Fault detection and isolation in hybrid process systems using a combined data‐driven and observer‐design methodology

The purpose of this paper is mainly aimed to provide an energy security strategy for the petroleum production and processing in the grand challenges. Fault detection and diagnosis is the central component of abnormal event management (AEM) [1]. According to the International Federation of Automatic Control (IFAC), a fault is defined as an unpermitted deviation of at least one characteristic property or parameter of the system from the acceptable/usual/standard condition [2-4]. Generally, there are three parts in a fault diagnosis system, detection, isolation, and identification [5, 6, 7]. Depending on the performance of fault diagnosis systems, they are called FD (for fault detection) or FDI (for fault detection and isolation) or FDIA (for fault detection, isolation and analysis) systems [5]. As the increasing needs for energy grows rapidly, energy security becomes an important issue especially in petroleum production and processing. The importance can be mainly considered in following perspectives: higher system performance, product quality, human safety, and cost efficiency [5, 8]. With this in mind, the purpose of this research is to develop a Fault Detection and Isolation (FDI) system which is capable to diagnosis multiple sensor faults in nonlinear cases. In order to lead this study closer to real world applications in oil industries, the system parameters of the applied system are assumed to be unknown. In the first step of the proposed method, phase space reconstruction techniques are used to reconstruct the phase space of the applied system. This step is aimed to infer the system property by the collected sensor measurements. The second step is to use the reconstructed phase space to predict future sensor measurements, and residual signals are generated by comparing the actually measured measurements to the predicted measurements. Since, in practice, residual signals will not perfectly equal to zero in the fault-free situation, Multiple Hypothesis Shiryayev Sequential Probability Test (MHSSPT) is introduced to further process those residual signals, and the diagnostic results are presented in probability. In addition, the proposed method is extended to a non-stationary case by using the conservation/dissipation property in phase space. The proposed method is examined by both of simulated data and real process data. The three tank model is modeled according to a nonlinear laboratory setup DTS200 and introduced for generating simulated data. On the other hand, the real process data collected from a sugar factory actuator system are also used to examine the proposed method. According to our results obtained from simulations and experiments, the proposed method is capable to indicate both of healthy and faulty situations. In the end, we have to emphasize that the proposed approach is not limited in the applications of petroleum production and processing. For example, our approach can also apply to enhance the quality of water and avoid the discharges, such as leakage, in the process of water resource management. Therefore, the proposed approach not only benefits the issue of energy security but also other issues in the grand challenges.

In this research, a hybrid system for fault detection and isolation (FDI) is proposed. The system incorporates the well-known benefits of residual generation approaches with a classification system for the automatic determination of thresholding. The architecture of the FDI system is divided in three levels: (i) the first one models the monitoring system signals (sensors) with a set of neural networks; (ii) the second makes the residual generation based on the difference of the actual output and the estimation of the neural network; ; (iii) and the third level assesses the set of residues using a classification system, at this stage the detection and fault isolation is done. The FDI system is evaluated in a simulated model of wind turbine with nine different types of issues. The results suggest a decrease in the number of false alarms, compared to a data-driven approach without loss in detection speed.

We present a geometric approach for fault detection and isolation (FDI) in robotic manipulators in presence of model uncertainty. A systematic procedure is introduced for representing robotic system model being affine with respect to faults and disturbances. The proposed residual generator has smooth dynamics with freely selectable functions and it does not require high gains or threshold adjustment for the FDI purpose. No assumption on amplitude of faults and their rate is used. The solvability conditions for the FDI problem lead to a quotient observable subspace unaffected by all unknown inputs except the faults. Simulation example demonstrates localization of faults in presence of uncertainty in link moment-of-inertia matrices and measurement noise.

This article has been retracted. A retraction notice can be found at https://doi.org/10.3233/JIFS-219433.

A methodology for fault detection and isolation (FDI) of electrical faults in a three-phase permanent magnet synchronous motor (PMSM) is presented. The considered faults are stator winding and current sensor faults. The former consist of wire degradation (characterized by a rise in the corresponding phase resistance), turn-to-turn short circuits and phase-to-phase short circuits. The sensor faults consist of gain changes and offsets in phase current measurements. A multi-model adaptive estimator of fault parameters is developed by resorting to specific signal models. These models characterize phase current measurements (i.e. the signals) in the healthy and in each faulty operating mode. Such signal models do not require knowledge of the three-phase motor parameters. The severity of the fault is evaluated through the estimators that take the form of three dual extended Kalman filters (DEKFs) and one Kalman filter (KF). The proposed approach is validated by using synthetic data of phase current measurements. These data were generated by a digital twin of an electromechanical actuator (EMA) driven by a PMSM. This motor drive is subject to closed-loop speed control and includes realistic sensor noise. To test the robustness of the approach, its performance is evaluated with different levels of sensor noise.

In this contribution different modelling stragies for fault detection and isolation (FDI) in view of the model complexity are studied at the benchmark problem Three Tank System. The particularities of models needed for model-based FDI in contrast to models used in control are pointed out. Of special interest is the question of complexity. This depends basically on the given situation such as the kind of plant, the kind and number of faults to be detected, the demands for fault isolation and robustness and the measurements available. The paper shows for a certain process that diagnostic models for controllable and observable plants comprise only a partial description of the input/output model and are therefore less complex than those for control. This issue is discussed in terms of different model-based FDI approaches — analytical and data-based. The results are demonstrated in a case study, where several model-based FDI approaches are applied to the Three Tank System, which has been declared as a benchmark within the COSY project of the European Science Foundation (ESF).

