ExpertIA: Evolución del modelado y control de proceso industrial: modelos avanzados combinando conocimiento experto con técnicas IA en el diseño y desarrollo

Objective:

ExpertIA proposes to start from the classical models of resolution or FEM models (finite differences for solving differential equations that describe the behaviour of many physical systems) with AI/ML/DL techniques that allow: 1) to reduce the computation time required by numerical models; 2) to reduce the propagation of errors in numerical equations; 3) to reduce the propagation of errors in numerical equations; 3) reduce the amount of data needed to train data models; 4) reduce the amount of data needed to train data models. 5) improving the explainability and interpretability of black box models; 6) ensuring that data driven models generate solutions that comply with the laws of physics (e.g. conservation of mass or energy), which is difficult to guarantee due to the very nature of black box models. The nature of data driven models computational times and simulation errors of classical approaches based on the resolution of parametric physics laws of parametric physical laws, which will allow their use in real-time adaptive control systems, or to emulate situations never observed before. The performance of real-time control systems that would allow for the reduction of raw material, faults or defects, energy, waste, etc. ExpertIA aims to provide the industrial fabric of the Basque Country with the knowledge of this new paradigm, known as "Physics Aware AI", which will lead to an increase in their competitiveness, improve the impact of their competitiveness, improve the impact of the real application of AI/ML/DL in traditional businesses (RIS3 Sectors) and alleviate the problem associated with talent in the digital world. The challenges inherent to this general objective can be broken down into the following: - Coexistence of physical models and data models where different paradigms are included, with the aim of improving predictive simulation models, both for process control and failure prediction. - Preserving, systematising and transferring highly specialised expert knowledge to the AI techniques for analysis of real process data o Retain and integrate the expertise of the master operator who knows the fine tuning and fine tuning of process adjustment of process parameters due to their acquired experience. o Anticipate the loss due to generational change: overcome the difficulty of transmitting knowledge that comes from intuition and experience. - Build on existing solutions without discarding previous modelling, improve the shortcomings of the disjunct solution while respecting the disjoint solution while respecting the most practical approach. Balance between utility (interpretability improvement), cost (reusing previous investment without destroying its result) and accuracy.