{"title":"随机森林回归预测UR10协作机器人末端执行器的异常状态,从有目的的故障数据","authors":"Ethan Wescoat, Matthew Krugh, Laine Mears","doi":"10.1016/j.promfg.2021.06.064","DOIUrl":null,"url":null,"abstract":"<div><p>Unexpected downtime from equipment failure has increased due to a production line’s mechanization to meet production throughput requirements. Manufacturing equipment requires accurate prediction models for determining future failure probability in maintenance scheduling. This paper explores using generated failure data under contrived failure scenarios in training a model for a robot with different combinations of data features. Failure data are generated by inducing an anomalous state in the robot arm. The anomalous state is created by attaching weights at the robot end-effector. A random forest regression model diagnoses the anomalous state and determines the anomalous state progression after gathering data. Three different regression models were trained to test accuracy based on different feature selections. The random forest regression predicted 92% of the robot joint operations through five-fold cross-validation, an anomaly in a robot joint 99% of the time, and the correct anomaly state-based on the confusion matrix, 85% of the time. In future research, the anomalous state will represent more targeted component failures on the system through purposeful permanent damage of the robots’ components. Future datasets generated will train other machine health algorithms for estimating component and system damage.</p></div>","PeriodicalId":91947,"journal":{"name":"Procedia manufacturing","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2021-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1016/j.promfg.2021.06.064","citationCount":"3","resultStr":"{\"title\":\"Random forest regression for predicting an anomalous condition on a UR10 cobot end-effector from purposeful failure data\",\"authors\":\"Ethan Wescoat, Matthew Krugh, Laine Mears\",\"doi\":\"10.1016/j.promfg.2021.06.064\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><p>Unexpected downtime from equipment failure has increased due to a production line’s mechanization to meet production throughput requirements. Manufacturing equipment requires accurate prediction models for determining future failure probability in maintenance scheduling. This paper explores using generated failure data under contrived failure scenarios in training a model for a robot with different combinations of data features. Failure data are generated by inducing an anomalous state in the robot arm. The anomalous state is created by attaching weights at the robot end-effector. A random forest regression model diagnoses the anomalous state and determines the anomalous state progression after gathering data. Three different regression models were trained to test accuracy based on different feature selections. The random forest regression predicted 92% of the robot joint operations through five-fold cross-validation, an anomaly in a robot joint 99% of the time, and the correct anomaly state-based on the confusion matrix, 85% of the time. In future research, the anomalous state will represent more targeted component failures on the system through purposeful permanent damage of the robots’ components. Future datasets generated will train other machine health algorithms for estimating component and system damage.</p></div>\",\"PeriodicalId\":91947,\"journal\":{\"name\":\"Procedia manufacturing\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2021-01-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://sci-hub-pdf.com/10.1016/j.promfg.2021.06.064\",\"citationCount\":\"3\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Procedia manufacturing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S2351978921000755\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Procedia manufacturing","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2351978921000755","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Random forest regression for predicting an anomalous condition on a UR10 cobot end-effector from purposeful failure data
Unexpected downtime from equipment failure has increased due to a production line’s mechanization to meet production throughput requirements. Manufacturing equipment requires accurate prediction models for determining future failure probability in maintenance scheduling. This paper explores using generated failure data under contrived failure scenarios in training a model for a robot with different combinations of data features. Failure data are generated by inducing an anomalous state in the robot arm. The anomalous state is created by attaching weights at the robot end-effector. A random forest regression model diagnoses the anomalous state and determines the anomalous state progression after gathering data. Three different regression models were trained to test accuracy based on different feature selections. The random forest regression predicted 92% of the robot joint operations through five-fold cross-validation, an anomaly in a robot joint 99% of the time, and the correct anomaly state-based on the confusion matrix, 85% of the time. In future research, the anomalous state will represent more targeted component failures on the system through purposeful permanent damage of the robots’ components. Future datasets generated will train other machine health algorithms for estimating component and system damage.