Research And Implementation Of Industrial Time Series Analysis Method Based On Machine Learning

The era of industry 4.0 brings industrial production into a new era,which is based on the Internet of things,cloud computing,big data,artificial intelligence and other information technology,and promotes the development of industrial production to intelligence,automation and digitalization,which has a profound impact on social development.Traditional periodic maintenance and periodic maintenance are changing to intelligent maintenance.By collecting massive time series data generated by industrial equipment in operation,combining with intelligent algorithm analysis,fault diagnosis and Remaining Useful Life(RUL)prediction,maintenance costs can be effectively reduced.Improve equipment reliability and safety.Based on the operational time series data of aviation turbofan engines published by NASA,this paper studies and applies machine learning algorithms to predict the RUL of aircraft engines and provide decision support for engine health monitoring and intelligent maintenance.The main work is as follows:(1)For the engine multi-dimensional sensor data,feature selection is carried out based on the analysis of its working principle to reduce the influence of irrelevant variables on the prediction accuracy,and the model convergence speed is improved by normalization and sliding window reconstruction of the data.The multi-channel TCN network extracts different sensor degradation information to strengthen sensor degradation features,concatenates the extracted features,and combines the attention mechanism to calculate the weights of different sensors(such as fans,compressors,turbines,etc.)to analyze the importance of different sensors on the prediction results.Predict the remaining life of the engine.(2)For longer flight cycle and sensor dimension more engine degradation data,more than the expansion of the characteristics of attention convolution model is not enough to fully extract engine degradation information,the time and space complexity,therefore puts forward the improved Transformer model Dilated-Conv Transformer.The Multi-head Prob Sparse Self Attention is used to enhance the correlation between sequences,and the dilated causal convolution is combined to enhance the learning ability of the model.Comparative experiments on the N-CMAPSS dataset show that the improved Transformer model has a significant improvement in the prediction accuracy and operation efficiency of RUL.Finally,combined with the parameter variation trend and degradation mechanism,the potential fault units are analyzed to provide decision support for aero-engine predictive maintenance.