AI for digital twins and for systems for monitoring and control is typically trained on data from system sensors. But such sensor data is usually limited and noisy and insufficient to make the AI understand the physics of the system. This will lead to erratic or improper behavior by the AI. We solve this problem by incorporating knowledge about the system's physics in the training process to make the AI comply with these physics.
We help developers of digital twins and systems for monitoring and control to train AI-models from the limited and noisy data coming from the sensors in the system.
We offer consultancy around the use of physics AI, including an assessment of the potential of physics AI, training and instruction.
We audit AI-models that are used in digital twins or in systems for monitoring and control to assess whether proper approaches have been used to guarantee compliance with the underlying physics and whether the model is fit for purpose.
4DCool allows optimal temperature control in data centers by providing a predictive 3D temperature model based on data from sensors in combination with a model for the physics of the indoor climate.
We trained an AI surrogate model for model-in-the-loop control of a large and complex conveyor belt. This model-in-the-loop control is more robust and reliable than the original rules-based control.