Simulation and Experimental Investigation of Structural Dynamic Frequency Characteristics Control

Xingwu Zhang, Xuefeng Chen, Shangqin You, Zhengjia He, Bing LiSensors 2012 · DOI

TL;DR

This paper introduces a novel Dynamic Frequency Characteristics Active Control (DFCAC) method that uses neural networks to not only suppress vibrations but also actively alter the dynamic frequency response of mechanical systems. This approach offers greater flexibility and adaptability compared to traditional vibration control methods, enabling tailored performance for various operating conditions.

What Problem Does This Paper Address?

Mechanical equipment like cars, airplanes, and machine tools typically operate with fixed dynamic frequency characteristics. However, for improved performance, stability, or adaptation to different working conditions, these characteristics may need to be modified. Existing active control methods primarily focus on reducing vibration amplitudes in the time or frequency domains, but they lack the ability to arbitrarily change the inherent dynamic frequency properties of a system. This paper addresses this gap by proposing a method that can actively control and modify these characteristics.

Methods

The core of the proposed DFCAC method is a neural network system comprising two main components: a Neural Network Identifier (NNI) and a Neural Network Controller (NNC).

  • Neural Network Identifier (NNI): This part identifies the system model of the controlled object. It takes the actuating parameters (amplitude, frequency, phase) and the system's vibration response (converted to the frequency domain via FFT) as inputs to learn the system's frequency-domain model. The NNI uses a linear transfer function for its output layer and employs the Widrow-Hoff learning rule for weight correction.
  • Neural Network Controller (NNC): This component is responsible for achieving the control target. It's a nonlinear network with three layers (input, hidden, output). The NNC uses the identified system parameters and the desired frequency response to generate new actuating parameters. It utilizes a gradient descent procedure, enhanced with a momentum factor, to minimize a global frequency error.
  • Error Criterion: A combined error criterion is used, incorporating both a global frequency error (J) and a frequency node error (E). This dual approach aims to improve the flexibility, adaptability, and anti-interference capabilities of the DFCAC method, preventing issues like slow convergence or unacceptable control results that can arise from using only one type of error.

Datasets

Not discussed in the paper. The paper focuses on simulation and experimental validation using a controlled flat plate.

Key Insights and Ideas

  • Beyond Vibration Suppression: The most significant insight is the shift from solely reducing vibration to actively controlling and modifying the dynamic frequency characteristics of a system. This opens up possibilities for tailoring equipment performance for specific needs.
  • Frequency-Domain Advantage: Operating entirely in the frequency domain, with FFT used only once per control cycle and no IFFT, significantly simplifies the control process and reduces computational time compared to time-domain methods.
  • Neural Network Synergy: The combination of an NNI for system identification and an NNC for control provides a robust and adaptive framework capable of handling complex, nonlinear dynamics.
  • Combined Error Criterion: Using both global and frequency node errors is crucial for achieving reliable and flexible control, avoiding the pitfalls of relying on a single error metric. This enhances robustness against noise and improves adaptability.

Key Results

The paper presents simulation and experimental results for a flat plate under various conditions.

  • Simulation (Working Condition 1): The DFCAC method successfully reduced the amplitude at 21 Hz by half, from 0.0002134 mm to 0.0001062 mm, with a smooth iterative process.
  • Simulation (Working Condition 2): The method achieved the objective of increasing vibration response to four times the original at 20 Hz and two times the original at 50 Hz, demonstrating smooth control with a high convergence rate.
  • Experiment (Working Condition 1): In the presence of motor interference (DC motor), the DFCAC method successfully increased vibration at 80 Hz from 0.01 mm to 0.02 mm, achieving the objective in a noisy environment. The error curve indicated a smooth control process, verifying anti-interference ability.
  • Experiment (Working Condition 2): The DFCAC method successfully altered vibration from 0.007 mm to 0.004 mm at 80 Hz and from 0.002 mm to 0.008 mm at 120 Hz, achieving the control objective with a smooth convergence process.
  • Experiment (Working Condition 3): The control process was stable and smooth, reducing vibration at 80 Hz from 0.012 mm to 0.006 mm.

Limitations and Future Work

  • Limitations:
    • The paper mentions that the peak at 0 Hz in one experimental condition was induced by DC interference from the hardware device, which could be eliminated by software or hardware measures. This highlights a potential sensitivity to certain types of noise or interference.
  • While the method is shown to be robust, the specific range of background noise it can effectively handle is not exhaustively detailed.
  • Future Work: The paper implicitly suggests that the DFCAC method's flexibility and adaptability make it suitable for a wide range of mechanical devices. Further research could explore its application to more complex systems, investigate optimal neural network architectures for different scenarios, and delve deeper into the theoretical bounds of its anti-interference capabilities. The ability to "arbitrarily change the dynamic characteristics" also opens avenues for exploring adaptive control strategies that dynamically adjust these characteristics based on real-time operational demands.

Summary updated

BibTeX
@article{zhang2012simulation,
  author = {Xingwu Zhang and Xuefeng Chen and Shangqin You and Zhengjia He and Bing Li},
  title = {{Simulation and Experimental Investigation of Structural Dynamic Frequency Characteristics Control}},
  journal = {Sensors},
  year = {2012},
  doi = {10.3390/s120404986},
}
Figure