Remaining Life Prediction Of Metal Band Saw Blade Based On Machine Learning

In the context of the "Made in China 2025" policy driven by innovation,the traditional manufacturing industry has begun to upgrade through industrial intelligent solutions.At present,the intelligence of metal band saws is mainly reflected in sawing automation and fault diagnosis,and the prediction of the remaining life of the band saw blade is in the exploratory stage.Replacing the band saw blade too early will result in waste of the band saw blade,while replacing the band saw blade too late will bring about a decline in sawing efficiency and quality.It is very important to accurately predict the life of the band saw blade.At present,the main methods for judging the remaining life of band saw blades are failure physics and data fusion,which have problems such as low efficiency and excessive interference in industrial practice.In order to collect the band saw blade cutting data needed for the research,this paper designs a complete set of remote monitoring system for metal band sawing machine.Through the intelligent gateway and the band sawing machine PLC docking,the band sawing machine cutting information is collected and uploaded to the cloud platform for analysis,and the APP and Web client are designed to realize the multi-end monitoring and control of the system.The system completes the online monitoring and data collection functions of the band saw,and provides data support for the prediction of the remaining life of the band saw blade.The remaining life of the metal band saw blade needs to be combined with the analysis of the total cutting area before failure,so this paper proposes a set of remaining life calculation schemes.By calculating the area of each blade data collected and optimizing the calculation plan,the accurate cutting area is obtained,and the reasonable failure point is found by analyzing and recording the whole life data of scrap metal band saw blades.In this way,the total cutting area before failure is obtained,and the normalized remaining life of the metal band saw blade after each cut is calculated as the output of supervised learning in machine learning.In order to obtain the input features related to the remaining life,this paper proposes a data-driven remaining life prediction feature selection scheme.Through the sorting and preprocessing of other data collected by the remote monitoring system,the data set is divided according to 4:1,and the selection of input features is completed by combining experimental tests such as the correlation with the remaining life.Finally,through the analysis of the three machine learning algorithm principles of K nearest neighbor algorithm,random forest algorithm and LightGBM algorithm,the training set is trained based on the grid search algorithm and the five-fold cross-validation,and the performance of the test set is verified.The verification results show that the LightGBM algorithm has higher accuracy than the K nearest neighbor algorithm,and is nearly 10 times faster than the random forest algorithm,which is suitable for industrial practice.This paper proposes a band saw blade prediction model based on the LightGBM algorithm.On the basis of the LightGBM algorithm,the Bayesian optimization algorithm is further used to improve its parameter tuning process and improve its prediction accuracy.The root mean square error of the final prediction result is 0.0276,and the coefficient of determination is 0.9908,to meet the demand for prediction of the remaining life of the band saw blade.