Search

To guarantee the safe operation of systems, it is necessary to use systematic techniques to detect and isolate faults for the purpose of diagnosis. The Fault Detection and Isolation (FDI) of nonlinear systems with coupling multiple energies became a difficult task. This why, we propose in this paper, the exploitation of the behavioral and structural properties of graphs (Bond Graph and Signed Directed Graph) combining with a Principal Component Analysis method (PCA). Therein, a coupled Bond Graph model is used for modeling methodology. A Signed Directed Graph (SDG) is then deduced. Fault detection is later carried out by graph coloring. The localization of the actual fault is performed based on a nonlinear PCA (NLPCA) and back/forward propagations on the SDG. The proposed approach is tested on the thermofluid case study and some simulations are provided.

Fault diagnosis has become a significant issue due to rising demands for reliability as well as higher system performance and product quality. There are various approaches for the goal of fault detection and isolation among which we focus on the unknown input observer since observer-based schemes have been successfully adopted in a variety of application fields. This paper presents a novel design of Unknown Input Observer (UIO) for fault detection and isolation which can be applied to non-minimum phase systems. Unlike many previous UIO methods developed in the literature, the proposed scheme can carry out real-time reconstruction of unknown input in the presence of unstable invariant zero (with respect to the relation between the output and the unknown input). We extend the system by augmenting low pass filter(s) in the input(s)-output(s) path to tackle the non-minimum phase problem and generate robust residuals. Finally, the effectiveness of the proposed design method is confirmed through demonstrating a simulation-base example.

Active fault diagnosis can improve the diagnosability of potential faults by injecting a suitable input into the system. This input can be designed using either a stochastic or a deterministic framework. The stochastic approach aims to maximize the probability of a correct diagnosis at a certain time, whereas the deterministic approach aims to guarantee diagnosis within a certain time interval. Recently, a hybrid stochastic-deterministic approach has been developed in which all uncertainties are described by uniform probability density functions (PDFs) with finite support on zonotopes. This method is able to provide a guaranteed diagnosis at a given time N, while approximately maximizing the probability of diagnosis at some earlier time M

In this article, concrete aerospace industrial achievements corresponding to a level 5 on the Technological Readiness Level (TRL) scale are presented. Starting from basic research levels up to industrial validation, it is shown how it is possible to bridge the gap between advanced methods advocated by the academia and the more and more demanding industrial needs. To illustrate this position, a focus is made on real-time on-board Fault Detection and Diagnosis (FDD) for the upcoming and future innovative Flight Control System (FCS) of civil aircraft. This is exemplified by three concrete cases presenting the research and development process, from theory up to simulator and flight tests: 1) detection of Oscillatory Failure Cases in FCS thanks to a derivative-free variant of an extended Kalman filter; 2) detection of control surface lock-in-place failure (a.k.a. jamming) via recursive parametric modeling; 3) flight parameter Data Fusion with dedicated fuzzy logic.

An enhanced Fault Detection and Isolation (FDI) technique based on 't' statistical test is presented here employing a residual based R-Adaptive Unscented Kalman Filter (AUKF). Although the use of AUKF is common for estimation problem, employing AUKF for FDI duty is quite rare. Adaptive Unscented Kalman filtering framework is chosen here due to its derivative free calculations in the algorithm and reportedly better accuracy. Monitoring the 't' statistic of the innovation sequence, measurement faults are detected and corresponding alarm signals are generated here. The presented 't' statistic based FDI technique comprising of consistency checking and abruptness checking is straight forward, demands less computational burden and minimizes false alarms. The superiority of the proposed AUKF based FDI method compared to non adaptive, nonlinear filtering based FDI techniques is demonstrated here by an extensive simulation study executed on a nonlinear LEO satellite planar model. The use of AUKF has provided faster convergence speed and is also found to be advantageous for the situations where the measurement noise statistics are unknown.

Reliability and sustainability of wind turbines increase in importance as wind turbines contribute with increasing power generation to the world's power grids. One possible way to achieve this is by using advanced fault detection and isolation methods in wind turbines based on the measurements provided to the control system. In this paper a Karhunen-Loeve basis approach is designed for detecting changes in frequency response from rotating parts like a gearbox. The potential of this method is shown by applying it to an established Wind Turbine FDI and FTC Benchmark model. These faults are industrial wind turbines detected using auxiliary condition monitoring systems, which uses their own sensors etc. Usage of the existing control system including sensors will thereby provide a potential cost reduction of the wind turbine.

