Online Tuning of PID controller using Black Box Multi-Objective Optimization and Reinforcement Learning

Abstract A PID Controller is the most widely used controller due to its ease and convenience of use. Manual tuning of a PID Controller is a time-consuming task. Hence, employing intelligent algorithms is necessary. The Cummins engine controller has a complex structure. To fine-tune it, a substantial amount of time is required. To reduce this time requirement, a black box approach was selected for online tuning. This would not only reduce the required time, but also reduce the efforts. Black box optimization would mean the engineers have to spend less time trying to understand the controller structure. With this aim in mind, a PID system simulation was set up in MATLAB. A function would randomize a system, resulting in a true black box to tune. This removed any bias the authors might have. The algorithm has shown promising results, with tuned controller gains in just over 20 iterations on average. This could then be extended to not only Cummins controllers, but other industrial controllers as well.