In the engineering process, simulation software such as iPhysics acts as a test bed for the design of the artificial intelligence (AI) to be developed. With the help of the digital twin, a virtual training center is created to prepare the algorithms for their task: e.g. monitoring the productivity of machines and systems. In principle, simulation is currently used by engineers as a tool to gain knowledge that can be transferred to reality in accordance with the VDI3633-1 guideline. In continuation of this approach, the engineer exposes the AI to its learning task in the simulated environment before operating the real machine.
Hardware-in-the-loop simulation and AI
The training environment, based on a hardware-in-the-loop (HIL) simulation, consists of the machine model, a real or virtual controller and the AI system. This allows the interaction of all three components to be tested. This is important to enable the artificial intelligence to test different algorithms and experience possible errors for itself with the aim of avoiding them in real operation. The learning objectives can also be defined and checked by the engineer without risk.
Ghost robots
In many industries that work with robot-controlled processes, various robots are located in a confined space and must function safely with and alongside each other. The more the self-programming machine becomes a reality, the more thoroughly it must be checked for collisions or lack of reachability. A real-time 3D model with collision detection can be used for this purpose.
A ghost robot runs parallel to the virtual machine and is one predefined time sequence ahead of the real machine. This additional tool can also be used during the planning and development stages to virtually run through various scenarios to see how the robots in question can best run together and side by side. The simulation of robot behavior also helps companies to comply with the Machinery Directive. This means that the speed of movements can be adapted to the environment, e.g. in the vicinity of scaffolding or safety fences, at an early stage in the development process.
Machine learning
The use of 3D simulation as a training environment for machine learning leads the engineer directly to the next evolutionary step: the fusion of both technologies in real operation. This is because the degrees of freedom of machine learning – and therefore the bandwidth and computing effort required later in operation – can be significantly reduced by using a 3D model of the machines and robots. This will be important if all machines in the production facilities are to be upgraded with AI. The cost-effectiveness of the computing power provided rises and falls with the number and complexity of the degrees of freedom to be learned by algorithms. The 3D real-time simulation with collision calculation on the machine can help to significantly reduce the degrees of freedom to be recorded and thus focus the AI on the essential things.