For green technologies to be sustainable in the overall cycle, the remanufacturing of worn-out devices and parts is crucial. The remanufacturing of wind turbine rotors, gearwheels, battery cells or hydrogen tanks, for example, minimizes the environmental impact by requiring fewer raw materials and energy-intensive processing steps compared to new production and avoiding additional material transport.

As the wear mainly affects the shape or surface properties, remanufacturing has so far been associated with a high workload. Even with a robot-based machining process, the current state of the art requires very frequent manual and therefore expensive adaptation of the robot program. This often makes new production more economical, although it is significantly less sustainable.

The aim of the RoboGrind research project is therefore to develop an AI-based, flexible automation solution that allows the robot to program and set itself up independently for the machining task. RoboGrind is an InvestBW-funded joint project between the University of Stuttgart, DHBW Karlsruhe and the company SHL and is coordinated by robotics software and solution provider Artiminds Robotics. The project results are to be used in the partners' offerings after the end of the project in September 2023.

Surfaces take center stage

The project focuses on the process steps of grinding, polishing and deburring in the areas of green mobility, green energy storage and green power generation. AI researcher Prof. Marco Huber from the IFF at the University of Stuttgart explains: "A cost-effective and flexible system for automated surface processing is required for the economical reconditioning of wind turbine rotor blades or electric motor gears. By using AI-based software solutions, it is possible to integrate object detection and measurement, force-controlled surface processing and downstream visual inspection in a single robot system."

In order to achieve the highest possible degree of autonomy and precision, a hybrid AI approach is being pursued that combines both knowledge-based and learning, data-based methods. "In this way, the robot should be able to anticipate deviations and surface conditions at runtime and adapt automatically.

This is achieved on the one hand through prior knowledge, which qualified workers can contribute, and on the other hand by means of sensor data, for example from force-torque sensors or vision sensors," says Dr. Darko Katic, technical contact for the RoboGrind project and Senior Team Leader Artificial Intelligence at Artiminds.