Use of Neural Networks to Solve Transient Heat Transfer in Film Cooling Experiments

The aim of this project was to investigate the application of Artificial Neural Networks (ANNs), and particularly Physics-Informed Neural Networks (PINNs) for modelling transient heat transfer in film cooling applications. The primary objective of the research was to evaluate the performance of the PINN framework in comparison to a traditional Crank-Nicolson MATLAB solver and assess its potential for accurately capturing the complex behaviour of film cooling heat transfer. To achieve this objective, the project employed a multi-stage methodology, which involved a comprehensive review of the literature on neural network-based solutions for partial differential equations, the development of a custom Python code to implement the PINN framework, and the verification of the achieved result via comparison with an established finite differencing method. The results of the investigation demonstrated that the PINN framework is an effective tool for modelling the general behaviour of transient heat transfer in film cooling contexts, despite certain inaccuracies observed at the internal boundary surface of the model. Despite these discrepancies, the PINN model achieved a low RMS error of 0.003729, and a maximum percentage error of 1.03%, indicating its potential for approximating complex heat transfer processes. The limitations of the PINN framework were also identified, including a potential suboptimal network architecture and and inherent limitations of PINNs when applied to intricate boundary conditions. In conclusion, this investigation has demonstrated the promise and limitations of using PINNs for film cooling applications, further advancing the state-of-theart in data-driven modelling and simulation techniques.