Research On Image Recognition And Evaluation Of Bolster Hanger Magnetic Particle Flaw Detection Based On Improved YOLOv5

As a key component of the spring suspension system for rocking pillow,the bolster hanger is prone to fatigue cracks due to factors such as excessive dynamic loads and long service life,which has a significant impact on railway safety.Currently,most of the magnetic particle inspection work for bolster hanger is done manually,which is easily influenced by subjective factors of the operators and has low work efficiency.At the same time,the existing defect detection algorithms have poor robustness and low inspection accuracy.Therefore,this paper proposes an improved magnetic particle inspection image recognition algorithm based on YOLOv5,and the main research contents are as follows:The main form of damage to the bolster hanger is fatigue cracks,this article mainly focuses on image recognition of fatigue cracks in bolster hanger,analyzes the characteristics of magnetic traces and unrelated magnetic trace images,and improves the quality of magnetic particle inspection images through data labeling,data augmentation,and image denoising based on bilateral filter.A magnetic particle testing dataset for bolster hanger is established.To address the issue of similarity between non crack features and crack features in the background region,the SimAM mechanism is first added to the YOLOv5 backbone network to enhance the model’s sensitivity to crack information and enhance its anti-interference ability against background regions;To address the issue of information loss during Neck partial feature fusion,the BiFPN network is introduced to fuse feature maps of different scales.By repeatedly upsampling and downsampling,features from different levels are integrated,achieving bidirectional connection and weighted fusion of features,thereby reducing the loss of underlying information.Build a flaw detection image acquisition system for bolster hanger,and use the improved YOLOv5-SA-BF model to perform performance testing on the inspection images.The experiment shows that the improved algorithm’s Map index reaches 98.08%,and the Recall index also increases to 96.43%.This model effectively solves the problems of background misjudgment and low detection accuracy,and meets the needs of magnetic particle inspection image recognition for bolster hanger.In order to accurately evaluate the quality of the bolster hanger,a three-dimensional model of the swing bolster crane is constructed based on defect identification.The number,type,size,and position of cracks are analyzed,and the transformation relationship between object and image space is obtained through size calibration to measure the crack length.Based on the magnetic particle testing standard of the bolster hanger,a visual inspection result evaluation report is intelligently output.At the same time,we have developed a magnetic particle flaw detection image recognition software for bolster hanger,which achieves defect detection and flaw detection evaluation functions,providing a solution for the intelligent magnetic particle flaw detection of bolster hanger,and has certain application reference value.