Forging defect detection method based on deep learning

Abstract

The invention discloses a forging defect detection method based on deep learning, which comprises the following steps: firstly, establishing a YOLOv4 algorithm model for detecting forging defects, then training the YOLOv4 algorithm model, and then carrying out defect detection on a picture of a to-be-detected forging by adopting the trained YOLOv4 algorithm model. The method is characterized in that after residual errors in a feature extraction network of a YOLOv4 algorithm model are connected, a CBAM attention module is inserted, and features are screened. The method has the advantages of being good in robustness, capable of efficiently and accurately conducting defect detection and the like.

Description

technical field [0001] The invention relates to the technical field of visual inspection, in particular to a method for detecting defects of forgings based on deep learning. Background technique [0002] Diesel engine is an important source of power for ships, and the quality of diesel engines directly affects the overall performance of ships. Defect detection of diesel machined workpieces is an important link to ensure product quality. However, due to the complex on-site conditions and comprehensive standards for defect detection, many stations still rely mainly on manual visual inspection. Take the connecting rod as an example. The connecting rod is one of the core components of the engine. It connects the crankshaft and the piston. The connecting rod is generally produced by forging. 100% defect detection, but its detection standards are more complicated, and currently it is usually manual visual inspection, with low efficiency and accuracy. If the surface defect of the...

AI Technical Summary

Problems solved by technology

This detection based on traditional image processing requires artificially designed feature extractors to extract defect features, and different feature extraction methods are designed according to different tasks. In the face of problems such as low contrast between defect and non-defect areas, high noise and defect similarity, limitations Larger, not widely applicable

Because the surface of the connecting rod is not smooth, and there are interference factors such as stray spots, the types, positions, and sizes of defects are also diverse. However, traditional detection requires artificially designed feature methods, which lacks pertinence for diverse defects and is less robust. Difference

Images