Defect Classification on Automobile Tire Inner Surfaces using Convolutional Neural Networks

This study examines a method of applying a convolutional neural network to classify defects occurring on various patterns on the inner surface of a tire by conducting three-dimensional (3D) shape measurement using the light-section method on the inner tire surface. Classification objects are divided into “good parts,” “quasi-good parts,” and “defective parts.” “Good parts” and “quasi-good parts” are classified as non-defective, but “defective parts” are classified as defects. The experimentally obtained results revealed that a two-step classifier for classifying defects can improve classification accuracy. Additionally, results show that various patterns are classified as non-defective with high accuracy, demonstrating that the convolutional neural network method is effective for objects with greatly variable shape elements such as rubber products. This method is anticipated as a method that is suitable for defect inspection automation.