In this paper, an adaptive fault diagnosis technique is used for fault detection in Lithium ion batteries. The monitoring setup consists of multiple models representing the different degree of parameter shift due to over-discharge in the Lithium ion battery. A recursive least square estimator along with equivalent circuit methodology is used to construct the non-linear battery models. Extended Kalman filters are used to generate the estimated terminal voltages for each system. The residuals are further evaluated using the conditional probability evaluation function, to generate probabilities that determine the presence of a particular operational condition. Using experimental data, it is shown that Li-ion battery performance shift due to over-discharge can be accurately detected in real time.

In order to obtain lower cost of energy for wind turbines fault detection and accommodation is important. Expensive condition monitoring systems are often used to monitor the condition of rotating and vibrating system parts. One example is the gearbox in a wind turbine. This system is operated in parallel to the control system, using different computers and additional often expensive sensors. In this paper a simple filter based algorithm is proposed to detect changes in a resonance frequency in a system, exemplified with faults resulting in changes in the resonance frequency in the wind turbine gearbox. Only the generator speed measurement which is available in even simple wind turbine control systems is used as input. Consequently this proposed scheme does not need additional sensors and computers for monitoring the condition of the wind gearbox. The scheme is evaluated on a wide-spread wind turbine fault detection and fault tolerant control benchmark model, in which one of the included faults results in a change in the gear box resonance frequency. This evaluation shows the potential of the proposed scheme to monitor the condition of wind turbine gear boxes in the existing control system.

In this paper a fault diagnosis framework based on detection with feature extraction methods and identification based on data-driven process topology methods was investigated. A simulation of a simple system consisting of two tanks with heat exchangers was used to generate data for normal operating conditions and a number of faults. Fault detection methods included principal component analysis and kernel principal component analysis feature extraction with Shewhart, cumulative sum and exponentially weighted moving average monitoring charts. Process topology information was extracted with linear cross-correlation, partial cross-correlation and transfer entropy. Connectivity maps were constructed to identify possible fault propagation paths to aid root cause analysis and changes in connectivity structure due to faults were exploited for fault identification. Kernel principal component analysis with a CUSUM chart gave the best detection performance, while connectivity graphs based on partial correlation gave an accurate representation of the system and assisted fault identification.

This paper presents a first principle model based methodology for simultaneous optimal tuning of a fault detection algorithm and a feedback controller. The key idea is to calculate the effect of stochastic input disturbances on the variability of the output variables by using a generalized polynomial chaos (gPC) expansion and a mechanic model of the process. A two-level optimization is proposed for simultaneously tuning the fault detection and controller algorithms. The goal of the outer level optimization is to find a trade-off between the efficiency for detecting faults and the closed loop performance, while the inner optimization is designed to optimally calibrate the fault detection algorithm. The proposed method is illustrated for a continuous stirred tank reactor (CSTR). The results show that the computational cost of the gPC-based method is significantly lower than a Monte Carlo (MC) simulation-based approach, thus demonstrating the potential of the gPC method for dealing with large problems.

In safety critical systems, the control system is composed of a core control system with a fault detection and isolation scheme together with a repair or a recovery strategy. The time that it takes to detect, isolate, and recover from the fault (fault recovery time) is a critical factor in safety of a system. It must be guaranteed that the trajectory of a system subject to fault remains in the region of attraction (ROA) of the post-fault system during this time. This paper proposes a new algorithm to compute the critical fault recovery time for nonlinear systems with polynomial vector fields using sum of squares programming. The proposed algorithm is based on computation of ROA of the recovered system and finite-time stability of the faulty system.

Active fault isolation of parametric faults in closed-loop MIMO systems are considered in this paper. The fault isolation consists of two steps. The first step is group- wise fault isolation. Here, a group of faults is isolated from other possible faults in the system. The group-wise fault isolation is based directly on the input/output signals applied for the fault detection. It is guaranteed that the fault group includes the fault that had occurred in the system. The second step is individual fault isolation in the fault group. Both types of isolation are obtained by applying dedicated auxiliary inputs and the associated residual outputs.

