Surface Defect Detection of Bearings based on Machine Vision

Machine vision assessment methodology has become increasingly appealing for manufacturing automation due to innovations in noninvasive technologies such as eddy current and ultrasonic testing, which have enhanced the circumstances for bearing defect identification. At this point, manual detection results in low lifespans and reliability. So, we present an innovative rider optimization-driven mutated convolutional neural network (RO-MCNN) technique for surface defect detection of bearings based on machine vision. To evaluate the effectiveness of the suggested approach, samples of the bearing surface with various defects were gathered. The raw data specimens are denoised using a Gaussian filter, and the defect-oriented surface patterns are then extracted using a local binary pattern (LBP) technique. Subsequently, the MCNN model is designed to identify and categorize the various sorts of defects. Experimental results obtained high accuracy (99.0%), F1-score (98.7%), recall (98.6%) and precision (98.5%), which validate the greater of RO-MCNN over existing methods, demonstrating its capability in robustly detecting and classifying bearing defects with high precision and reliability, thereby advancing the efficacy of machine vision in industrial defect assessment. The MCNN model’s performance is improved and the loss function is decreased by using the RO method. The results of the experiments showed that the suggested RO-MCNN technique outperforms current strategies in terms of bearing defect type detection and classification.