Data-Driven Feedforward Force Control of a Single-Acting Pneumatic Cylinder with a Nonlinear Hysteresis Characteristic

Xiaofeng Wu, Hongliang Hua, S. K. Feng, Yanli Zhao, Yuhong Yang, Zhenqiang LiaoActuators 2025 · DOI

TL;DR

This paper tackles the challenge of precise force control in single-acting pneumatic cylinders, which are plagued by nonlinear hysteresis. They introduce a data-driven neural network approach to accurately model and control the actuating force, achieving excellent tracking performance with minimal error.

What Problem Does This Paper Address?

Pneumatic systems are widely used in automation for tasks like robotic grasping and industrial polishing due to their advantages in size, safety, and cost. However, achieving accurate force control is difficult, especially with single-acting pneumatic cylinders. These cylinders, while simpler than their double-acting counterparts (using a spring for retraction), suffer from significant nonlinear hysteresis. This hysteresis, caused by factors like friction and the spring's own nonlinear characteristics, makes it hard to predict and control the exact force output. Existing feedforward control methods struggle to compensate for these inherent nonlinearities, leading to performance degradation. This paper aims to develop a robust feedforward force control strategy that can accurately model and overcome the hysteresis in single-acting pneumatic cylinders.

Methods

The core of this work is a data-driven neural network modeling approach.

  • Neural Network Modeling: They use a neural network to learn the complex relationship between control inputs (like actuating pressure, piston displacement, velocity, and motion state) and the resulting actuating force. This is a departure from traditional physics-based models, which are difficult to develop accurately for such nonlinear systems.
  • Input Layer Analysis: A key contribution is the systematic analysis of different input variable combinations for the neural network. This helps identify the most crucial factors influencing the actuating force, leading to a more efficient and accurate model.
  • Hidden Neuron Optimization: The paper investigates the trade-off between the number of hidden neurons in the neural network and its impact on both modeling accuracy and real-time execution speed. This is crucial for implementing the control on embedded microcontrollers.
  • Instruction Smoothing: To further improve step force control, they introduce an "instruction smoothing module." This module gradually adjusts the control signal, preventing abrupt changes that can lead to overshoot and steady-state errors.

Datasets

The study utilizes data collected from a custom-built single-acting pneumatic system. This system includes:

  • A single-acting pneumatic cylinder.
  • A micro linear potentiometer for measuring displacement.
  • A force sensor for measuring actuating force.
  • A micro-controller (Arduino UNO) for data acquisition and control.

The data was generated by performing loading and unloading experiments under various conditions to capture the hysteresis and nonlinearities. This experimental setup and the collected data serve as the benchmark for evaluating their proposed method.

Key Insights and Ideas

  • Data-Driven Dominance: The paper strongly suggests that for complex nonlinear systems like pneumatic cylinders with hysteresis, a data-driven neural network approach is more effective than traditional analytical modeling.
  • Input Variable Importance: Not all potential input variables are equally important. Identifying the critical ones (actuating pressure, displacement, velocity, and motion state) significantly improves model accuracy and efficiency.
  • Real-time Constraints: The research highlights the critical balance between model complexity (accuracy) and computational cost (real-time execution) for embedded control systems. They demonstrate how to optimize this trade-off.
  • Instruction Smoothing for Stability: The instruction smoothing technique is a practical way to mitigate overshoot and steady-state errors in step force control, making the system more predictable and stable.

Key Results

  • High Modeling Accuracy: The neural network model achieved excellent modeling accuracy, with good-fit parameters (R) reaching up to 0.99962 for predicting target voltage instruction and 0.99995 for predicting actuating force.
  • Fast Execution: The optimized neural network models could be executed on a microcontroller within 2 ms, meeting real-time requirements.
  • Precise Force Tracking: The actuating force achieved ideal tracking of the target via the feedforward approach.
  • During loading and unloading, the control error amplitude was less than 0.5 N.
  • The overall RMS value of the control error was approximately 1 N.
  • Step Force Control Improvement: Instruction smoothing significantly reduced percent overshoot (up to 88.1% reduction for 50 N force) and steady-state error (up to 80.3% reduction for 50 N force) as the smoothing steps increased.

Limitations and Future Work

  • Unmodeled Dynamics: The paper acknowledges that some dynamic effects, particularly the inertia force of the piston rod during switching, were not fully modeled and contributed to control errors.
  • Experimental Setup Specificity: While the method is data-driven, its direct transferability to significantly different pneumatic systems might require re-calibration and re-training.
  • Future Work Suggestions: The authors hint at exploring hybrid actuation systems that combine the strengths of different actuator types to overcome individual limitations and enhance overall performance across a broader range of applications. Further research could also focus on more advanced control strategies to address the unmodeled dynamic effects.

Summary updated

Abstract

Pneumatic force control has a broad application background in the automation field, such as in industrial polishing, robotic grasping, and humanoid robots. Nonlinear hysteresis characteristics are one of the major factors that affect the feedforward force control performance of a pneumatic system. The primary motivation of this paper is to develop an accurate feedforward actuating force control method for a single-acting pneumatic cylinder with a nonlinear hysteresis characteristic. A data-driven neural network modeling method is presented to achieve accurate actuating force modeling. The modeling accuracy of the neural network model under different configurations of the input layer is quantitatively analyzed to determine the essential modeling variables. The real-time execution speed of neural network models with different numbers of hidden neurons is evaluated to achieve a balance between the modeling accuracy and the real-time computing speed of the neural network model. Then, a single-acting pneumatic system is fabricated to experimentally verify the effectiveness of the proposed modeling and control method. The experimental results reveal that the actuating force can achieve ideal tracking of the target. In both the loading and the unloading process, the amplitude of the control error is less than 0.5 N. The overall RMS value of the control error is about 1 N. An instruction smoothing operation could reduce the percentage overshoot and steady-state error of the feedforward step actuating force control.
BibTeX
@inproceedings{wu2025data,
  author = {Xiaofeng Wu and Hongliang Hua and S. K. Feng and Yanli Zhao and Yuhong Yang and Zhenqiang Liao},
  title = {{Data-Driven Feedforward Force Control of a Single-Acting Pneumatic Cylinder with a Nonlinear Hysteresis Characteristic}},
  booktitle = {Actuators},
  year = {2025},
  doi = {10.3390/act14040162},
}
Figure