Executive Summary : | Discrete power electronic devices, such as MOSFETs/IGBTs, are crucial in power management circuits for safety-critical applications like automotive, locomotive, aerospace, and power grids. Estimating their Remaining Useful Lifetime (RUL) is essential for maintaining safety. Data-driven approaches using Neural Networks (NNs) have gained popularity, but they can sometimes yield non-realistic RUL estimates, leading to misleading predictions. To address this issue, a project proposes a Physics Informed Neural Networks (PINN) approach for RUL estimation in power electronic converters. This approach incorporates physical rules into the loss function as regularization terms, providing more accurate and reliable RUL estimates.
Power electronic converters (PEC) are used in nonlinear environments, and existing control mechanisms are sensitive to nonlinearities, parametric variations, system disturbances, and various load conditions. Tuning PI gains is a complex task that requires complete knowledge of the system and depends on system parameters. With the changing landscape of the power sector, more adaptive control mechanisms are needed for better control and optimization. DL control provides a solution for these shortcomings, offering advanced optimization techniques and the highly adaptive nature of neural networks. DL algorithms can also mitigate cyber-attacks and secure information, which is crucial in today's communication-dominated world. |