Research On Multivariable Manufacturing Process Monitoring And Diagnosis Method Based On Machine Learning

In the modernized manufacturing process,the degree of automation and industrial complexity of the manufacturing process are constantly increasing and it cannot meet the requirements of the modernized manufacturing process which has been relying on traditional multivariate statistical process control to monitor whether the manufacturing process is abnormal.With the development of information acquisition technology,a large number of data has been generated in the manufacturing process.How to make full use of these data to control quality in the multivariable process has become an urgent problem for researchers.In the background of the big data era,many machine learning algorithms have been proposed and successfully applied in various fields.Therefore,how to apply machine learning algorithm to quality control in the manufacturing process to realize intelligent monitoring and diagnosis has become a popular topic in this research field.Combining the feature engineering and machine learning algorithms,this paper systematically studies the intelligent monitoring and diagnosis method for quality control of the multivariable manufacturing process.The main work of this paper is as follows:(1)Research on on-line monitoring model of multivariable manufacturing process based on support vector data description.The first research content is monitoring the quality of manufacturing process.Two different multivariable control charts based on support vector data description,namely D control chart and D-MCUSUM control chart,are proposed.The two proposed control charts are suitable for monitoring the small shift and large shift of the manufacturing process respectively.Due to the unsupervised learning characteristic of support vector data description,the two proposed control charts have strong self-learning and applicability simultaneously.Simulations and application examples also prove the effectiveness of the two proposed control charts in monitoring whether the manufacturing process is abnormal.(2)Research on multivariable manufacturing process monitoring model based on hybrid control chart model.Based on the characteristics that the proposed D control chart and D-MCUSUM control chart are only sensitive to specific shift ranges,this paper combines the monitoring characteristics of the two control charts through a dynamic control limit,and proposes a CDD control chart which is sensitive to a broader shift range.In addition,the general algorithm for generating the dynamic control limit and the design method of the dynamic control limit is also studied,and the selection of window size for the CDD control chart is discussed as well.The comparative study of simulation and industrial examples show that the CDD control chart is superior to the two proposed control charts in monitoring a broader shift range.Then a quality monitoring model of manufacturing process based on unsupervised learning algorithm is established.(3)Research on diagnosis of abnormal signals in the multivariate manufacturing process with random forests model.In the field of diagnosis of out-of-control signals in multivariable manufacturing process,a random forests recognition model with features is proposed,and the proposed intelligent model is successfully applied in a typical two-variable manufacturing process.This model can effectively identify abnormal sources in manufacturing process because of the advantages of feature engineering and ensemble learning of random forests.The model can quickly and accurately analyze the abnormal alarm of multivariable control charts and then adjust them.Based on the above research,this paper establishes a complete intelligent quality control system for the multivariable manufacturing process,which can realize real-time monitoring and diagnosis of the process quality.This paper also provides some new methods and ideas for quality control of the multivariable manufacturing process,and has important theoretical and practical significance.