Objective. Automatize the inspection of critical assembly points
Our client, a large automotive seats manufacturer, would like to automate the inspection of 5 critical points in the assembly of its products. Until now this inspection has been carried out visually by an operator. The new system must be able to record all inspections and detect when one of the assemblies has been carried out correctly and when it has not.
The variability between the different models and the fact that there are several review points adds complexity to the design of the solution. They needed a reliable, robust and economical solution for this case.
Solution. Computer vision with ATRIA algorithm with a collaborative robot
The first step consisted in collecting all the requirements, cycle time, models and cases that could happen for each of the 5 control points that we were going to implement. It was very important to know what was considered a ‘good’ assembly and a ‘bad’ one to subsequently be able to train the computer vision system.
Afterwards, we begin with the definition and design of the equipment. We selected the camera that would be in charge of 4 of the inspections and the laser depth sensor that was going to be in charge of the fifth. We proposed using a collaborative robot to carry the camera and laser with the objective of capturing images to inspect the different points.
The next step consisted in programming the robot’s trajectories to be efficient and achieve good image taking. Later, we installed and integrated the system at the customer’s factory so we could begin the rest of the programming. This is where we carry out the development of communications between all elements of the system and the client’s internal system.
At this point we were able to begin taking images to train the computer vision system. In this case we designed and trained our own algorithm based on convolutional neural networks.
The final step consisted in the validation of the communications and inspections by our client and the training we gave about the system and the developed algorithm.