| In industrial manufacturing,the occurrence of equipment failures and the corresponding maintenance work has always been a problem that needs to be solved in industry.With the development of intelligent manufacturing,industrial equipment is becoming more sophisticated and complex,and traditional equipment maintenance methods are no longer able to adapt to the development of modern industry.With the rapid development of technologies such as big data,artificial intelligence and the Internet of Things in recent years,the fault prediction methods have become an important solution in the field of industrial equipment maintenance.The high-dimensional feature redundancy and category imbalance of industrial equipment data can lead to a low accuracy of algorithms in predicting faults.In this paper,we investigate the feature selection problem and data imbalance learning problem in fault prediction using integrated learning algorithms to improve the accuracy of fault prediction and reduce the training time complexity in response to the characteristics of industrial equipment data.The main research content and results include:(1)To address the problem of high feature dimension and feature redundancy in industrial equipment data,an RF-RFE feature selection method combining random forest algorithm(RF)and recursive feature elimination(RFE)is proposed to remove irrelevant and redundant features in industrial equipment data.The RF-RFE feature selection method uses the RF algorithm to train the model,calculates the importance measure of the data features,and then selects the data features through the backward search and deletion strategy of the RFE,and obtains the optimal feature subset as the input features of the algorithm training,which enhance the generalization of the model and improve the accuracy of fault prediction.The experimental results verify the effectiveness of the proposed method.(2)To address the class imbalance problem of industrial equipment data,a fault prediction algorithm named as ASNM-LightGBM(Adaptive Synthetic and NearMiss sampling-LightGBM)based on data sampling and ensemble learning is proposed.Real industrial equipment data is extremely unbalanced,that is,the number of fault samples is far less than normal samples,resulting in low accuracy of the LightGBM ensemble learning algorithm in predicting faults.In order to improve the accuracy of fault prediction,the LightGBM algorithm is optimized by using the ASNM combined sampling algorithm.The ASNM algorithm uses the ADASYN adaptive synthesis technology to increase the number of fault samples,and then uses the NearMiss algorithm to undersample the normal samples,and achieve data category balance through combined sampling.Finally,it is combined with the LightGBM ensemble learning algorithm to obtain a high fault prediction accuracy.The experimental results show that the algorithm proposed in this paper has high recall rate and AUC value.(3)The industrial fault prediction system has been designed.The functional modules of the system have been implemented using the Java development techniques,and we proposed the algorithm have been applied to the fault prediction module of system,which further validate the effectiveness and practicality of the algorithm. |
