Explainable attention-based fused convolutional neural network (XAFCNN) for tire defect detection: an industrial case study

Abstract

In the tire manufacturing industry, ensuring the quality of tires is of utmost importance because defective tires have the potential to fail explosively, particularly in high-speed driving situations such as races. To address this issue, thorough visual inspections after production are essential. Nevertheless, detecting defects in tires is a challenging task due to the various textures and structures they have. This paper introduces an Explainable Attention-based Fused Convolutional Neural Network (XAFCNN) model for tire defect detection. Special Attention Module (SAM) is used to prevent overfitting and improve local feature mapping, enabling the detection of small objects. The Grad-CAM heatmap method is utilized to offer significant visual indicators that enrich our comprehension of how the suggested model analyzes data and discerns relevant features within the tire samples. The model was trained on a dataset comprising 38,710 X-ray images of defective tires and 83,985 defect-free tire images, encompassing 15 defect types across 50 design patterns. The results demonstrate the model's exceptional performance, achieving a recall rate of 86.85%, a precision of 98.5%, an F1 score of 92.31%, and an overall accuracy of 95.40%. Notably, the model excels in identifying diverse tire defects and complex textures, making it highly effective in real-world scenarios. By providing both a substantial dataset and a high-performing model, this research advances automated tire defect detection, meeting the industry's imperative for accurate and reliable inspections, and ultimately enhancing human safety.