Abstract
In the low-speed operation of the DC motor dual closed-loop speed control system, the PI controller of the speed loop is prone to experiencing integral saturation due to the persistent presence of large errors. This leads to increased overshoot and prolonged adjustment time, seriously restricting system performance. To address this issue, this paper proposes an improved speed loop controller. This controller adopts a multi-level integral separation strategy, managing the integral action in stages based on the error magnitude, and innovatively utilizes the genetic algorithm (GA) to automatically optimize the multi-level separation thresholds offline. Experiments conducted on the Matlab/Simulink platform for a 583.3 rpm step start-up condition demonstrate that the proposed GA-optimized method effectively suppresses integral saturation, achieving near-zero overshoot and shortening the adjustment time by approximately 18% compared to traditional PI control. This method provides an automated optimization solution for high-performance low-speed control of DC motors.
Introduction
In industrial DC motor control, the Proportional-Integral (PI) controller is widely utilized due to its simplicity and effectiveness. However, it suffers from a significant drawback known as integral saturation or "windup." During startup or low-speed operation, large initial speed errors cause the integral term to accumulate excessively. By the time the motor reaches its target speed, this accumulated error forces the system to overshoot, leading to long settling times and potential mechanical stress. While standard integral separation techniques—which disable the integral term during large errors—offer some relief, manually tuning the thresholds for these transitions is often inefficient and sub-optimal.
System Architecture
The control system is designed as a dual-closed-loop structure consisting of an outer speed loop (ASR) and an inner current loop (ACR). The speed loop generates the current reference, while the current loop regulates the motor's armature voltage.
The mathematical model of the separately excited DC motor is implemented in MATLAB/Simulink, incorporating parameters such as rated current, torque, and electromagnetic time constants to ensure a high-fidelity simulation environment.
Proposed Methodology
Multi-Level Integral Separation
To provide a smoother transition than binary ON/OFF separation, this research implements a three-stage multi-level integral separation strategy. The integral action coefficient ($\gamma$) is adjusted dynamically based on the absolute error ($|e|$):
- Large Error ($|e| > \epsilon_1$): $\gamma = 0$. The integral term is completely removed to prevent windup during the initial acceleration phase.
- Medium Error ($\epsilon_2 < |e| \le \epsilon_1$): $0 < \gamma < 1$. A weakened integral action is applied to begin closing the steady-state gap without causing excessive accumulation.
- Small Error ($|e| \le \epsilon_2$): $\gamma = 1$. Full integral action is restored to ensure high precision and eliminate steady-state error.
Optimization via Genetic Algorithm
Determining the optimal thresholds ($\epsilon_1, \epsilon_2$) is critical for performance. This paper employs a Genetic Algorithm (GA) to perform an offline global search for these parameters.
The GA evaluates potential solutions using a fitness function that balances overshoot and settling time: $$Fitness = \omega_1 \cdot \text{Overshoot} + \omega_2 \cdot \text{Settling Time}$$ By evolving the thresholds over multiple generations, the algorithm identifies the optimal boundaries that minimize both metrics simultaneously.
Results and Discussion
The proposed GA-optimized multi-level separation method was benchmarked against three standard control schemes during a 583.3 rpm step start-up simulation. The results are summarized in the table below:
| Scheme | Control Strategy | Overshoot ($\sigma\%$) | Settling Time ($t_s$) |
|---|---|---|---|
| A | Traditional PI | 10.74% | 0.340s |
| B | Single-stage Separation | 6.28% | 0.337s |
| C | Manual Multi-level | 0.63% | 0.312s |
| D | GA-Optimized Multi-level | $\approx 0\%$ | 0.280s |
The data indicates that the GA-optimized approach (Scheme D) virtually eliminates overshoot and reduces the adjustment time by approximately 18% compared to the traditional PI controller (Scheme A). Furthermore, it outperforms manually tuned multi-level separation, proving that automated optimization can discover more effective control boundaries than human intuition.
Conclusion
This research demonstrates that combining multi-level integral separation with Genetic Algorithm optimization significantly enhances the dynamic performance of DC motor speed control. The proposed method effectively mitigates integral saturation, providing a near-perfect response with minimal overshoot and faster settling times. While the current optimization is performed offline and relies on an accurate system model, future work will explore online adaptive algorithms to maintain performance under varying load conditions and parameter drift.
Summary updated