Published February 27, 2026 | Version v1
Journal Open

AUTONOMOUS ACTIVE VIBRATION CONTROL: A DEEP REINFORCEMENT LEARNING APPROACH FOR DYNAMIC PID TUNING IN NON-LINEAR SYSTEMS

Authors/Creators

Description

PID controllers are ubiquitous in industrial control applications due to their simplicity. However,
static gain control in PID controllers is restricted in dealing with non-linear dynamics, leading to
overshooting, oscillations, or inefficiencies. This paper proposes an active vibration control system
that is self-sustaining in nature. The control signals are adapted using the Deep Reinforcement
Learning (DRL) algorithm. A Soft Actor Critic algorithm was designed to self-adjust the
proportional, integral, and derivative gains every 10 ms based on the changes in the system
dynamics. The system was tested in an in-house Open AI-based customized environment that
considers the electromechanical properties of the DC motor. The test output reveals that the control
system removes overshooting, brings down the settling time by 40%, and optimizes energy
efficiency by 17%. Moreover, the adaptive gain process discovers that the control algorithm
independently learns the variable-structure control logic. The control logic hardens or softens the
system as required. The experiment validates the efficacy of the novel control logic based on the
superior power of DRL algorithms as contrasted with the Ziegler-Nichols PID logic in self-healing
non-linear industrial dynamics.

Files

AUTONOMOUS ACTIVE VIBRATION CONTROL A DEEP REINFORCEMENT LEARNING APPROACH FOR DYNAMIC PID TUNING IN NON-LINEAR SYSTEMS.pdf

Files (378.7 kB)