Characterization of Kalman filter residuals in the presence ofmismodeling

The mean and covariance of a Kalman filter residual are computed for specific cases in which the Kalman filter model differs from a linear model that accurately represents the true system (the truth model). Multiple model adaptive estimation (MMAE) uses a bank of Kalman filters, each with a different internal model, and a hypothesis testing algorithm that uses the residuals from this bank of Kalman filters to estimate the true system model. At most, only one Kalman filter model will exactly match the truth model and will produce a residual whose mean and standard deviation have already been analyzed. All of the other filters use internal models that mismodel the true system. We compute the effects of a mismodeled input matrix, output matrix, and state transition matrix on these residuals. The computed mean and covariance are compared with simulation results of flight control failures that correspond to mismodeled input matrices and output matrices     

基于混合卡尔曼滤波器组故障传感器定位方法

为了在发动机性能蜕化与传感器故障并存的情况下实现故障传感器的定位与部件蜕化情况的估计,并实现故障诊断基准数据的修正,构建了1种包含了机载模型与线性卡尔曼滤波器的组合结构混合卡尔曼滤波器组。该卡尔曼滤波器组能够在之前所描述的故障/蜕化耦合情况下定位故障传感器,并得到较为准确的部件蜕化估计结果。为了验证了混合卡尔曼滤波器组的有效性,进行了相关仿真。仿真结果表明:混合卡尔曼滤波器组能够在发动机动态过程中遭遇传感器故障与部件蜕化并存的情况下完成故障定位与蜕化估计。     

Control of a large space structure using MMAE/MMAC techniques

The application of moving-bank multiple model adaptive estimation and control (MMAE/MMAC) algorithms to an actual spade structure (Space Integrated Controls Experiment (SPICE)) being examined at Phillips Laboratory at Kirtland AFB, NM, is presented. The structure consists of a large platform and a smaller platform connected by three legs in a tripod fashion. Kalman filtering and LQG (linear system, quadratic cost, Gaussian noise) control techniques are utilized as the primary design tools for the components of the MMAE/MMAC. Implementing a bank of filters or controllers increases the robustness of the algorithms when uncertainties exist in the system model, whereas the moving bank is utilized to reduce the computational load. Several reduced-order models are developed from the truth model using modal analysis and modal cost analysis. The MMAE/MMAC design with a substantially reduced-order filter model provides an excellent method to estimate a wide range of parameter variations and to quell oscillations in the structure.     

Reduced state estimators for consistent tracking of maneuvering targets

Linear Kalman filters, using fewer states than required to completely specify target maneuvers, are commonly used to track maneuvering targets. Such reduced state Kalman filters have also been used as component filters of interacting multiple model (IMM) estimators. These reduced state Kalman filters rely on white plant noise to compensate for not knowing the maneuver - they are not necessarily optimal reduced state estimators nor are they necessarily consistent. To be consistent, the state estimation and innovation covariances must include the actual errors during a maneuver. Blair and Bar-Shalom have shown an example where a linear Kalman filter used as an inconsistent reduced state estimator paradoxically yields worse errors with multisensor tracking than with single sensor tracking. We provide examples showing multiple facets of Kalman filter and IMM inconsistency when tracking maneuvering targets with single and multiple sensors. An optimal reduced state estimator derived in previous work resolves the consistency issues of linear Kalman filters and IMM estimators.     

Kalman filtering with state equality constraints

Kalman filters are commonly used to estimate the states of a dynamic system. However, in the application of Kalman filters there is often known model or signal information that is either ignored or dealt with heuristically. For instance, constraints on state values (which may be based on physical considerations) are often neglected because they do not fit easily into the structure of the Kalman filter. A rigorous analytic method of incorporating state equality constraints in the Kalman filter is developed. The constraints may be time varying. At each time step the unconstrained Kalman filter solution is projected onto the state constraint surface. This significantly improves the prediction accuracy of the filter. The use of this algorithm is demonstrated on a simple nonlinear vehicle tracking problem     

