Monitoring and tracking machine learning models for fire sprinkler design in real-time using Simvue

Machine learning, governed by an optimisation objective, has evolved to handle complex modelling tasks. Managing and monitoring these optimisation processes is challenging, especially when dealing with large-scale simulations or data processing tasks, across domains and compute resources. To address this challenge, the UKAEA has developed Simvue - a generic and configurable real-time tracking and monitoring framework for any simulation or data processing tasks, which can be leveraged to monitor, and control machine learning based optimisation frameworks, providing a unified platform for managing the entire ML workflow. The platform is applicable to both current and legacy applications and is also language- and hardware-agnostic, capable of scaling on to any cloud based or bare-metal architecture. In this poster, we demonstrate how Simvue can be applied to track and monitor an optimisation procedure for fire sprinkler system design, which involves iterative exploration of the simulation design space, parameterised by different sprinkler head designs, droplet size distributions, and water surfactants. Simvue enables researchers to define the optimisation workflow, set up the necessary simulations, and monitor the progress of the optimisation process in real-time across different compute resources. By leveraging Simvue to manage the ML-based optimisation framework, users can streamline the process of identifying optimal fire sprinkler designs, potentially leading to substantial cost savings. The integration of Simvue with ML-based optimisation techniques demonstrates the versatility and effectiveness of the framework in managing complex data processing tasks across various domains with multi-fidelity resources. A recording of this session is available on YouTube: https://youtu.be/MjhhCI0GOU8