Research And Application Of Extruder Anomaly Detection Based On Time Series Analysis

China has always been a major producer and consumer of aluminum profiles,of which extruded materials play an important role in China's aluminum industry.As an important equipment on the aluminum profile production line,the extruder's structure tends to be automated,complicated and large in the development process.In the production process,there are many abnormalities that are difficult to observe directly in the extruder equipment.If the abnormalities are not handled in time,it is easy to cause other related abnormalities,which will have an unpredictable impact on product quality and normal production of the enterprise.Great economic loss.In the research,it is found that due to abnormal data transmission of the collection program and failure of the sensing equipment,the problem of missing data in the production data collected by the energy management system is particularly serious,which seriously affects the effectiveness and integrity of the anomaly detection algorithm.Therefore,in order to effectively apply the anomaly detection algorithm under study,it is necessary to first solve the problem of missing data in the collected data.This thesis takes the aluminum profile extruder as the research object,uses the various data collected during the production process of the extrusion equipment as the basic data unit of the study,and studies the problem of the lack of production data and the detection of abnormal production of the extruder.The research content is as follows:(1)In order to have a more specific understanding of the production process and abnormal state of the extruder,firstly,the structure and process of the production system of the extruder are analyzed,and the cooperative working relationship among the components in the extruder system is elaborated;on this basis,the abnormal state of the production of the extruder and the related influencing factors are further analyzed.(2)In order to solve the problem of missing data in the energy management system,a method for filling in missing data of large unequal lengths based on Naive Bayes is proposed.In view of the fact that there are many industrial big data collection equipment and high repair efficiency requirements,a method of establishing a probabilistic model based on the Naive Bayes method is proposed to ensure that the model can better fit the changing law of data while completing data filling more efficiently task.In view of the fact that the missing data segments have different lengths in the production environment,the proposed method uses continuous prediction and then uses the constraint screening method to simultaneously repair the missing data segments with different lengths in one data repair process.The experimental results show that the proposed data filling algorithm can not only complete a large length of missing data filling tasks,but also has high accuracy.(3)In order to detect the equipment status abnormality that may exist during the operation of the extruder,an abnormality detection method of extruder flow data based on the isolated forest algorithm is proposed.In view of the problem that the noise in the data set cannot accurately reflect the state of the device,the algorithm extracts the sequence characteristics of the data to represent the state of the device in a certain period of time,reducing the impact of noise on the effect of anomaly detection.At the same time,the semi-space isolated forest(HS-Trees)algorithm has been further improved to solve the problem of delay in the results feedback in the original algorithm,so that it can better adapt to streaming data scenarios.Experimental results show that the proposed algorithm can not only detect the abnormal state of the extruder in operation in real time,but also have high accuracy.(4)Combining the proposed data filling method and anomaly detection method,an extruder data filling and anomaly detection system is designed and implemented on the basis of an energy management system of an aluminum profile production enterprise in Foshan.Through the demonstration of specific implementation projects,the feasibility of the proposed data filling method and anomaly detection method in practical application is illustrated.