Study on Defect Identification Method of Pipeline Weld Based on Integrated Machine Learning
To improve the automatic identification of different defect types in pipeline weld,based on the excavated X-ray weld images,and through comparing the variance information of the pooled domain and characteristic-graph,the pooled layer operation of the traditional convolutional neural network is modified by using the square difference pooling method,and then the selection of CNN model hyperparameters is realized by using whale algorithm.Finally,a integrated machine learning model is formed for the classification of weld defects,and are compared and verified with other models.The results show that the whale algorithm can optimize parameters such as the number and size of convolutional kernel,the number and size of pooled kernel,the type of convolutional kernel activation function,and the learning rate in a short time.In the classification results of integrated machine learning model,the classification accuracy of unfused and unwelded defects is the highest.The proposed model is superior to other CNN models and SVM models in classification accuracy,training time and stability.The research results can provide practical reference for the identification of weld defects in other pressure vessels or pressure pipelines,and have strong universality and expansibility.
万方数据
维普
