Research On Ultra-narrow Gap Welding Quality Assessment Method Based On Deep Learning

In recent years,with the rapidly development of science and technology,the demands of various metal structures are high strength,large-scale and thick plate.At present,the commonly used welding methods have some shortcomings.Flux strip restrained arc ultra-narrow gap welding method is single-pass multi-layer welding,which meets the high efficiency,energy saving and green welding concept.Therefore,it is of great significance to apply this method to welding production.However,the process of ultra-narrow gap welding is more complex than other welding methods.The arc stability in the welding process is affected by various factors,deviation welding and slag inclusion are easy to occur.Aiming at the problems of unobservable welding quality,difficulty in on-line detection and prone to defects in ultra-narrow gap welding,combined with welding methods and welding technology,the deep learning theory is applied to the evaluation of ultra-narrow gap welding quality,in order to realize the non-destructive evaluation of ultra-narrow gap welding quality,and lay a foundation for further on-line evaluation and prediction of ultra-narrow gap welding quality.This paper mainly introduces the ultra-narrow gap welding test platform and designs the signal acquisition system.On the basis of welding test platform and signal acquisition system,welding test was designed,a large number of welding process signals were collected,and the corresponding other welding process parameters were recorded.By comparison,the median filtering algorithm with better filtering effect is selected to process the signal,and 22 characteristic parameters are extracted from the welding process parameters and the welding process signals as the sample features of the welding quality evaluation.The shortcomings of shallow neural network are analyzed,and the evaluation models of ultra-narrow gap welding quality are established by using BP neural network and support vector machine respectively.The evaluation accuracy is relatively low,which further proves that the traditional neural network is not suitable for the welding process with complex process and unclear mechanism.In this paper,the theory of deep learning is introduced,and a model of ultra-narrow gap welding quality evaluation based on deep learning is established under the framework of TensorFlow to evaluate the welding quality.In order to ensure the accuracy of the evaluation results,50 experiments were carried out on BP,SVM and DNN respectively.By comparison,the deep neural network has achieved a higher accuracy of model recognition.In summary,compared with the traditional shallow neural network,the model based on deep learning can better evaluate the welding quality of ultra-narrow gap welding.In addition,the experimental results also show that the extracted characteristic parameters can better reflect the welding quality.