Zhangming He

Data-Driven Design for static Model-Based fault diagnosis

This paper focuses on Data-Driven Design FOR Model-Based fault diagnosis, called D34MB for short. When the objective model is static, LVD (Latent Variable Detection) methods can be realized based on the LVE (Latent Variable Extraction) and LVR (Latent Variable Regression) techniques. A unified weight-framework for D34MB are proposed in this paper, which shows that all D34MB methods share the same procedures, i.e., LVE, LVR and LVD. The detection theorems shows that D34MB methods based on RRR (Rank Reduction Regression) and CCA (Canonical Correlation Analysis), compared with PCA (Principal Component Analysis) and PLS (Partial Least Square), tend to ensure higher calibration accuracy in terms of MSE as well as better detection performance in terms of FDR (Fault Detection Rate). In the case study, TEP (Tennessee Eastman Process) validates the correctness of our theoretical results.

Data-driven diagnosing for unanticipated fault by a general process model

The improvement of unanticipated fault detection and diagnosis (UFDD) capability is a difficult point, and is also a tendency for research and application. In this paper, a general process model (GPM) for unanticipated fault diagnosis is established. And combined with the characteristics of monitoring data, the corresponding diagnosis methods are researched. The model and the methods are used for online unanticipated fault detection, isolation and recognition. The GPM for the unanticipated fault diagnosis is designed, by adopting a three-layer progressive structure, which is comprised of an inherent detection layer (IDL), an unanticipated isolation layer (UIL) and an unanticipated recognition layer (URL). Several key problems in the GPM are analyzed, including the establishment and evaluation of detection statistics, the extraction of fault feature direction, and the design of fault isolation criterion and the calculation of contribution factor. The proposed model and methods are driven by pure data and they can detect and diagnose the unanticipated fault. The proposed approach is evaluated by using an example of a satellites attitude control system, and excellent results have been obtained.

Fault diagnosis for satellite's attitude determination system based on model error prediction and EMD

As one of the main parts for a satellites attitude measure and control system, star trackers and gyros tend to be damaged in the actual on-orbit operation. That often makes the satellite fail to finish the given task. In this paper, an approach is proposed to detect, identify and diagnose the fault for gyro and star tracker in a satellites attitude determination system. The model errors, including the model uncertainty and the fault factors, are estimated by predictive filter based on a satellites attitude kinematics equation firstly. And then the estimated model error is decomposed into some intrinsic mode functions (IMF) and one trend component by the empirical mode decomposition (EMD). Furthermore, the fault model for the trend component is established by the time series analysis. And the time series models parameters and residual variance are extracted as the feature vector, which is used to construct the discriminant function for the fault diagnosis of gyro and star tracker. Finally, promising simulated experimental results demonstrate the validity and effectiveness of the proposed method. Compared with the traditional fault diagnosis method, the detection and identification precision of the approach are improved significantly.

Maximum Correntropy Unscented Kalman Filter for Ballistic Missile Navigation System based on SINS/CNS Deeply Integrated Mode

Strap-down inertial navigation system/celestial navigation system (SINS/CNS) integrated navigation is a high precision navigation technique for ballistic missiles. The traditional navigation method has a divergence in the position error. A deeply integrated mode for SINS/CNS navigation system is proposed to improve the navigation accuracy of ballistic missile. The deeply integrated navigation principle is described and the observability of the navigation system is analyzed. The nonlinearity, as well as the large outliers and the Gaussian mixture noises, often exists during the actual navigation process, leading to the divergence phenomenon of the navigation filter. The new nonlinear Kalman filter on the basis of the maximum correntropy theory and unscented transformation, named the maximum correntropy unscented Kalman filter, is deduced, and the computational complexity is analyzed. The unscented transformation is used for restricting the nonlinearity of the system equation, and the maximum correntropy theory is used to deal with the non-Gaussian noises. Finally, numerical simulation illustrates the superiority of the proposed filter compared with the traditional unscented Kalman filter. The comparison results show that the large outliers and the influence of non-Gaussian noises for SINS/CNS deeply integrated navigation is significantly reduced through the proposed filter.

