Integrated platform for simulations and machine learning

Recent advances in computing and data storage have spurred a trend of machine-learning democratisation. Nowadays, data are hailed as the new transformative commodity, however, some researchers are sceptical whether machine learning alone is sufficient to tackle complex problems.  On the other hand, the coupling of machine-learning methods with physics-based simulation approaches has grown in popularity in recent years in both scientific and industrial settings. Nevertheless, the data-centric engineering paradigm lacks a common, easy-to-use solution framework to help researchers and industry practitioners couple machine-learning and simulations. In this work, we develop a cloud native integrated platform that couples first-principles, physics-based simulation approaches to machine learning pipelines, within a standardised, low-code solution framework. We showcase the ability of the platform to run fluid flow simulations, developed using OpenFoam, on a scalable cloud cluster programmatically and also through a user interface. We show how simulation metadata and outputs are automatically logged and used for downstream machine-learning pipelines. The solution platform is generic, easily extensible, and can handle different data-centric engineering algorithms like calibration, optimisation, and risk analytics, among others.