An MMAE failure detection system for the F-16

A multiple model adaptive estimation (MMAE) algorithm is implemented with the fully nonlinear six-degree-of-motion, Simulation Rapid-Prototyping facility (SRF) VISTA F-16 software simulation tool. The algorithm is composed of a bank of Kalman filters modeled to match particular hypotheses of the real world. Each presumes a single failure in one of the flight-critical actuators, or sensors, and one presumes no failure. For dual failures, a hierarchical structure is used to keep the number of on-line filters to a minimum. The algorithm is demonstrated to be capable of identifying flight-critical aircraft actuator and sensor failures at a low dynamic pressure (20,000 ft, 0.4 Mach). Research includes single and dual complete failures. Tuning methods for accommodating model mismatch, including addition of discrete dynamics pseudonoise and measurement pseudonoise, are discussed and demonstrated. Scalar residuals within each filter are also examined and characterized for possible use as an additional failure declaration voter. An investigation of algorithm performance off the nominal design conditions is accomplished as a first step towards full flight envelope coverage     

Interlaced Kalman filtering of 3D angular motion based on Euler'snonlinear equations

A novel Kalman filtering technique is presented that reduces the mean-square-error (MSE) between three-dimensional (3D) actual angular velocity values and estimated ones by an order of magnitude (when compared with the MSE resulting from direct measurements) even under extremely low signal-to-noise ratio conditions. The filtering problem is nonlinear in nature because the dynamics of 3D angular motion are described by Euler's equations. This nonlinear set of differential equations state that the angular acceleration in one axis is proportional to the torque applied to that axis, and to the products of angular velocity components in the other two axes of rotation. Instead of using extended Kalman filtering techniques to solve this complex problem, the authors developed a new approach where the nonlinear Euler's model is decomposed into two pseudolinear models (primary and secondary). The first model describes the time progression of the state vector containing the linear terms, while the other characterizes the propagation of the state vector containing the nonlinearities. This makes it possible to run two interlaced discrete-linear Kalman filters simultaneously. One filter estimates the values of the state vector containing the linear terms, while the other estimates the values of the state vector containing the nonlinear terms in the system. These estimates are then recombined, solving the nonlinear estimation process without linearizing the system. Thus, the new approach takes advantage of the simplicity, computational efficiency and higher convergence speed of the linear Kalman filter form and it overcomes many of the drawbacks typical of conventional extended Kalman filtering techniques. The high performance and effectiveness of this method is demonstrated through a computer simulation case study     

Adaptive Gaussian sum squared-root cubature Kalman filter with split-merge scheme for state estimation

The paper deals with state estimation problem of nonlinear non-Gaussian discrete dynamic systems for improvement of accuracy and consistency. An efficient new algorithm called the adaptive Gaussian-sum square-root cubature Kalman filter（AGSSCKF） with a split-merge scheme is proposed. It is developed based on the squared-root extension of newly introduced cubature Kalman filter（SCKF） and is built within a Gaussian-sum framework. Based on the condition that the probability density functions of process noises and initial state are denoted by a Gaussian sum using optimization method, a bank of SCKF are used as the sub-filters to estimate state of system with the corresponding weights respectively, which is adaptively updated. The new algorithm consists of an adaptive splitting and merging procedure according to a proposed split-decision model based on the nonlinearity degree of measurement. The results of two simulation scenarios（one-dimensional state estimation and bearings-only tracking） show that the proposed filter demonstrates comparable performance to the particle filter with significantly reduced computational cost.     

Incremental maneuver estimation model for target tracking

An incremental model for maneuver detection and estimation for use in target tracking with the Kalman filter is described. The approach is similar to the multiple Kalman filter bank, but with a memory for the maneuver status for the track under consideration. The advantage of this approach is that the target acceleration can be more accurately estimated. The maneuver-detection model has shown good maneuver-following capability. Moreover, it needs only a finite number of Kalman filters to handle all possible maneuver values and it responds quickly as maneuver occurs. When there is an abrupt maneuver change the model can still track the targets in short time     

采用序列概率比方法检测航空发动机传感器软故障

提出基于卡尔曼滤波和序列概率比方法进行某型涡扇发动机控制系统传感器软故障检测新方法.研究了采用修正的序列概率比方法处理滤波残差,检测传感器软故障;并将该方法与残差加权二乘算法WSSR(Weighted Sum of Squared Residual)检测传感器软故障过程进行了对比.仿真结果表明,序列概率比方法较WSSR法所需决策时间短,适合于航空发动机传感器软故障检测.      

