Combining the advantages from electric motor and the pneumatic actuator, the hybrid pneumatic-electric actuator is expected to be safe, low-cost, clean, high power to weight ratio, and to provide precise position control. In this thesis, the modeling and control of an improved hybrid pneumatic-electric actuator prototype is presented.
The actuator’s main components consist of a low-friction pneumatic cylinder, two on/off solenoid valves, and a small DC motor. The cylinder and motor are connected to a common output shaft using gears. The shaft rotates a single-link robot arm. Its position is measured by an incremental encoder. The prototype was improved by incorporating faster switching valves, flow controls, a faster valve drive circuit, a high resolution encoder rather than the existing linear potentiometer, more accurate pressure sensors and stronger gears.
A system dynamic model without the valve dynamic was developed identified and validated using open-loop experiments. The valve models for a discrete input and PWM input were then developed and validated separately. The use of bipolynomial function and artificial neural network fitting methods for modeling the valve mass flow rates were compared. The combined system model with valve dynamics was validated experimentally.
Two model-based nonlinear position controllers, using the backstepping and discrete-valued model predictive control (DVMPC) methods, were designed, simulated and extensively tested. Testing was done with the actuator operating using the cylinder alone, the motor alone and in hybrid mode using the cylinder and motor together. Operating in the hybrid mode reduced the root-mean-square error (RMSE) up to 80%.
A stability analysis for the backstepping control including the valve modeling error, friction model error, and electric motor torque modeling error was performed. Compensation terms were designed to improve the performance for the two controllers. Additional stability analyses were performed for backstepping controller with a feedback term and the DVMPC with motor control. A payload estimation algorithm was proposed and shown to enhance the robustness of the DVMPC operating in vertical configuration.
Simulations and experiments demonstrated that the model-based controllers performed well for both vertical and horizontal configurations. Regarding robustness to payload mismatch, if the payload was within the load capacity of the hybrid actuator, the model-based controllers were both insensitive to the payload variations in horizontal configuration. The backstepping controller was also robust to the payload variations in the vertical configuration.
In experiments, the backstepping control in hybrid actuation mode produced a RMSE of 0.0066 radian for a 2 Hz sine wave desired position trajectory with a 0.3 radian amplitude. With DVMPC, this value decreased to 0.0045 radian. These tracking errors were shown to be 30 to 50% less than those produced by a modified linear position plus velocity plus acceleration controller.