Numerical simulation and machine learning - Contribution of artificial intelligence to numerical modelling

Numerical simulation has become widely used in industry over the past two decades, and is now used in a wide range of engineering applications (mechanical, thermal, acoustic, hydrodynamic, etc.). It benefits from constant innovations stemming from academic research in a variety of fields (physical modeling, applied mathematics, computer science and algorithms, etc.): simulations can account for increasingly complex physical phenomena (such as multiphysical couplings, non-linear behavior, etc.) with growing precision and efficiency. Calculations help to optimize the design of many products and improve their reliability and durability. However, numerical simulations have a number of limitations that restrict their use in certain cases, particularly in terms of robustness, computational resource requirements (computing, storage, etc.) and energy consumption.

Alongside numerical simulation, machine learning techniques are developing, with highly interesting predictive capabilities: based on the growing availability of data (from test results, measurements, sensors, calculations, etc.), machine learning algorithms can be used to build numerical models to complement the models used for physics simulation.

This article, aimed primarily at young engineers and researchers in numerical simulation, offers a brief state-of-the-art on the coupling between numerical simulation and machine learning techniques, which is becoming one of the most interesting avenues for overcoming certain current computational limitations and taking numerical simulation to the next level.

Readers will find these references in the "Further reading" section of this article. An additional bibliography and links to websites provide useful resources for furthering knowledge on the subject.