Digital tracking filters with high order correlated measurement noise

The existing algorithms for the design of digital filters with colored measurement noise involve a restriction on the dimension of the measurement error model. Kalman filter equations and state space partition are used to formulate an optimal tracking filter without such restrictions. The input to the new filter are two consecutive measurements, and it is initialized by using the first available measurements and the error model correlation matrix. Several examples illustrate the filter formulation and initialization.     

Convolution Particle Filter for Parameter Estimation in General State-Space Models

The state-space modeling of partially observed dynamical systems generally requires estimates of unknown parameters. The dynamic state vector together with the static parameter vector can be considered as an augmented state vector. Classical filtering methods, such as the extended Kalman filter (EKF) and the bootstrap particle filter (PF), fail to estimate the augmented state vector. For these classical filters to handle the augmented state vector, a dynamic noise term should be artificially added to the parameter components or to the deterministic component of the dynamical system. However, this approach degrades the estimation performance of the filters. We propose a variant of the PF based on convolution kernel approximation techniques. This approach is tested on a simulated case study.     

航空发动机双重传感器故障诊断逻辑研究

针对航空发动机控制系统的双重传感器故障,提出了一种采用双路容错设计的卡尔曼滤波器故障检测隔离系统.故障检测隔离系统由一系列卡尔曼滤波器组成,每个滤波器都假定2路传感器故障,而以故障支路外的测量值作为输入量.当双重传感器故障发生时,只有不包含故障传感器信息的滤波器保持较低的估计残差,其他滤波器都会产生较大的估计残差,如此双重传感器故障便可以被隔离.利用滤波器组估计残差的特征,进一步设计合理的运算逻辑,系统就可以同时对传感器单一故障进行检测和隔离.为了验证故障诊断系统的有效性,在发动机慢车状态分别对传感器发生双重故障和单一故障的情况进行仿真.仿真结果表明:故障诊断系统能够准确有效地对传感器双重故障和单一故障进行检测和隔离.      

Kalman filtering for matrix estimation

A general discrete-time Kalman filter (KF) for state matrix estimation using matrix measurements is presented. The new algorithm evaluates the state matrix estimate and the estimation error covariance matrix in terms of the original system matrices. The proposed algorithm naturally fits systems which are most conveniently described by matrix process and measurement equations. Its formulation uses a compact notation for aiding both intuition and mathematical manipulation. It is a straightforward extension of the classical KF, and includes as special cases other matrix filters that were developed in the past. Beyond the analytical value of the matrix filter, it is shown through various examples arising in engineering problems that this filter can be computationally more efficient than its vectorized version.     

Kalman Filter Compensation for a Special Class of Systems

Filter compensation techniques for several special but practical cases are discussed. A general set of bias and covariance equations for linear filters with modeling errors is first summarized. A method for relating the modeling errors to the selection of the covariance of "process noise" for model error compensation is suggested. A performance ordering for cases when the true system becomes a subsystem of the model used for the filter construction is given. A bias correcting filter is derived for the case when the filter is matched only to a subsystem of the actual system.     

