Perpustakaan Digital ITB

The accurate simulation of fluid flow in porous media plays a crucial role for various engineering and scientific applications in petroleum engineering. Traditional numerical methods for solving the governing flow equations in porous media are often computationally demanding and time-consuming, yet provide discrete solutions. In recent years, Physics Informed Neural Networks (PINNs) have emerged as a promising alternative to address these challenges. The flow equations in porous media describe the movement of fluids through a porous medium, considering factors such as fluid velocity, pressure, and porosity. PINNs offer a unique approach by combining the power of artificial neural networks (ANN) with the physical laws governing fluid flow. The key idea behind PINNs is to incorporate known physical principles into the neural network architecture. This is achieved by embedding the governing equations of fluid flow as additional loss terms during the training process. By enforcing these physics constraints, the neural network learns to approximate the solution to the flow equations, resulting in accurate predictions. The flexibility of PINNs extends beyond their ability to approximate flow solutions. They can effectively handle situations with limited data, making them valuable when experimental or field data is scarce. PINNs also provide a framework for incorporating prior knowledge and can handle uncertain or noisy input data. These features make PINNs a versatile tool for studying flow behavior in porous media, with potential applications in petroleum engineering.