In the mechanical cutting process, the surface defects of the workpiece are an important indicator of cutting quality and also reflect the condition of both the machine tool and the cutting tool. Effective detection of defects on the surface of the workpiece plays an important role in adjusting the processing conditions promptly, reducing losses, improving the utilization rate of the workpiece, and maintaining the normal operation of the equipment. To address the challenge of detecting surface defects on workpieces, an inspection method based on an improved Single Shot Multibox Detector (SSD) model is proposed. The method simplifies the detection model and reduces the computation by proposing a DH-MobileNet network instead of a VGG16 network in the SSD structure. The inverse residual structure is also used for position prediction, and null convolution is used instead of a down-sampling operation to avoid information loss. A scanning electron microscope was used to obtain the surface image of the workpiece. A dataset of workpiece surface defects was constructed and expanded, then used to train and test the model for detecting three common types of high-frequency defects: peel-off, chip adhesion, and scratches. The effect was compared with YOLO, Faster R-CNN, and the original SSD model. The detection results show that the method can detect the defects on the surface of the workpiece more accurately and quickly, which provides a new idea for defect detection in real industrial scenarios.
