Four “eyes” see better than two

Robots do monotonous workflows and less pleasant, repetitive tasks with brilliance. Combined with image processing, they become “seeing” and reliable supporters of humans. They are used in quality assurance to check components, help with the assembly and positioning of components, detect errors and deviations in production processes, and thus increase the efficiency of entire production lines. An automobile manufacturer is taking advantage of this to improve the cycle time of its press lines. Together with the latter, VMT Vision Machine Technic Bildverarbeitungssysteme GmbH from Mannheim developed the robot-based 3D measuring system FrameSense for the fully automatic loading and unloading of containers. Pressed parts are thus safely and precisely inserted into or removed from containers. Four Ensenso 3D cameras from IDS Imaging Development Systems GmbH provide the basic data and thus the platform for process automation.

The actual workflow that FrameSense is designed to automate is part of many manufacturing operations. A component comes out of a machine – here a press – and runs on a conveyor belt to a container. There it is stacked. As soon as the container is full, it is transported to the next production step, e.g. assembly into a vehicle.

Up to now, employees have been responsible for loading the containers. This actually simple subtask is more complex than one might think at first glance. In addition to the actual insertion process, the first step is to determine the appropriate free space for the part. At the same time, any interfering factors, such as interlocks, must be removed and a general check of the “load box” for any defects must be carried out. All these tasks are now to be taken over by a robot with a vision system – a technological challenge. This is because the containers also come from different manufacturers, are of different types, and thus vary in some cases in their dimensions.

For their fully automatic loading and unloading, the position of several relevant features of the containers must be determined for a so-called multi-vector correction of the robot. The basis is a type, shape, and position check of the respective container. This is the only way to ensure process-reliable and collision-free path guidance of the loading robot. All this has to be integrated into the existing production process. Time delays must be eliminated and the positioning of the components must be accurate to the millimeter.

To counter this, VMT uses four 3D cameras per system. The four sensors each record a part of the entire image field. This can consist of two containers, each measuring approximately 1.5 x 2 x 1.5 meters (D x W x H). Two of the cameras focus on one container. This results in data from two perspectives each for a higher information quality of the 3D point cloud. These point clouds of all four sensors are combined for the subsequent evaluation. In the process, registrations of relevant features of the container take place in ROIs (Regions of Interest) of the total point cloud. Registration is the exact positioning of a feature using a model in all 6 degrees of freedom. In other ROIs, interference contours are searched for which could lead to collisions during loading. Finally, the overall picture is compared with a stored reference model. In this way, the containers can be simultaneously checked for their condition and position in a fully automated manner. Even deformed or slanted containers can be processed. All this information is also recorded for use in a quality management system where the condition of all containers can be traced. The calibration as well as the consolidation of the measurement data and their subsequent evaluation are carried out in a separate IPC (industrial computer) with screen visualization, operating elements, and connection to the respective robot control.

The main result of the image processing solution is the multi-vector correction. In this way, the robot is adjusted to be able to insert the component at the next possible, suitable deposit position. Secondary results are error messages due to interfering edges or objects in the container that would prevent filling. Damaged containers that are in a generally poor condition can be detected and sorted out with the help of the data. The entire image processing takes place in the image processing software MSS (Multi Sensor Systems) developed by VMT. FrameSense is designed to be easy to use and can also be converted to other components directly on-site.

On the camera side, VMT relies on Ensenso 3D cameras – initially on the X36 model. The current expansion stage of FrameSense is equipped with the Ensenso C variant. The reasons for the change are mainly the better projector performance – thanks to a new projection process – as well as a higher recording speed. In addition, the Ensenso C enables a larger measuring volume. This is an important criterion for FrameSense, because the robot can only reach the containers to be filled up to a certain distance.

The Ensenso C addresses current challenges in the automation and robotics industry. Compared to other Ensenso models, it provides both 3D and RGB color information. Customers thus benefit from even more meaningful image data. The housing of the robust 3D camera system meets the requirements of protection class IP65/67. It offers a resolution of 5 MP and is available with baselines from current to approx. 455 mm. This means that even large objects can be reliably detected. The camera is quick and easy to use and addresses primarily large-volume applications, e.g. in medical technology, logistics or factory automation.