← back

Price

—

Travel

—

Access

—

Social

—

Modern engineering systems increasingly involve complex interactions across multiple physical scales, domains and operating conditions, making accurate and computationally efficient modelling a major challenge. Conventional physics-based approaches, including finite element and multiphysics methods, can become computationally expensive and difficult to apply to highly nonlinear, heterogeneous and multiphase systems. This seminar presents a Physics-Informed Machine Learning (PIML) framework that combines governing physical principles with data-driven learning to address these limitations. The approach is demonstrated through multiscale and multiphase modelling of complex engineering processes, with an application to predicting coupled transport phenomena and microscale structural evolution in thermal processing of deformable porous materials (DPM). The developed PIML models integrate physical laws, experimental observations and computational modelling to predict dynamically changing complex properties and morphological variations of DPM across different spatial and temporal scales, while significantly reducing computational cost and dependence on extensive training data. Beyond the specific application to complex thermal processes, the seminar will discuss the broader potential of PIML as a general methodology for complex engineering systems; with particular attention to the connections between physics-informed learning, digital twins and real-time monitoring. These concepts offer promising opportunities for developing intelligent engineering systems that can combine physical models and sensor data to continuously predict, diagnose and optimise system behaviour. The seminar will conclude by discussing emerging challenges and opportunities in advanced PIML, including generalisation under changing operating conditions, uncertainty quantification, and the transition towards real-time engineering decision-making.

• Talks