Research On CPS Of NC Turret Reliability Test

Numerical control(NC)turret is one of the important parts of CNC machine tool,which has a direct impact on the availability of CNC machine tool.Therefore,it should give high priority to carry out reliability test.For one thing,there is no interaction with existing reliability test systems of NC turret,and the process of reliability test has been still offline.For another thing,there is no anomaly detection method during reliability test,which means that the present reliability data are not able to give support for fault data mining,fault early warning and preventive maintenance.In view of the above problems,this paper develops the Cyber-Physical System(CPS)of NC turret reliability test.The architecture of NC turret Reliability Test-CPS(RT-CPS)emphasizing data cycle is desigened;and the cell level RT-CPS in the laboratory is also built.Furthermore,a working interval classification method and an anomaly detection method considering multi-working process and multi-working condition during NC turret reliability test is proposed and calibrated.The main contents of this paper are as follows:1.A RT-CPS architecture is designed,which bridges data among field test,bench test and virtual test.This architecture includes cell level,system level and system of system level.The key technologies about RT-CPS are summarized.In addition,the cell level RT-CPS of NC turret is built in the laboratory,including six parts: test unit,monitoring unit,primary decision-making unit,control execution unit and communication unit.The primary decision-making unit is divided into two parts: working interval extraction and classification,and anomaly detection.2.The method of working interval extraction and classification is proposed.Taking the vibration signal as the research object,this paper presents the dynamic double threshold method to detect working endpoint.Secondly,there are two classification methods of working interval.The first one is based on discrete feature,where the support vector machine-stepwise regression(SVM-SR)is introduced to choose the best features.Another method is based on shape feature,where the creative approaches about linear segmentation standard and symbol determination are proposed.Finally,the classification process of cell level RT-CPS under different features is summarized.3.The relationship model between signal and working conditions is established.The mathematical models and data-driven models are discussed respectively between different working interval and relevant working conditions,such as release/lock signal and oil pressure,transposition signal and transposition interval as well as unbalanced moments,and loading signal and loading force.Through the above models,the theoretical working conditions can be deduced from the signal features,and vice versa.4.The anomaly detection method considering working conditions changing is presented.Comparing the observed features with the theoretical features,this paper proposes the multivariate Gaussian distribution model(MGDM)based on deviation features,and obtains the probability density threshold.This model can identify the cause of abnormal signal as abnormal working condition or abnormal devices.Meanwhile,compared with the MFDM without considering working conditions,this algorithm is proved to be advanced.5.The validity of the above methods is verified by experiments.Firstly,the parameters about working interval extraction are calculated and calibrated by single and multiple samples.Secondly,the accuracy of working interval classification is compared between two classification methods in three training sample levels.At the same time,the accuracy of working interval's classification between the methods proposed in this paper and traditional classification methods is compared.Thirdly,the accuracy of anomaly detection under known and unknown working conditions is calculated.