Research On The Recognition Method Of Torque Wrench Indicating Value

At present,torque wrench is widely used in the assembly of various important equipment,which ensures that the torque of bolt and nut is within the specified range.Torque wrench often occurs errors that require regularly verification.With more and more applications of big-scale torque wrenches,it is necessary to design automatic verification system with electric loading function.The automatic verification system needs the automatic identification of the indicating value as the basis,but the traditional method can not meet the requirements of the high accuracy of the automatic verification system.In this paper,the convolution neural network with outstanding performance in the field of pattern recognition is used to identify the indicating value.According to the identified indicating value image,a suitable model of convolutional neural network is selected.Firstly,the paper studies the function and usage of convolutional layer and pooling layer that the most typical convolutional layer in the feature extraction stage.Then briefly describes the compositions of classification stage.The size of the input image and the size of the convolution kernel are selected according to the actual situation of torque wrench to determine the number of network layers.The scheme of extracting comprehensive features from simple features is given and the classifier type is selected.Finally the indicating value identification model of convolution neural network is determined.Indicating value recognition model based on convolution neural network need train its weight and bias.By calculating the minimum value of the model loss function,the weight and bias of the model are updated by the backward propagation process.Aiming at the problem of gradient disappearance in convolutional neural networks,the type of activation function suitable for identification of indicating value is selected.Because of the inability of the activation function to control the output amplitude,Xavier weight initialization method that can remedy this problem is selected.Then the value of the learning rate which causes the loss function to drop rapidly is selected for the torque value recognition problem of the torque wrench.Finally,by training the model,the loss rate and the correct rate of the model are obtained,so as to confirm that the model meets the requirement of torque wrench indicating value.Torque wrench automatic verification system is designed to realize the functions of indicating value identification.The system can realize torque loading or unloading,automatic verification,verification data management and maintenance,verification certificate generation and printing.The part of indicating value recognition makes use of the obtained network model for character recognition.Realizing the automatic control of the torque loading mechanism is on the basis of the value identification,in order to achieve the verification automation.Finally,the paper designs interface of verification system.Finally,the paper summarizes the main work of the dissertation,looks forward to the further research direction of the torque wrench indicating value recognition,and discusses the issues to be studied in the next stage.