Data-driven Intelligent Fault Diagnosis Method Of Industrial Robot

Industrial robot is an indispensable role to realize intelligent production and promote industrial upgrading.In order to ensure the healthy and stable operation of industrial robots and ensure the long-term and stable productivity of industrial production line,it is necessary to establish a reliable fault diagnosis system for industrial robots.With the improvement of intelligent level,there are more and more data-driven fault detection methods in the field of robot fault diagnosis.However,many methods rely on the data of external sensors.Moreover,the previous way of shallow learning for fault diagnosis method depends on the manual derivation of signal features and the selection of appropriate classifier combination.It depends heavily on expert experience.The optimization process of fault diagnosis model is timeconsuming and its generalization ability is poor,which is difficult to meet the actual needs of industrial production fault diagnosis.At the same time,in the task of robot fault diagnosis,there are some problems,such as lack of fault data,imbalance of categories and so on.In this paper,aiming at the related problems of robot fault diagnosis,taking multi axis industrial robot as the research object,and focusing on the fault diagnosis of its core components,a fault diagnosis algorithm based on improved ML-SRIPCNN-1D is proposed.The algorithm uses the initial operation data of industrial robot to input the model.It does not require experts to use knowledge to extract features,then it can identify faults.The algorithm uses the initial operation data of industrial robot to input the model.It does not require experts to use knowledge to extract features,then it can identify faults.The main contents of this paper are as follows:(1)In view of the lack of fault data and the difficulty in obtaining fault data of industrial robots,an experimental platform was set up in the factory,a component replacement experiment was set up to collect the fault data of industrial robots,and the fault data set of industrial robots was established.(2)Aimed at the problems of low accuracy of fault diagnosis and its dependence on expert experience and knowledge,this paper proposed a fault diagnosis model of industrial robot based on improved SRIPCNN-1D.Firstly,I analyzed the operation data of the robot in the original time domain,and enhanced the fault data by random sampling and using Mixup algorithm.Secondly,SRIPCNN-1D was used to calculate the original operation data of industrial robot end-to-end.The algorithm achieved more than 99% diagnosis accuracy on the single fault data set of industrial robots,and the effectiveness was verified by comparing with WDCCN and CNN-1D models.(3)Aimed at the composite fault diagnosis problem of industrial robot,an improved MLSRIPCNN-1D fault diagnosis model was proposed.ML-SRIPCNN-1D model had the characteristics of fewer model layers,fewer training parameters and strong model expression ability.It was suitable for online diagnosis of robot faults.The model had achieved more than98% diagnostic accuracy on the composite fault data set of multi axis industrial robot.At the same time,it was compared with MLCNN,WT-MLCNN,T-FSM-MLCNN and single fault diagnosis model to verify the effectiveness of the proposed model.And we also studied how different sample frequencies affect the fault diagnosis effect of the model established by the algorithm.The experimental results showed that the model is still effective by adjusting the sampling interval to 1 second.(4)Established a fault diagnosis system for industrial robots,designed and implemented the overall architecture and main functional modules of the fault diagnosis system for industrial robots.The effectiveness of the algorithm was verified by connecting the actual running industrial robot.