Research articleSensor fault detection in a class of nonlinear systems using modal Kalman filter
Introduction
The growing need for safe, optimal and efficient operation of many complicated and expensive industrial systems is one of the main incentives to find the best solution for fault detection and identification in such systems. Fault occurrence is inevitable; hence, a fast and reliable fault detection method, in order to prevent disaster event, is an essential need.
Regarding the fault detection methods used, two general categories can be identified: data-driven methods and model-based methods. Data-driven methods, utilize the information collected from the system for fault detection purposes. These approaches mostly rely on real-time or historical data and use artificial intelligence algorithms, statistically process data or consist of expert systems. The main advantage of these methods is that no accurate model of the system is required and they have been successfully applied for fault detection purposes [1]. However, the use of recorded data streams for fault detection requires a large database, and the enormous computational effort makes these methods impractical for online applications.
On the other hand, model-based methods assume that the accurate model of the system and its parameters are precisely known. These methods are based on the difference between the process measurement and the model measurement, through the generation of “residual” signal. By evaluating the generated residual, fault occurrence could be detected [2]. When a precise mathematical model of the system is available, model-based methods are more powerful, in comparison to data-based ones, and provide better real-time performance. Different model-based techniques, such as parity space [3], parameter estimation [4] and state estimation [5], have been used in literature, in order to detect fault occurrence. Among the existing model-based methods which use state estimation to generate “residual” signal, Kalman filter (KF) variants have found widespread applications.
Kalman filter has been a powerful tool, which provides optimal state estimation for linear systems with Gaussian noise, since the 1960s [6]. The algorithm consists of two steps: the state update which produces the a priori estimate of the state, which does not include observation information from the current timestep, and the measurement update which provides the a posteriori estimate of the state, by combining the current a priori prediction with current observation information to decreases state error covariance and produce the a posteriori state estimate. Its recursive nature makes it a suitable option for implementations. Optimality of linear Kalman filter (LKF) for linear systems state estimation encourages other modifications of this filter to be developed, such as extended Kalman filter (EKF), for nonlinear systems state estimation. The main idea of EKF is the linearization of the nonlinear system, and then applying LKF to obtain the state estimation. This linearization takes place at any time step around the last estimated state. To overcome the issue of linearization errors, another modification of Kalman filter, called Unscented Kalman Filter (UKF) is presented [7]. In this algorithm, the state probability distribution is represented by using a set of samples, called sigma points.
In the 1970s, the use of this method for fault detection was presented by Mehra and Peschon [8]. Since then, it has been used widely as a residual generator for fault detection. EKF and UKF are also successfully used for nonlinear systems fault detection [9], [10]. Faults can occur in process components, actuators or sensors. In this work, sensor fault detection is addressed, with the assumption of non-faulty performance of other components of the system.
Sensors are used in almost every industrial process and they are of great importance to ensure a safe and reliable operation of the systems. The information, which the sensors provide, is essential to have an accurate assessment of the condition of the monitored system and take proper actions in case of any abnormality. Faults in sensors can cause degradation in system performance and even make the control system unstable. In addition, the defective output of the sensor can lead to incorrect decisions and component replacements. Therefore, constant monitoring of sensors performance and sensors fault detection is essential and various methods are used to detect sensor faults in nonlinear systems. [11], [12], [13], [14]. Kalman filter and its modifications, as well-known optimal filters, are widely used to detect sensor faults for linear and nonlinear systems in the presence of measurement noise and disturbances [15], [16], [17]. In this paper, the ability of a new modification of KF, called modal Kalman filter (MKF) for sensors fault detection in a class of certain nonlinear systems will be investigated.
In [18] and [19] a method, called modal series, is introduced for investigation and modeling of autonomous nonlinear systems. In this proposed approach the modal series are used to transform the state estimation of a nonlinear system into an infinite number of state estimations of linear systems. In this method, system modeling is more accurate than the linearization of nonlinear systems. In [20] the approach is extended to find the solution of nonautonomous nonlinear differential equations. In [21] an extended form of modal series is used to present a nonlinear model predictive control approach.
Modal Kalman filter is introduced in [22]. In this method, in order to estimate the states of the nonlinear system, instead of using the first linear term of the Taylor series extension, higher-order terms are used. This filter, based on modal series, converts the state estimation of a nonlinear system into a series of linear systems state estimation. Since Kalman filter is an optimal and effective state estimator for linear systems, it is used to estimate the state of each linear system in this series. In other words, Kalman filter is used more than once at each step in nonlinear system state estimation. As it has been discussed in [22], the estimation error in this case is much lower than one by using KF, and in its extended modification, the error is lower than EKF. The main contribution of this paper is to extend the formulas of MKF, which are presented in [22], to make it possible to use this filter for state estimation of physical nonlinear plants with inputs, and to use this filter for the purpose of sensors fault detection in nonlinear systems.
The remainder of this paper is organized as follows: modal Kalman filter and its mathematical formulation are introduced in Section 2. In Section 3 the MKF-based sensor fault detection method is explained. In order to evaluate the proposed method and to verify its effectiveness, simulation results are given in Section 4, and the performance of MKF in sensor fault detection is compared with KF and EKF. Finally, in Section 5 conclusions are presented.
Section snippets
Problem formulation
The following nonlinear differential equation is considered: where , and are states vector, a nonlinear smooth function and initial states condition, respectively. Assuming as a stable operating point for the nonlinear system (1), it is shown in [23] that the solution of the nonlinear system (1), the modal series of , can be obtained as follows: where are continuous-time vectors that satisfy the following linear
Sensor fault detection
Sensor faults classification, based on the impact on sensor readings, can be defined as follows:
Bias: The constant value added to the sensor’s output signal.
