As a trained control engineer, I understand that when the plant model is well understood (or even with many uncertainty in the plant dynamic model), designing a control system using traditional methods such as PID control, H-infinity control, or model predictive control is simple (MPC). However, as machine learning and artificial intelligence have advanced, neural network-based approaches for plant control systems have grown in popularity.

In this regard, I'd like to solicit the thoughts and experiences of other control engineers on the occasions in which it is appropriate to pick a neural network-based strategy over conventional control techniques. What benefits and drawbacks come with implementing plant control systems using neural networks? Are there any particular uses or circumstances where a neural network-based approach would be more effective than conventional control techniques? How does one choose the best strategy for a specific plant control issue? Any ideas or information would be greatly appreciated.