Anomaly Detection And Location Of Injection Molding Machine Based On Data Driving

With the rapidly expansion of modern manufacturing industry,plastic products are popularly used in daily life,and people put forward more and more strict requirements for the stability of plastic products production.Injection molding machine is the core equipment in the production of plastic products,but also the main energy consumption equipment in the process of industrial production,its operation will directly affect the product quality and energy efficiency.Once the injection molding machine in the production process of abnormal work and can not be timely dealt with,will lead to injection parts product defects,enterprise energy costs increase,system parts wear and other problems,for the enterprise to bring significant economic losses.Therefore,the process monitoring of injection molding machine can timely find the abnormal of injection molding machine,which is of great significance for ensuring the reliable operation of equipment,safe production and reducing energy consumption.With the evolution of Internet of Things technology and storage technology,as well as the wide application of intelligent instrument and modern control technology,anomaly detection method based on data driven industrial equipment has become a current research hotspot.Based on horizontal injection molding machine as the research object,this thesis analyzes the common abnormal situation in the process of injection molding machine production,in view of the injection molding process is a batch process,have the characteristics of nonlinearity,multi-modal,partial inferior,based on the improved entropy component analysis algorithm of injection molding machine for nuclear anomaly detection methods,and put forward the combination of fault tree and Bayesian network abnormal positioning method,finally,Based on the research,the module of abnormal detection and location in injection molding process is developed by using Java language and Matlab.The thesis contents are as follows:（1）The structure of injection molding machine and the process principle of injection molding process are analyzed in detail,and the influencing factors of injection molding process are analyzed,and the common abnormal types and reasons of injection molding machine are summarized to provide prior knowledge for the following text.（2）Injection molding process is a batch process,with multi-mode,strongly nonlinear,mixed distribution and other characteristics,which leads to the unsatisfactory anomaly detection effect of traditional kernel entropy component analysis（KECA）algorithm,thus,the KECA algorithm was improved,The local neighbourhood standardized LNS was introduced to unify the multi-mode data into a single mode,and then feature extraction was carried out based on the multi-direction kernel entropy component analysis algorithm（MKECA）.Finally,support vector data description（SVDD）was used for anomaly detection,which improved the accuracy of anomaly detection.The experimental results show that the proposed algorithm is effective and accurate in the abnormal detection of injection molding machine.（3）In order to locate the abnormal source quickly,a method of abnormal location for injection molding machine based on fault tree and Bayesian network was proposed,which made full use of the advantage of Bayesian network（BN）in solving uncertain problems to solve the problem of complicated and difficult to locate the abnormal cause of injection molding machine.Firstly,the fault tree model of injection molding machine system is transformed to Bayesian network,and the initial topology of Bayesian network based on expert knowledge is established.Secondly,the K2 algorithm is used to learn the Bayesian network structure according to the abnormal data of injection molding machine,and the maximum likelihood estimation（MLE）algorithm is used to learn the parameters,and the best Bayesian network is obtained.Finally,causality diagnosis is carried out through the junction tree（JT）inference engine,and the biggest possible cause of the anomaly is deduced.The validity and feasibility of the Bayesian network are verified by the actual data set of enterprises,which has practical significance for the rapid location of the anomaly of enterprise injection molding machine.（4）On the basis of the energy management system of a large injection molding enterprise in South China,the abnormal detection and positioning microservice module is designed and developed.By compiling the Matlab.M file and packaging it into JAR package for Java program to call,the mixed programming is realized.After the corresponding micro-service is registered and published in the registry,it has been preliminarily applied in the production of enterprises,which proves the feasibility of the proposed method in practical application.