This paper investigates an online effective method for tool condition monitoring. Acoustic emission signal of a system which is acquired by a sensor mounted to the spindle of the milling machining center is used as the fault indicator because it is easily to be installed, inexpensive and practical for use in industrial environment. Time-frequency analysis is selected for signal processing step based on its ability to reveal time and frequency variant characteristics of faulty signal. S-transform is used as a powerful time-frequency method for this purpose. Because of the high dimension of the time-frequency results, it is desirable to use a local region of interest in time-frequency domain instead of using the entire information, for fast and accurate monitoring and detection when any abnormal/fault operating condition might occur. Such a strategy also helps to reduce the computation cost which is necessary for online applications and improves the interpretation resolution for law quality signals. An optimization method based on genetic algorithm is used for finding the most discriminative local area as the region of interest in time-frequency domain. For feature generation step, a correlation coefficient between each signal and the healthy signal is assigned to the signal using a 2-D correlation analysis. Curve fitting approach is then used to determine a function to approximate the fault value based on the correlation coefficients. Experimental results based on a milling machine under different operating conditions show that this method has a high accuracy for fault detection. It is also concluded that the accuracy of the local feature extraction is higher than the conventional ways.

The object of this paper is to address data-driven fault detection design for systems with unsteady trend, which shows cyclicity, monotonicity and non-zero mean. Firstly, mean theorem and covariance theorem are proposed and proved. The former is the mean property of projection matrix, and the latter is the recursive formula for covariance matrix of regression residual. Secondly, an improved fault detection statistic, called Least Square T 2 (LST 2 ), is proposed. It can partly solve the detection problem for systems with unsteady trend. The improvement can also partly cope with the limitations of the traditional multivariate detection methods, such as Principal Component Analysis (PCA). Thirdly, based on the two theorems, a recursive algorithm and a moving window algorithm of LST 2 are given, thus both time and space complexity are greatly reduced for online detection. The effectiveness of the presented detection statistic is evaluated with an application of monitoring satellite attitude control system. The case study result shows that the false alarm rate of LST 2 is much lower than that of T 2 based on PCA, while LST 2 is more sensitive to fault.

A sequential method for Control Performance Diagnosis using a classification tree to predict possible root-causes of poor performance is presented. The classification tree methodology is used to combine process pre-assessment (nonlinearities detection, delays estimation and controller assessment), control performance assessment (CPA) and analysis of variance (ANOVA) into an integrated framework. An initial process data set is analysed and the results are used as decision thresholds for the classification tree. The methodology is capable to identify root-causes such as poor tuning, inadequate control structure, nonlinearities, process mismatch and disturbance changes. The proposed methodology is applied to individual loops of a tandem cold rolling mill.

This paper design a method for fault detection and isolation based on observers for systems modelled as Descriptor-Linear Parameter Varying (D-LPV) with Unmeasurable Scheduling Functions (USF). The first contribution of the paper is deal with the USF problem by transforming the into an uncertain D-LPV system with an estimated scheduling parameter. As a second contribution, a robust LPV observer is designed with H ∞ performance to supply a robust state estimation against uncertainties provided by USF. Sufficient conditions to guarantee the robustness and convergence are obtained via linear matrix inequalities (LMIs). Finally, an observer bank based on robust H ∞ observers is used to generate residuals and perform sensor fault detection and isolation. A numerical example demonstrates the effectiveness of our methods.

This work focuses on the design of a fault detection and fault-tolerant control framework for spatially distributed processes modeled by highly-dissipative partial differential equations (PDEs) subject to external disturbances and sensor faults. The main objective is to devise a suitable sensor reconfiguration strategy to reduce the performance degradation due to the errors resulting from the sensor faults. Initially, a finite-dimensional system that captures the slow dynamics of the PDE is derived and used to design a obverse-based output feedback controller. Using Lyapunov techniques, the fault-free and faulty behavior of the closed-loop system are characterized in terms of the sensor spatial placement, the size of the disturbances, the magnitude of the sensor faults as well as the controller and observer design parameters. Based on the fault-free closed-loop dynamics, the Lyapunov stability bound is used as an alarm threshold to declare the presence of sensor faults. To suppress the performance deterioration, a performance-based sensor reconfiguration policy is developed whereby the supervisor determines either to continue using the current faulty sensors or to switch to suitable backup sensors based on a comparison between the sizes of the achievable terminal sets. A singular perturbation formulation is used to analyze the implementation of the sensor fault-tolerant control structure on the infinite-dimensional system. Finally, the results are illustrated through an application to a representative diffusion-reaction process example.

