With the constant growth of competitiveness at an industrial level, the need for faster and more precise processes arises. Currently, the key to success in the industry is the implementation of automatic and digital systems. The product quality control is a central section of a production line, requiring rigorous product checks and the exclusion of potential defects, with the aim of obtaining the highest degree of customer satisfaction. In the combustion devices industry it plays an extremely important role. In the painting process, several types of defects can appear: e.g. scratches, dents, lack or excess of paint, etc. Currently, the detection of these defects is performed, in most cases, manually by specialized workers, which implies the possibility of errors of human nature as well as consequences for the workers who place themselves many times in unergonomic positions. In order to eliminate this factor, an automatic system for detecting defects in the painting process of combustion device covers is proposed. It is based on the deflectometry inspection method that is widely applied in defect detection systems, but has not been used, as far as we know, for the detection of defects in combustion device covers. Some laboratory tests were carried out with a television and various components in order to find the best configuration to the image acquisition. YOLOv8 algorithm was used to detect the various defects in a single image, allowing both the positioning and counting of defects. A dataset with images of crater type defects was built using image processing and image augmentation to train the models.