Defect Detection in Disc Brake Pads Using an Improved AlexNet Convolutional Neural Network

In the last few years, with the rapid advancement of the automotive industry, disc brake pads have been increasingly widely used in automotive braking systems. However, due to long-term friction and high temperature effects, they are prone to defects such as cracks, wear, and dents. To raise the accuracy and efficiency of disc brake pad defect detection, the study first optimized image preprocessing and, based on the AlexNet network, introduced a multi-scale convolution module, batch normalization, and global average pooling for structural improvement. Finally, a new type of disc brake pad defect detection model was proposed. The experiment findings denoted that the highest defect detection accuracy of the model was 97.12%, and the mean detection time was 10.53 milliseconds. Compared to other advanced methods, this new model had the lowest missed detection rate of only 1.78%, the lowest resource consumption rate of 36.45%, and the lowest model complexity of 53.23%. This denotes that the new model can efficiently and accurately detect various kinds of defects in disc brake pads, which is suitable for practical industrial inspection scenarios and provides a reliable technical support for the field of disc brake pad defect detection.