Since incipient faults often have small amplitudes and obscure symptoms, those observer-based or output residual-based fault indicators are ineffective for incipient faults, especially in the presence of system disturbances. The ToMFIR-based fault detection approaches use both controller and output residuals to construct a fault indicator, which make them sensitive to incipient faults. This paper extends the ToMFIR-based fault detection approaches to a class of closed-loop nonlinear systems under the consideration of system disturbances. The restriction on the fault type is removed compared with the existing approaches, and ToMFIR-based fault estimation algorithm has been derived. Verification results in a near space hypersonic vehicle (NSHV) simulation system can demonstrate the effectiveness of the proposed approach.

In this paper, an output feedback controller in the structure of model predictive control is developed focusing on model discrimination and fault detection and isolation in addition to a tracking performance. The development of the controller is based on a system identification method that is used for an update of a model of an underlying system. For the systems whose dynamics may change at an unknown time, this identification method uses only a relatively small amount of data recently collected from the underlying system. The signal generated by the controller is applied to the system and makes the system to produce input-output data such that models are more distinguishable and faults are more detectable. Since the optimization problem in the controller is nonconvex, a suboptimal solution via a semidefinite relaxation technique is pursued.

In this paper, a robust H ∞ fault detection filter is designed for a class of discrete-time nonlinear networked control systems via T-S fuzzy model with multiple bounded state delay and random packet dropout induced by the limited bandwidth of communication networks. Our aim in this paper is to analyze and design a robust H ∞ full-order fault detection filter, such that the filtering error dynamics is asymptotically mean-square stable with a prescribed H ∞ performance level. Sufficient conditions for the existence of the desired filter are presented. Finally, an example is given to illustrate the effectiveness and applicability of the proposed new design method.

We consider distributed methods for detection of cyber attacks and faults in power networks. The approach is based on grouping of buses in the system, and for each group, we design filters for detection and isolation of faults applied to the buses within the group. The scheme is distributed in the sense that it uses only locally available data such as the generated power, loads, and power flows. Specifically, a fault detection method based on a geometric approach is applied to two different settings. Depending on the availability of phasor measurement units, we develop indirect and direct methods and compare their characteristics in performance and computation. A numerical example is provided to illustrate the results.

In order to sustain safety and high product quality, the data-driven fault detection tools are increasingly used in industrial processes. The quality variables are the key index of the final product. Obtaining them in high frequency is time-consuming in the laboratory because they require the efforts of experienced operators. Meanwhile, process variables such as the temperature, the pressure, and the flow rate can be readily sampled in high frequency; hence the sample size between the process and the quality data is quite unequal. To effectively integrate two different observation sources, the high-rate process measurements and low-rate quality measurements, a semi-supervised regression model with probabilistic latent variables is proposed in this article to enhance the performance monitoring of the variations of the process and the quality variables. The corresponding statistics are also systematically developed and a TE benchmark problem is presented to illustrate the effectiveness of the proposed method.

Consensus and diffusion based observers have been developed to address distributed state estimation over a sensor network. In this paper, the way sensor faults affect the state estimation of such distributed state observers is investigated. Next, a methodology to design a distributed system for sensor fault detection and isolation (FDI) is proposed. The aim is to achieve a specified fault detection (FD) performance while limiting the complexity of the algorithm at each node of the sensor network. To this end, proper selection of the measurements used for FDI at each node is performed thanks to a fault distinguishability measure based on the Kullback-Leibler divergence.

