IMAGE ENHANCEMENT IN INDUSTRIAL WELDING ENVIRONMENT WITH IMAGE PROCESSING TECHNIQUES

With the increase and acceleration of production capacity, computer-based control mechanisms are becoming increasingly common in industrial applications. The use of intelligent welding robots in the welding industry is increasing due to their instant decision-making and application capabilities. For this reason, computer vision systems and image processing algorithms are increasingly used. Although visual limitations in sensors and industrial environmental conditions (arc, noise, dust, etc.) cause problems in robotic welding applications, computer-controlled systems achieve much more efficient results than operator-controlled systems. One of the most important points here is the applicability and stability of the algorithm to the system. In this study, considering the computational load of image processing algorithms and the negative effects of this computational load in moving environments, a more stable and efficient image feature extraction algorithm was tried to be created for robotic welding applications. After the welding process, object recognition was performed by performing object feature matching with the help of samples taken from the weld images. A new algorithm was created to recognize welding processes that differ from each other in some aspects with multiple samples and even to detect different types of welds. This algorithm reduces the images to gray level and performs a pre-processing step to remove noise with a filtering process, then detects the weld points with the help of predetermined templates and decides how accurately these points are made. Thanks to the NCC Template Matching method used in the algorithm, the running time of the algorithm is accelerated and more accurate results are obtained by introducing more than one template Experimental method aimed both to calculate the accuracy rate in case the same type of weld operations are different from each other and to recognize the operations performed with different types of welds. While the detection level was around 60% in images without image preprocessing, the detection rate exceeded 70% in images with image preprocessing. In the experiments conducted on the images taken with the Template Matching algorithm, it was observed that the detection rate increased to around 75% at different threshold values. In addition, with the region of interest selection and NCC method, the running time of the algorithm was reduced to 190 ms on average. Considering the results obtained in the experiments, the algorithm significantly improved the accuracy rates of spot welding and the differentiation of different weld types. By using sufficient light welds and correct experimental equipment, the success rate of the Template Matching algorithm has been increased and the processing load has been alleviated. The effect of external environmental conditions, which is considered the biggest disadvantage of the algorithm, is minimized with lighting elements.