基于卷积神经网络的盘式刹车片表面缺陷检测

Surface Defect Detection of Disc Brake Pad Based on Improved Convolutional Neural Network

汽车刹车片质量关系车辆驾驶安全,目前汽车刹车片表面缺陷主要采用人工抽检和机器视觉检测的方法,人工抽检存在效率低、易出现漏判与误判表面缺陷的问题,机器视觉检测则需要依靠被检测对象的特定特征进行检测。为此,以业界广泛应用的盘式刹车片为研究对象,提出一种改进AlexNet卷积神经网络模型,即AlexNet6_BN模型,对盘式刹车片进行表面缺陷检测。对经典AlexNet卷积神经网络模型中的卷积层进行调整,增加了1层卷积层和1层池化层,调整首层卷积核大小为13×13以提取更为显著的样本特征,并在每层卷积层后用批量标准化代替原来的局部响应归一化以加快网络收敛速度。实验测试表明,改进后的网络模型对刹车片表面缺陷识别的准确率均高于AlexNet、VGG16等经典网络,其检测准确率达到97%。

The quality of automobile brake pad was related to the safety of vehicle driving.At present,the surface defects of automobile brake pad are mainly inspected by manual sampling method and machine vision.Manual sampling inspection has the problems of low efficiency and prone to omission and misjudgment of surface defects.And machine vision inspection needs to rely on the specific characteristics of objects to be detected.Taking disc brake pads widely used in the industry as the research object,an improved AlexNet convolution neural network model was proposed to detect surface defects of disc brake pads to improve the defection efficiency and detection accuracy.And the convolution neural network model was named AlexNet6_BN model.AlexNet6_BN convolution neural network added 1 convolution layer and 1 pooling layer on the basis of AlexNet network,which has six convolution layers and four pooling layers.And to extract more significant sample features,the size of the convolution kernel of the first layer was adjusted to 13×13 in AlexNet6_BN network.To accelerate the network convergence speed,the original local response bormalization was replaced by batch normalization after each convolution layer in AlexNet6_BN network.Experimental results show that the improved network model can improve the efficiency,and the accuracy of the improved network model is higher than AlexNet,VGG16 and other classical networks.The detection accuracy reaches 97%.