Research On Resistance Spot Welding Quality Detection Method Based On Deep Learning

Resistance spot welding is the most commonly used welding method in automobile manufacturing,which directly affects the reliability and safety of automobile.Due to the complex environment of spot welding,the quality problems of the solder joints of the vehicle often occur,which brings great hidden dangers to the safety of the body,so it is necessary to detect the quality of spot welding.However,the traditional detection method uses manual destructive sampling detection,which is not only inefficient,but also consumes a lot of manpower and material costs.With the rapid development of computer technology and artificial intelligence technology,data-driven spot welding quality inspection has become a research hotspot.However,the spot welding data collected in the actual production environment cannot be effectively tested because of the scarce number of fault samples.In addition,current spot welding quality discrimination methods based on data-driven have problems such as insufficient ability to extract timing information from spot welding data.These problems affect the establishment of spot welding quality inspection model.In order to solve the above problems,This paper focuses on the theory and method of resistance spot welding quality detection based on deep learning.the key work of this paper is as follows:1.In order to better solve the problem of insufficient spot welding fault samples,a spot welding fault sample generation model based on ISSL-GAN is proposed for the problems that traditional GAN is not easy to converge under the condition of scarce fault samples and low quality of generated fault samples.Firstly,the model uses the results of unsupervised clustering training to provide hidden features with fault category information to the generator,which effectively improves the sample generation performance of the model.Then,based on the ACGAN framework,the LSTM network is introduced into the generator structure and discriminator structure of ACGAN,and the loss function of ACGAN is modified,so that the true and false classification task and the fault classification task are unified,so that the generator focuses on the generation of high-quality fault samples.Finally,through statistical indicators and experimental verification,it is proved that the ISSL-GAN model has relatively good fault sample generation ability under the condition of scarce fault samples,and can expand the data set to effectively improve the spot welding quality detection ability.2.Aiming at the problem of insufficient extraction of data time series information by the current spot welding quality inspection model,a spot welding quality inspection model based on ABi LSTM-1DCNN is proposed.The model takes one-dimensional convolutional neural network and bidirectional long-term short-term memory network as the basic network structure,introduces the attention mechanism and residual connection mechanism,and improves the data temporal feature extraction ability and deep feature mining ability.At the same time,the model performs preprocessing operations such as missing value filling and batch effect elimination on the data to solve the problem of data loss and batch effect and improve the quality of samples.Experiments show that the model can effectively extract spot welding data information and has higher accuracy in spot welding quality inspection.3.In order to meet the requirements of real-time monitoring of spot welding quality in spot welding production,based on the above researched algorithm,a spot welding quality monitoring software system is designed,which realizes the recording and accurate analysis of spot welding data and spot welding quality data,and can monitor spot welding quality in real time,which improves the automation and intelligence level of spot welding inspection in the welding production environment.