In many application domains there is an increasing trend for mixed-criticality systems with functions of different assurance levels on shared computing platforms. Today's CAN-based platforms do not support the requirements of mixed-criticality systems. A single CAN bus provides low cost, real-time support and flexibility for applications where the communication service is not safety-relevant. Fault-tolerance extensions for CAN impose incompatibility to legacy applications, high cost and overhead for the entire CAN communication. This paper introduces a CAN infrastructure for fault isolation and selective fault-tolerance, which permits a balanced trade-off between cost and fault-tolerance for each subsystem of a mixed-criticality system. We introduce replicated CAN routers that perform fault isolation based on a priori knowledge of the permitted behavior of CAN nodes. Fault masking is supported selectively through the redundant transmission of messages from safety-critical subsystems. The CAN routers perform input agreement on pending messages for replica deterministic behavior, as well as output agreement on the delivery status of messages. Software layers hide the fault-tolerance mechanisms to establish compatibility to legacy software. The benefits of the proposed communication infrastructure are demonstrated in a simulation of an example system.

Principal component analysis (PCA) and Partial least square (PLS) are powerful multivariate statistical tools that have been successfully applied for process monitoring. They are efficient in dimension reduction and are suitable for processing large amount of data. Nevertheless, their application scope is restricted to static processes where the dynamics are ignored. In order to achieve improved monitoring performance for dynamic processes, in this paper, we propose an effective dynamic monitoring scheme based on the canonical variate analysis (CVA) technique. Different from the standard PCA- and PLS-based techniques which rely on mean-extraction for residual generation, the proposed CVA-based scheme takes process dynamics into account as well. The properties of all three methods are then compared in detail and finally, the improvements of the proposed method are demonstrated on the well-accepted Tennessee Eastman benchmark process.

We propose a robust actuator fault tolerant control strategy for systems with linear parameter varying (LPV) uncertainty model description. The scheme employs a set-based robust fault detection and identification (FDI) approach and a bank of virtual actuators (VA). An interesting feature is that the virtual actuators are used both for FDI and controller reconfiguration (CR) tasks. The robust FDI method is based on the separation of relevant sets defined for measurable residual signals, which are computed based on the virtual actuator signals and taking into account model uncertainty, noises and process disturbances. For CR, each VA is designed to operate adequately in combination with a nominal controller (designed for the fault-free plant) to achieve correct reconfiguration for a particular fault situation in a considered range of fault scenarios. The resulting robust fault tolerant control scheme ensures boundedness of the closed-loop system trajectories under a wide range of actuator fault scenarios. The performance of the scheme is illustrated through a simulation example.

To avoid structural damage, a wind turbine is equipped with a safety supervisor that triggers an emergency shutdown procedure in case of internal faults or large wind gusts. This paper leverages the (compositional) barrier certificate framework for the design and optimization of such a supervisor. The problem is formulated as a sum of squares problem and is solved using semi-definite programming. Both a direct and compositional approach are successfully implemented and verified for the NREL 5MW reference turbine. In conclusion, the methods derived in this paper are indeed viable for the syntheses and optimization of safety systems for future wind turbines.

In this paper a fault detection system and a fault tolerant controller for a wind farm model is designed and tested. The wind farm model is taken from the wind farm challenge which is a public available challenge where a wind farm consisting of nine turbines is proposed. The goal of the challenge is to detect and handle different faults occurring in the individual turbines on farm level. The fault detection system is designed such that it takes advantage of the fact that within a wind farm several of the turbines will be operating under similar conditions. To enable this the turbines are grouped into several groups of similar turbines, then the turbines within each group are used to generate residuals for the turbines in the group. The generated residuals are then evaluated using dynamical cumulative sum. The designed fault detection system is cable of detecting all three fault types occurring in the model. All the detections are not within the requirement of the challenge thus room for improvement. To take advantage of the fault detection system a fault tolerant controller for the wind farm has been designed. The fault tolerant controller is a dispatch controller which is estimating the possible power at each individual turbine, using these estimates to set the reference to the turbines accordingly. The fault tolerant controller has been compared to the reference controller from the challenge. And the fault tolerant controller is better than the reference controller on all measures both under normal and faulty conditions. Thus a fault detection system and a fault tolerant controller has been designed and combined. The fault tolerant control system has then been tested and compared to the reference system and shows improvement on all measures.

This paper addresses the problem of fault isolation in processes which are working in different operating points. Due to nonlinearities and set-point changes, the statistical model which is obtained from data is different from one operating point to another. Therefore the classical multivariate statistical process monitoring approaches may not be suitable for monitoring and diagnosis purposes. For that, a data-driven fault isolation method is proposed which splits the process into several local models. Based on the local models, a probabilistic approach is proposed to determine the contribution of each variable to the fault detection index and find the risky variables which are responsible for the fault. Finally, the proposed method is demonstrated through its application on a laboratory setup of continuous stirred tank heater.