Drift: A variable with time, , that is added to the sensor’s signal. This value usually changes linearly with time.
Loss of effectiveness: In this case, the signal is multiplied by a coefficient
Hard Failure: the sensor output stays constant at zero or a non-zero value.
Bias and hard failure are abrupt faults, which occur
Mixing tank
A mixing tank process example, as shown in Fig. 3, is used in this work to evaluate the performance of the proposed methodology in sensor fault detection. Nonlinear equations of the system are according (47), (48), where is thestate vector. The elements of indicate the height and the temperature of the liquid in the tank, respectively. is the input vector, represents the flow of two liquid inlets, with temperatures and .
Conclusion
This paper presents a model-based sensor fault detection scheme, for a class of certain nonlinear systems, using modal Kalman filter. This method offers a great enhancement in accurate state estimation, which can make the residuals smoother in the non-faulty mode. Generation of smoother and lower residuals results in more sensitivity to fault occurrence; as a result, the false alarm rate and the detection delay are noticeably reduced. Two simulation examples, with different fault types,
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (30)
- et al.
Residual-based fault detection using soft computing techniques for condition monitoring at rolling mills
Inf Sci
(2014) Fault detection and isolation using parity relations
Control Eng Pract
(1997)- et al.
A bounded-error approach to simultaneous state and actuator faultestimation for a class of nonlinear systems
J Process Control
(2017) - et al.
An innovations approach to fault detection and diagnosis in dynamic systems
Automatica
(1971) - et al.
Robust sensor fault detection and isolation of gas turbine engines subjected to time-varying parameter uncertainties
Mech Syst Signal Process
(2016) - et al.
From data to diagnosis and control using generalized orthonormal basis filters. part II: Model predictive and fault tolerant control
J Process Control
(2006) - et al.
Robust sensor fault detection based on nonlinear unknown input observer
Measurement
(2014) - et al.
Fault detection inmulti-sensor networks based on multivariate time-series models and orthogonal transformations
Inf Fusion
(2014) - et al.
Fault prediction for nonlinear stochastic system with incipient faults based on particle filter and nonlinear regression
ISA Trans
(2017) - et al.
Robust Model-Based Fault Diagnosis for Dynamic Systems
(1999)
Recursive least squares method in parameters identification of DC motor models
Ser: Elec: Ener: Vol work
A new approach to linear filtering and prediction problems
J Basic Eng
Unscented filtering and nonlinear estimation
Proc IEEE
Hydraulic actuator leakage fault detection using extended kalman filter
Int. J. Fluid Power
A sensor fault detection and isolation method in interior permanent-magnet synchronous motor drives based on an extended kalman filter
IEEE Trans Ind Electron
Cited by (11)
Interval estimation for nabla fractional order linear time-invariant systems
2022, ISA TransactionsCitation Excerpt :State estimation has been regarded as a critical problem in the control community and extensively investigated for application fields such as controller design [1], reliability analysis [2], and fault detection [3].
Model-based fault diagnosis methods for systems with stochastic process – A survey
2022, NeurocomputingCitation Excerpt :However, the linearization error generated during the first-order derivative process of the nonlinear function cannot be ignored, and the derivation of Jacobian matrices is not applicable in practice. Instead of utilizing the first linear term of Taylor expansion, modal Kalman filter (MKF) converts the nonlinear filter into a set of linear filters based on the modal series, and obtains a smaller estimation error [71]. The unscented Kalman filter (UKF) is constructed on the basis of unscented transform technique, a set of sigma points is utilized to represent each state variable and approximate the posterior information of the output variable with a higher-order accuracy when propagated in the nonlinear systems [72], which means that the UKF can cope with higher-order nonlinear systems.
Fuzzy extended state observer for the fault detection and identification
2022, ISA TransactionsCitation Excerpt :Meanwhile, model-based methods assume an accurate model and estimate the fault through a residual signal by employing an observer. This signal represents the difference between the process and the model measurements [6]. In this context, significant researches have been made in observer-based methodologies applied to dynamical systems.
Quantized recursive filtering for networked systems with stochastic transmission delays
2022, ISA TransactionsDetection of Unit of Measure Inconsistency in gas turbine sensors by means of Support Vector Machine classifier
2022, ISA TransactionsCitation Excerpt :The same data features can be also exploited for detecting, and subsequently classifying, sensor faults, whose most frequent types are reported in Table 1. For example, based on the definition reported in [32], a bias fault can be detected by comparing its standard deviation and mean value to the healthy signal. In fact, the standard deviation of the faulty signal is comparable to the standard deviation of the healthy signal, whereas its mean value can be lower or higher than that of the healthy signal.
Robust M-estimation-based maximum correntropy Kalman filter
2022, ISA TransactionsCitation Excerpt :The Kalman filter (KF) is derived from the mean-square-error (MSE) under the condition of a linear model and Gaussian noises [1,2]. It is widely used in target tracking [3,4], navigation [5], multi-sensor filtering fusion estimation [6], sensor fault detection [7], dynamic system fault diagnosis [8], communication networks discrete-time estimation [9], and digital chaotic systems [10]. In practice, the noise does not obey a Gaussian distribution in fields, such as radar receiver noise [11], colored electromagnetic interference in transmission line inspection [12], and long-tail noise in radio systems [13].