Caterpillar is a manufacturing company in which quality is a top priority. In the production process of several products, there are quality gates that supervise the quality of pieces through visual inspections. This visual inspection accounts for approximately 12% of the total production costs. However approximately 28% of defects in the production lines are missed due to human error. Accounting that there may between 248 and 500 inspection points in a particular production line missing defects represent an important cost. In this report, a prototype of an inspection cell that detects the presence of welds in a metal beam was developed. It involves 3 components, the simulation of the prototype cell, the actual physical prototype and an offline inspection script. The simulation was done in CoppeliaSim since it accurately reflects the real dimensions and behaviour of the physical prototype. The physical prototype cell involved designing a jig that connected the camera to the tip of the UR10e, and programming an inspection routine that takes picture of both sides of metal beam, focusing on the individual welds. Lastly, the offline inspection script included a pipeline that takes in a set of pictures of welds, counts the welds and yields a spreadsheet with the results. Additionally, the model used for detecting welds was Yolov8, which is a computer vision deep neural network designed to detect and classify objects. The results of the models where outstanding having a perfect score in the test data. These successful result can be the foundation of a new computer-aided inspection points in the quality gates of Caterpillar. Additionally, due to advantages of collaborative robots and computer vision, it can be implemented incrementally over a time span and reduce production costs, both in labor and in missing defects.