Automated Quality Inspection in Manufacturing with Computer Vision Systems Using Transfer Learning on ResNet50

Quality inspection is a necessity for each manufacturing system to need, as it marks the degree of quality that should be attained, estimate the degree of product defects, and enhance manufacturing effectiveness. Inspection methods that have involved the use of manned equipment and procedures that were not only time-consuming but also likely to involve a lot of errors are now changing due to the improvements in the computer vision systems from machine learning and deep learning. Transfer learning on ResNet50 is discussed in this paper to propose an optimized automatic inspection system. This work uses ResNet50 deep convolutional neural network that is pre-trained on natural images, then fine tuned on manufacturing datasets to detect and classify defects on some industrial products. The proposed system benefits greatly from the transfer learning paradigm since this approach greatly cuts the computational cost and data demand compared with other approaches. Experimental results show that, the proposed system can effectively detect the defects for the particular part and with high accuracy and with the recall values for each kind of defect are conspicuously high. Therefore integration of computer vision with the manufacturing process not only helps in increasing the inspection accuracy but also helps in decreasing the inspection time hence leading to real time inspection. Moreover, the study looks at the generalizability of the system to other manufacturing domains and the inclusive applicability of the approach. This work details how computer vision and transfer learning is reshaping the manufacturing landscape today, and provides a basis for anticipating future trends in AOI.