Simultaneous input-state estimation with direct feedthrough based on a unifying MMSE framework with experimental validation

Xi Liu,Yang Wang,Erik I. Verriest

DOI: https://doi.org/10.1016/j.ymssp.2020.107083

IF: 8.4

2021-01-01

Mechanical Systems and Signal Processing

Abstract:<p>This paper addresses the simultaneous input-state estimation problem for discrete-time linear stochastic systems with unknown input. A unifying minimum-mean-square-error (MMSE) estimation framework is presented for a comprehensive characterization and direct comparison among a few popular input estimation approaches. In this work, the dynamical model of the input is assumed to be unknown and the input is treated as a random variable at each time step. To account for the unknown input dynamics, we propose an input estimator that adopts a white Gaussian input model with a finite covariance, short-named as the FIC (finite input covariance) estimator. Theoretical formulation of the FIC estimator is first compared with two other estimators, one using a Gaussian random walk input model combined with augmented Kalman filter (AKF) and another one using a deterministic input model with weighted least squares (WLS) estimation. Based on the unifying MMSE framework presented in this paper, it is proved that when the input covariance of the FIC estimator approaches infinity and the feedthrough matrix has full-column rank, the estimator is equivalent to the well-known WLS estimator.</p><p>The FIC estimator is validated and compared with the AKF and WLS estimator using simulated measurements from 2-story shear structure and experimental measurements from a full-scale concrete frame. The 2-story shear structure is excited by two different types of input and the corresponding acceleration responses are used to compare estimator performance. With only acceleration measurements, the FIC estimator eliminates a low-frequency drift error in the estimated input and states with a tight estimation confidence interval. Detailed discussion on the effect of estimator covariances on input estimation is also provided. <em>A priori</em> knowledge of the statistical property of the unknown input can provide insights in the tuning of FIC estimator covariances. In addition, field acceleration measurements from a full-scale concrete frame under shaker excitation are used to compare the estimation results and validate the proposed FIC estimator. The FIC estimator is shown to provide better estimates with tighter estimation confidence interval in comparison to the AKF and WLS estimators.</p>

engineering, mechanical

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to simultaneously estimate the state and input in a discrete - time linear stochastic system when the input is unknown. Specifically, the paper proposes a method based on the minimum mean - square error (MMSE) framework for comprehensively characterizing and directly comparing several popular input estimation methods. In this work, the input dynamic model is considered unknown, and the input is regarded as a random variable at each time step.

### Main Contribution Points:
1. **Unified MMSE Framework**: The paper proposes a unified MMSE framework for simultaneously estimating the state and input. This framework can comprehensively characterize and directly compare several popular input estimation methods.
2. **Finite Input Covariance (FIC) Estimator**: To deal with the unknown input dynamics, the paper proposes an input estimator that adopts a white Gaussian input model and finite covariance, called the FIC estimator.
3. **Theoretical Analysis**: Through the unified MMSE framework, it is proved that when the input covariance of the FIC estimator approaches infinity and the feed - forward matrix \( \mathbf{D} \) has full column rank, the FIC estimator is equivalent to the known weighted least - squares (WLS) estimator.
4. **Experimental Verification**: The FIC estimator is verified by simulation and experimental data and compared with the augmented Kalman filter (AKF) and the WLS estimator. The results show that the FIC estimator performs better in eliminating low - frequency drift errors and providing a tighter estimation confidence interval.

### Mathematical Formulas:
- **State Update Equation**:
\[
\mathbf{x}_{k + 1}=\mathbf{A}\mathbf{x}_k+\mathbf{B}\mathbf{u}_k+\mathbf{w}_k
\]
- **Measurement Equation**:
\[
\mathbf{y}_k=\mathbf{C}\mathbf{x}_k+\mathbf{D}\mathbf{u}_k+\mathbf{v}_k
\]
- **Input Estimation Gain**:
\[
\mathbf{L}_{u_k}=\mathbf{R}_u\mathbf{D}^T\left(\mathbf{R}_{\tilde{y}_k}+\mathbf{D}\mathbf{R}_u\mathbf{D}^T\right)^{-1}
\]
- **Input Estimation Error Covariance**:
\[
\mathbf{R}_{u_k}=\mathbf{R}_u-\mathbf{R}_u\mathbf{D}^T\left(\mathbf{R}_{\tilde{y}_k}+\mathbf{D}\mathbf{R}_u\mathbf{D}^T\right)^{-1}\mathbf{D}\mathbf{R}_u
\]
- **State Estimation Gain**:
\[
\mathbf{L}_{x_k}=\mathbf{R}_{x_k|k - 1}\mathbf{C}^T\left(\mathbf{R}_{\tilde{y}_k}+\mathbf{D}\mathbf{R}_u\mathbf{D}^T\right)^{-1}
\]

### Experimental Verification:
The paper uses the simulation data of a two - layer shear structure and the experimental data of a full - scale concrete frame to verify the performance of the FIC estimator. The results show that the FIC estimator is superior to the AKF and the WLS estimator in eliminating low - frequency drift errors and providing a tighter estimation confidence interval.

### Conclusion:
The paper proposes a FIC estimator based on a unified MMSE framework for simultaneously estimating the state and input. This estimator performs well in dealing with unknown input dynamics, especially in the case where only acceleration measurement data are available, and can effectively eliminate low - frequency drift errors.