Hierarchical adaptive Kalman filtering for interplanetary orbitdetermination

A modular and flexible approach to adaptive Kalman filtering has recently been introduced using the framework of a mixture-of-experts regulated by a gating network. Each expert is a Kalman filter modeled with a different realization of the unknown system parameters. The unknown or uncertain parameters can include elements of the state transition matrix, observation mapping matrix, process noise covariance matrix, and measurement noise covariance matrix. The gating network performs on-line adaptation of the weights given to individual filters based on performance. The mixture-of-experts approach is extended here to a hierarchical architecture which involves multiple levels of gating. The proposed architecture provides a multilevel hypothesis testing capability. The utility of the hierarchical architecture is illustrated via the problem of interplanetary navigation (Mars Pathfinder) using simulated radiometric data. It serves as a useful tool for assisting navigation teams in the process of selecting the parameters of the navigational filter over various operating regimes. It is shown that the scheme has the capability of detecting changes in the system parameters and switching filters appropriately for optimal performance. Furthermore, the expectation-maximization (EM) algorithm is shown to be applicable in the proposed framework     

Multiple-model adaptive estimation using a residual correlationKalman filter bank

We propose a modified multiple model adaptive estimation (MMAE) algorithm that uses the time correlation of the Kalman filter residuals, in place of their scaled magnitude, to assign conditional probabilities for each of the modeled hypotheses. This modified algorithm, denoted the residual correlation Kalman filter bank (RCKFB), uses the magnitude of an estimate of the correlation of the residual with a slightly modified version of the usual MMAE hypothesis testing algorithm to assign the conditional probabilities to the various hypotheses that are modeled in the Kalman filter bank within the MMAE. This concept is used to detect flight control actuator failures, where the existence of a single frequency sinusoid (which is highly time correlated) in the residual of an elemental filter within an MMAE is indicative of that filter having the wrong actuator failure status hypothesis. This technique results in a delay in detecting the flight control actuator failure because several samples of the residual must be collected before the residual correlation can be estimated. However, it allows a significant reduction of the amplitude of the required system inputs for exciting the various system modes to enhance identifiability, to the point where they may possibly be subliminal, so as not to be objectionable to the pilot and passengers     

Performance Bounds for Continuous-time Filters in the Presence of Modeling Errors

We are concerned with obtaining bounds on the performance of Kalman-type, linear, continuous-time filters susceptible to modeling errors. Limiting the discussion to stationary performance, we obtain bounds on the performance index, the mean square error of estimates for suboptimal and optimal (Kalman) filters. The bounds are expressed in terms of the model matrices and the range of errors of the matrices. The results are useful to a designer in comparing the performance of a suboptimal filter with that of the optimal filter when he has information on the range of modeling errors. The tightness of the bounds is shown by an application of the results in the estimation of the motion of an aircraft carrier at sea.     

Adaptive Tracker Field-of-View Variation Via Multiple Model Filtering

Adaptive estimation using multiple model filtering is investigated as a means of changing the field of view as well as the bandwidth of an infrared image tracker when target acceleration can vary over a wide range. The multiple models are created by tuning filters for best performance at differing conditions of exhibited target behavior and differing physical size of their respective fields of view. Probabilistically weighted averaging provides the adaptation mechanism. Each filter involves online identification of the target shape function, so that this algorithm can be used against ill-defined and/or multiple-hot-spot targets. When each individual filter has the form of an enhanced correlator/linear Kalman filter, computational loading is very low. In contrast, an extended Kalman filter processing the raw infrared data directly and assuming a nonlinear constant turn-rate dynamics model provides superior tracking capability, especially for harsh maneuvers, at the cost of a larger computational burden.     

基于B样条模型的参数化卡尔曼滤波器设计

以标准B样条函数为基础,建立了以位置、样条系数为状态变量的参数化卡尔曼滤波器,用于解决外测数据的实时滤波问题。按照时间更新跨节点与否,分2种情况给出了状态转移方程。在时间更新跨样条节点时,使用样条函数的一阶连续导数条件,估计新增样条节点系数,由此实现滤波器在跨节点处的平滑过渡。通过仿真数据对新方法进行验证,并与已有的2类典型滤波方法进行比较,结果表明,本方法的滤波精度与另一类直接基于弹道信号表示的样条递推滤波方法精度相当,且可表现出更优的收敛性。新方法具有样条参数化模型的相同优点,可对时域信号全时段建模,可利用先验信息设计弹道优选节点而实现滤波性能优化,缺点在于状态更新的策略较为复杂。