TL;DR
This paper presents a novel recursive inverse method, combining a Kalman filter and recursive least squares, to accurately identify the inertia force exerted by a moving mass on a cantilever beam. The method demonstrates superior performance in noisy environments compared to traditional approaches and is validated through simulations and a field experiment.
What Problem Does This Paper Address?
Identifying the inertia force of a moving mass on a structure like a cantilever beam is crucial for structural health monitoring and control. Directly measuring this force is often impractical as it requires placing sensors in a way that could alter the structure's behavior or be damaged. Existing indirect methods, while useful, can struggle with noisy sensor data, limiting their real-world applicability. This paper aims to bridge this gap by developing a robust identification method that can accurately estimate the inertia force even under significant measurement noise.
Methods
The core of the proposed solution is a recursive inverse method that integrates two key components:
- Kalman Filter: This is used to process the measured vibration data (deflection) from the cantilever beam. It helps in estimating the system's state (modal displacements and accelerations) and generating a "residual innovation sequence." The Kalman filter is particularly adept at handling noisy measurements, providing a smoothed and more reliable estimate of the system's internal state.
- Recursive Least-Squares (RLS) Algorithm: This component takes the output from the Kalman filter (specifically, the innovation sequence) and uses it to estimate the unknown inertia force. The "recursive" nature means it updates its estimate incrementally as new data arrives, making it suitable for real-time applications.
The authors first establish a discretized state-space model of the cantilever beam and moving mass system based on the Euler-Bernoulli beam theory. This model transforms the complex differential equations governing the beam's vibration into a set of discrete-time equations that can be processed by the Kalman filter and RLS algorithm.
Datasets
The paper utilizes two types of datasets for evaluation:
- Simulated Data: Deflection data was generated using a finite element method (FEM) simulation. This allowed for controlled testing under both "none-noise" and "noise" conditions, with varying levels of simulated Gaussian noise.
- Field Experiment Data: A custom experimental setup was built to mimic the cantilever beam and moving mass scenario. Deflection data was collected using laser displacement sensors.
Key Insights and Ideas
- Robustness to Noise: The recursive inverse method shows a significant advantage over traditional least-squares methods when dealing with noisy sensor data. While both methods perform well in ideal, noise-free conditions, the Kalman filter's pre-processing is critical for the RLS to maintain accuracy in the presence of noise.
- Adaptive Estimation: The recursive nature of the algorithm allows it to adapt and refine its estimate of the inertia force as more data becomes available, making it suitable for dynamic scenarios.
- Model Simplification: The use of modal superposition and the Euler-Bernoulli beam model simplifies the complex physics into a manageable state-space representation, which is key to the algorithm's tractability.
- Parameter Sensitivity: The paper highlights the importance of the scale parameter $\gamma$ in the RLS algorithm, which balances noise elimination capability with fast adaptive capability. A careful selection of $\gamma$ is crucial for optimal performance.
Key Results
- Simulation Results:
- In none-noise simulations, the recursive inverse method achieved very low Relative Percentage Error (RPE), with values as low as 0.68% for a moving mass speed of 5 m/s.
- In noise simulations (with standard deviations of $10^{-7}$ and $10^{-6}$), the method demonstrated strong performance. For a speed of 10 m/s and noise $\sigma = 10^{-7}$, the RPE was 2.3375%. Even with higher noise ($\sigma = 10^{-6}$), the RPE was 11.7648%, while a modified IMII method failed significantly (RPE of 215.3404%).
- Field Experiment Results:
- The method successfully identified the inertia force in real-world experiments. For a speed of 10 m/s, the RPE was 1.145%.
- The initial part of the experimental identification was unstable due to impact waves from the mass entering and exiting the cantilever, but the algorithm converged to accurate results once these transient effects subsided.
Limitations and Future Work
- Initial Instability: The paper acknowledges that the initial phase of the identification process, both in simulations and experiments, can be unstable. This is attributed to the algorithm not yet converging and, in experiments, to impact waves from the mass.
- Impact Wave Mitigation: In field experiments, the authors suggest extending the fixed end of the cantilever to allow the moving mass to dwell longer, potentially helping to dissipate impact waves and improve initial stability.
- Assumption of Stable Velocity: The model assumes a stable velocity for the moving mass. If the cantilever is much longer or the contact surface is not smooth, the mass might lose energy, violating this assumption and requiring model modifications.
- Sensor Selection: The accuracy of the method is highly dependent on the quality of the deflection sensors. High resolution, stability, and low measurement noise are critical.
- Future Work: The authors plan to focus on improving the experiment setup and further refining the identification method itself.
Summary updated