Comparison Between Statistical and Observer-Based Approaches for Fault Detection and Isolation in a Chemical Process

In this paper, a robust unknown input observer-based actuator fault detection and isolation (FDI) scheme is applied on an example aircraft fluid thermal management system (FTMS). System dynamics are modeled using a graph-based approach. A linearized state space model is used in a bank of unknown input observers (UIOs) to generate a set of structured robust (in the sense of disturbance decoupling) residuals. Simulation results show successful actuator FDI in the presence of unknown inputs.

Modeling and fault diagnosis design for HVAC systems using recurrent neural networks

Functional redundancy techniques for fault detection and isolation (FDI) in dynamic systems requires close interaction between system instrumentation, modeling and analysis. Effective FDI requires detailed and accurate models to trade and analyze system behavior, including transient phenomena that result from faults. It also requires appropriate instrumentation technology to provide the measurements to capture and analyze system behavior. Models and measurements must be matched carefully to provide sufficient observability and effective analysis. In this paper we demonstrate the development of FDI systems for complex applications by presenting the modeling and instrumentation of an automobile combustion engine cooling system. We have developed a qualitative parameter estimation methodology for FDI. A system model is represented as a graph that captures the dynamic behavior of the system. To demonstrate the applicability a small leak is artificially introduced in the cooling system and accurately detected and isolated.

To monitor the dynamics and non-stationarity inherent in industrial processes, we propose a novel incipient fault detection and isolation scheme grounded in a probabilistic perspective, using the Cauchy–Schwarz (CS) divergence. Our innovation lies in the utilization of marginal CS divergence for incipient fault detection and the conditional CS divergence for fault isolation. This approach neither require prior parametric assumptions about the underlying data distribution nor depend on historical fault data, while simultaneously providing explanatory diagnostics. Beyond this, we develop a change point detection-base diagnosis technique for practical engineering applications. This online process monitoring technique guarantees timely intervention to uphold process stability and safety. We demonstrate the compelling performance, higher detection rate and lower alarm rate, of the CS divergence over prevalent divergence-based approaches, such as Kullback–Leibler divergence, Wasserstein distance and Mahalanobis distance. We also illustrate the explanatory insights offered by conditional CS divergence in fault isolation on synthetic data, benchmarks of continuous stirred-tank reactor process and continuous stirred-tank heater process and even a real-world continuous catalytic reforming process. Code of this CS divergence based fault detection and isolation is available at https://github.com/Feiya-Lv/Incipient-Fault-Detection-and-Isolation-with-Cauchy--Schwarz-Divergence.

Subspace identification for FDI in systems with non-uniformly sampled multirate data

This paper deals with actuator fault diagnosis of neutral delayed systems with multiple time delays using an unknown input observer. The main purpose is to design an observer that guarantees the asymptotic stability of the estimate error dynamics and the actuator fault detection. The existence conditions for such an observer are established. The main problem studied in this paper aims at designing observer‐based fault detection and isolation. The designed observer enhances the robust diagnosis performance, including rapidity and accuracy, and generates residuals that enjoy perfect decoupling properties among faults. Based on Lyapunov stability theory, the design of the observer is formulated in terms of linear matrix inequalities, and the diagnosis scheme is based on a bank of unknown input observers for residual generation that guarantees fault detection and isolation in the presence of external disturbances. A numerical example is presented to illustrate the efficiency of the proposed approach.

A novel distributed fault detection and isolation (FDI) methodology for a network of nonhomogeneous multi-agent systems is proposed in this paper. An FDI filter is designed such that the effects of the disturbances and control inputs on the residual signals are minimized (for the purpose of fault detection) subject to the constraint that the transfer matrix function from the faults to the residuals is equal to a pre-assigned diagonal transfer matrix (for the purpose of fault isolation). It is shown that by using this approach each agent not only can detect and isolate its own faults but also is able to detect and isolate faults of its neighboring agents. Moreover, the method covers isolation of possibly simultaneous occurring faults in the system. Sufficient conditions for solvability of the problem are obtained in terms of linear matrix inequality (LMI) feasibility conditions. Extended LMIs characterization is used to reduce the conservativeness of the solution by introducing additional variables, to eliminate the coupling of the Lyapunov matrices with the system matrices. Simulation results illustrate the effectiveness of the proposed design methodology.

Model-based fault detection and diagnosis is an interesting method, due to economical and safety related matters. However, in practice it is very difficult to achieve accurate models for complex nonlinear plants. If the plant structure is not precisely known, the diagnosis has to be based primarily on data or heuristic information. The inherent characteristics of fuzzy logic theory make it suitable for fault detection and isolation (FDI). In this paper is proposed a modification to the regularity criterion algorithm, aiming the reduction of computational time without loss of accuracy. The fuzzy models obtained using the modified regularity criterion algorithms are optimized by a real-coded genetic algorithm. An industrial valve simulator is used to obtain several abrupt and incipient faults in the system. These faults are some of the possible faults in the real system. The fuzzy models used in the FDI system were able to detect and isolate the simulated abrupt and incipient faults