Research On Fault Diagnosis Technology Of Industrial Robot Based On Deep Learning

Industrial robots have a wide range of applications in the manufacturing industry and related fields by virtue of their own high-efficiency and high-intensity working mode.Failure of industrial robots will often result in the stagnation of production lines,the consumption of human and material resources of the enterprise,and the serious loss of people's lives and property.Due to the long cycle and low diagnosis efficiency of traditional fault diagnosis methods,the accuracy of fault recognition is difficult to guarantee.Research on intelligent fault diagnosis systems for industrial robots has become an important part of industrial production.Based on deep belief network and multi-source information fusion technology,this thesis establishes a fault diagnosis model suitable for industrial robots.Through the analysis of the vibration signals of industrial robot joints and end effectors,a fault diagnosis model is established.The main research contents of this thesis are as follows:Analyze the common fault forms,fault characteristics and fault causes of industrial robots,determine the vibration signal of the industrial robot as the research object,complete the creation of the industrial robot fault diagnosis model and conduct experimental verification.The wavelet transform and the theory of information energy entropy are used to extract the features of the vibration signal of the industrial robot,and the energy entropy normalized feature vector of the vibration signal is constructed.Take KR-3-R540 robot,data acquisition and storage system,data analysis system and acceleration sensor as experimental equipment.The acceleration sensor is attached to the robot shell,and the vibration signal of each joint and end effector of the KR-3-R540 robot collected by the acceleration sensor is obtained through the data acquisition and storage system,and analyzed by the data analysis system.Due to the interference of the motion frequency and resonance frequency of the industrial robot,it is necessary to filter the vibration signal collected by the acceleration sensor.Use wavelet transform and information energy entropy to process the filtered vibration signal.By comparing the processing results of the vibration signal under different parameters,the parameters such as the optimal wavelet decomposition layer number and the number of wavelet bases,as well as the energy entropy calculation equation and the normalization equation are determined.Taking the normalized vector of energy entropy corresponding to the vibration signal as the experimental sample,it is divided into a training set and a test set to complete the forward unsupervised training and reverse optimization fine-tuning of the DBN network parameters and the test of the model fault recognition accuracy.The DSmT theory in the multi-source information fusion technology is used to fuse the decision-making level of the fault diagnosis model.The output layer of the DBN network is used as the fault evidence to calculate the evidence conflict metric.According to the high conflict evidence between the evidences,the combination rules and decision rules of DSm T under the open identification framework are selected for decision-making layer fusion.Design a robot experiment under multiple concurrent fault conditions to verify the applicability of the industrial robot fault diagnosis model based on multi-source information fusion under multiple concurrent fault conditions.The fault recognition accuracy rate can reach 94.3%.