Research On Bolt Assembly Monitoring Based On SEMG Signal

At present,the main method of mechanical assembly monitoring is to fix various types of sensors on the production line or assembly tools,and monitor the assembly process of operators by collecting various types of data.This type of monitoring method has poor portability and requires a large assembly space.In response to this problem,this research applies wearable sensors and proposes a bolt assembly monitoring method based on surface electromyography(s EMG)signal.The neural network is used to estimate assembly force/torque,identify the type of bolt assembly process and detect assembly connector and tool.The specific research content is as follows:(1)The test bench for bolt assembly monitoring was designed and built,where s EMG signal and Inertial signals were collected by a wearable device Myo armband,image information and torque were recorded by a Realsense camera and torque sensor.Three data sets namely torque classification data set,torque regression data set,bolt assembly data set were established.The torque classification data set contains s EMG signal and torque label;the torque regression data set contains s EMG signals,Inertial signals and assembly torque;the bolt assembly data set contains s EMG signals and image information.Finally,according to the type of data and the frequency of collection,different methods of preprocessing are carried out.(2)The multi-segmentation parallel CNN model(MSP-CNN)is proposed to estimate the granularity classification of assembly torque based on s EMG signal.Firstly,a two-dimensional convolutional neural network(2D CNN)is applied to predict the torque through a classification method.Secondly,the influence of s EMG signal preprocessing method and pooling method on the prediction accuracy is compared.Thirdly,the error between the predicted torque of the 2D CNN model and the real torque is analyzed.Finally,the error of the predicted torque of MSP-CNN proposed by the research is analyzed.The results show that compared to maximum pooling,average pooling can improve the accuracy of CNN torque classification and recognition.Moreover,the MSP-CNN model can improve the accuracy of torque monitoring as well as solve the problems of non-convergence and slow convergence of 2D CNN model.(3)A method of applying regression neural network to monitor assembly torque is proposed to regress the assembly torque based on s EMG signal and Inertial signals.Heterogeneous kernels-based temporal convolutional network(Het-TCN)and the other based on a two-stream convolutional neural network(two stream CNN)are proposed and compared with CNN,long short-term memory(LSTM),and temporal convolutional network(TCN)models.The results show that,compared with using s EMG signals alone,performing an assembly torque data regression using a combination of s EMG and Inertial signals can significantly reduce the regression error.The proposed Het-TCN model has the best regression performance,with an average error of 3.31 N·m,an RMSE of 9.18%,and a coefficient of determination R~2 of 0.72.Experiments show that this method can be used for torque monitoring of bolt assembly.(4)An integrated monitoring method using neural network to identify the bolt assembly process,detect the serial number of the assembly connector and estimate the working angle of the assembly tool is proposed.Firstly,the two-stream convolutional neural network is used to identify the bolt assembly process using s EMG signals and optical flow characteristics.The accuracy of the test set reaches 88.3%.Secondly,the yolov5 detection algorithm is applied to identify the type and position of the assembly connectors and tools in the image.The results show that the yolov5s model requires the shortest training steps and has the best detection real-time performance,with an average detection speed of 35ms.Finally,the detection prediction maps and the feature point matching algorithm are applied to detect the serial number recognition of the assembly connectors.Two detection prediction maps at different times are used to estimate the working angle of the assembly tool.In this paper,the quantitative monitoring of bolt assembly torque and the qualitative monitoring of bolt assembly type are realized by using the neural network model with the s EMG signal as the main input information.The method has strong universality and is less affected by working environment.