optimization algorithm.Compared with traditional support vector regression(SVR),BP neural network and autoregressive integrated moving average(ARIMA)algorithm on the same set of data,the accuracy of LSTM algorithm is verified.Finally,the PyQt toolkit in Python is used to design a comprehensive analysis platform for oil physical and chemical performance.The software includes login registration module,online monitoring module and off-line data analysis module.Focusing on the off-line data analysis module,it integrates the common methods of data preprocessing,data visualization,data analysis and prediction in the process of data analysis.It has rich functions and uses modular programming to expand flexibly.