Classification of wellhead device defects based on CNN-GA-BP model

During the long-term operation of wellhead devices, various types of defects appear due to scouring and corrosion, poor construction, and other reasons. Timely and accurate identification of wellhead device defects is an important factor in ensuring wellhead integrity. In this paper, we design a wellhead device defect classification model based on a combination of convolutional neural network (CNN) and genetic algorithm optimization-bp neural network (GA-BP) using the finite element simulation data of leakage detection and experimental data as samples. Based on the characteristics of leakage defect data with three dimensions: axial, radial and circumferential, the three-axis components of leakage defects are used instead of the color image R, G and B channels, and input to the convolutional neural network to achieve data fusion and automatic feature extraction of the three-axis components of leakage defects. In order to improve the problem that bp neural networks are prone to fall into local optima, a genetic algorithm is used for optimization to obtain optimal initial weights and thresholds. The GA-BP model is used instead of the self-contained softmax classifier in CNN to accurately classify the extracted features and achieve intelligent classification and recognition of wellhead device defects. Based on the leakage data for experimental analysis and comparison of various classification methods, the CNN-GA-BP combination model established in this paper has the highest classification accuracy, reaching 91.25%. The experimental results illustrate that the CNN-GA-BP model established in this paper has a good classification and recognition effect on various types of defects of wellhead devices.