Maximum Correntropy Criterion Kalman Filter for α-Jerk Tracking Model with Non-Gaussian Noise

As one of the most critical issues for target track, α -jerk model is an effective maneuver target track model. Non-Gaussian noises always exist in the track process, which usually lead to inconsistency and divergence of the track filter. A novel Kalman filter is derived and applied on α -jerk tracking model to handle non-Gaussian noise. The weighted least square solution is presented and the standard Kalman filter is deduced firstly. A novel Kalman filter with the weighted least square based on the maximum correntropy criterion is deduced. The robustness of the maximum correntropy criterion is also analyzed with the influence function and compared with the Huber-based filter, and, moreover, the kernel size of Gaussian kernel plays an important role in the filter algorithm. A new adaptive kernel method is proposed in this paper to adjust the parameter in real time. Finally, simulation results indicate the validity and the efficiency of the proposed filter. The comparison study shows that the proposed filter can significantly reduce the noise influence for α -jerk model.

D34MB - a FDI toolbox with new features

A toolbox is developed in the MATLAB/GUI environment for FDI (Fault Detection and Isolation) which is an important field of automatic control theory and engineering. It provides an interactive interface for FDI. In this GUI (Graphical User Interface), a variety of methods are available for the FDI tasks, e.g., the traditional data-driven and model-based methods. a new feature of it is D34MB (Data Driven Design for Model Based) fault diagnosis. Some other new methods, e.g., robust, adaptive and local approach are also implemented. The sufficient benchmarks toolbox and the GUI feature ensure user-friendliness for operating this toolbox. Among all the methods, D34MB is emphasized, which is why the whole toolbox is called D34MB FDI toolbox. The new feature for static and dynamic model, the design, and the application of the toolbox is illustrated in this paper.

Robust diagnosis framework based on sliding least squares estimate and directional discrepancy

Proposed in this paper is a diagnosis framework based on SLSE (Sliding Least Squares Estimate) and directional discrepancy. This framework can be used when linear dynamic equation and beforehand baseline data are not available. It is efficient, for SLSE works by recursive computation, removing the oldest datum while accepting the newest. Its computational complexity is low and asks for little memory space. Detection statistic is given by the prediction residual and isolation statistic by directional discrepancy. Both detection and isolation criterion are based on hypothesis-testing procedure. Another feature of the framework is that it can diagnosis unanticipated fault, which has never happened before, without wrongly assigning it as some anticipated. The proposed method is especially fit for real-time diagnosis of stochastic system without accurate dynamic equation or complete fault patterns. The application to fault diagnosis of satellite control system demonstrates its validity.

A new fault diagnosis method based on component-wise expectation-maximization algorithm and K-means algorithm

In this paper, a fault diagnosis method is proposed based on component-wise expectation maximization algorithm and k-means algorithm, and it is applied to diagnosing the fault of the satellite attitude determination control system. First, Gaussian mixture model and the its traditional parameter estimation algorithm are reviewed. The component-wise expectation-maximization algorithm is used to estimate the parameters of Gaussian mixture model, which can lower the computational complexity of parameter estimation. Moreover, fault diagnosis, including detection and isolation, is carried out based on Gaussian mixture model, component-wise expectation maximization algorithm and k-means algorithm. Finally, the traditional method and our proposed method are applied for fault diagnosis on the satellite attitude determination control system. The simulation result shows that the new proposed method can, lower the computational complexity significantly, while the traditional and the new methods have nearly the same performance.

Data fusion for multi-structure and unequal-precision estimation

Data fusion for parameter estimation with multi-structure and unequal-precision is considered in this paper. Matrix tools e.g., congruent transformation, trace function and matrix differential, are used to analyze the estimation performance. Theoretical results reveal that: the single equipment estimate, the optimal fusion estimate, and the joint estimate are some special cases of the fusion estimate. Moreover, the precisions of different fusion estimation methods are compared, the relations among which are provided in the following theorems. The performance of the four estimates are validated by the simulation of trajectory calculation for the V-2 missile.

Research on prediction model and method for equipment state and fault in test range

The state and the fault prediction for the measure & control equipment in the test range plays an important role in the test and evaluation for the large-scale weapon system. It is also a research development in the state or fault process region. The objective of this paper is to address a state or fault prediction method for the equipment in the test range, which is combined with the physical network relationship for the predicted equipment and the mathematical analysis for the monitored state. According to the life rules of the electronic parts and the components of the predicted equipment, a distribution function for the abnormality state is constructed, and then the state prediction method for the discrete variables is especially researched. The effectiveness of the proposed prediction model and method is validated by a large optical system. The state or fault prediction model can realize the long-distance manage & control for the test ranges equipment, and can also provide the technical support for enhancing the equipment reliability during the test task.