This paper deals with fault detection (FD) problem for linear discrete-time systems subject to random intermittent measurements, probabilistic actuator failures, norm-bounded model uncertainty, and stochastic model uncertainty. By taking into account the probabilistic actuator failures, a new reference residual model is proposed to formulate the FD issue as an H∞, model-matching problem. The corresponding reference residual is generated through maximizing a stochastic H–/H∞ or H∞/H∞ performance index via solving an algebraic Riccati equation (ARE). By the aid of the linear matrix inequality (LMI) techniques, a fault detection filter (FDF) is constructed such that the residual is sensitive to fault but insensitive to unknown inputs, mixed model uncertainties, random intermittent measurements and stochastic actuator failures. An illustrative example is given to demonstrate the effectiveness of the proposed method.

Reliability of the railway vehicle suspension system is of critical importance to the safety of the vehicle. It is very desirable to monitor the health condition and the performance degradation for rail vehicle suspension systems online, which offers the important information of the suspension system and it is critically important for the condition based maintenance rather than scheduled maintenance in the future. Advanced fault diagnosis method is one of the most effective means for the health monitoring of rail suspension systems. In this paper, taking the lateral suspension system as an example, the fault isolation issue for different component faults occurring in the suspension system is concerned. The sensor configuration for obtaining the state information for fault diagnosis and the mathematical model for the lateral suspension system are presented. Three different methods, Dempster-Shafer (D-S) evidence theory, Fisher Discrimination Analysis (FDA) and Support Vector Machine (SVM) techniques are applied to the fault isolation problem, respectively. Simulation study is carried out by means of the professional multi-body simulation tool, SIMPACK. The simulation results show that these methods can isolate the considered component faults effectively with a high accuracy. The proposed methods provide an effective alternative for the health monitoring of rail vehicle suspension systems.

The paper proposes a new method for a kind of parametric fault online diagnosis with state estimation jointly. The considered fault affects not only the deterministic part of the system but also the random circumstance. The proposed method first applies Kalman Filter (KF) and Maximum Likelihood (ML) technique to identify the fault parameter and employs the result to make fault decision based on the predefined threshold. Then this estimated fault parameter value is substituted into parameterized state estimation of KF to obtain the state estimation. Finally, a robot case study with two different fault scenarios shows this method can lead to a good performance in terms of fast and accurate fault detection and state estimation.

This paper deals with a multi-objective fault detection observer design problem in time and frequency domain for time invariant systems. Firstly, to improve the abilities of designed observer, different design criteria are proposed, which are evaluated by some suitable performance indices in time and frequency domain. Details about the relationships among different criteria are analyzed for two cases when the fault appears and disappears. With selected residual evaluation function and threshold, a formula with an envelope in time domain to realize fast fault detection is proposed for some typical faults. The designed observer considers the trade-off between fast transients of the residual for specified faults and the traditional criterion H – /H ∞ for general faults and disturbances. Compared with H – /H ∞ frequency design method, the effectiveness of the proposed method is demonstrated by the numerical simulation with a vehicle lateral dynamics system.

This note considers the problem of learning a warning system from observations of a system which has not encountered any error as yet. That is, can we infer a warning rule from a system remaining in normal operation regime, without making any (stochastic) assumptions? While this problem appears to be paradoxical, methods which were studied in online learning theory can be used to find such a strategy. To illustrate that this is indeed possible, the classical linear PERCEPTRON rule is reinterpretated.

A novel risk-based fault detection method has been developed. The proposed method provides a dynamic process risk indication based on the probability of happening a fault and its consequence. In this method instead of generating an alarm based on residuals or signals an alarm is activated only when the calculated risk of operation exceeds the acceptable threshold. This is an important concept as it can funnel the attention and effort of operators to the faults which poses the most operational or safety risk. Application of this new risk-based approach provides early warning of the fault as well as the associated risk with the fault. Methodologies were developed to apply the concept with model based fault detection algorithm as well as multivariate history based fault detection techniques. In this paper we show the model based approach by combining Kalman filter with the risk based approach. The history-based approach was demonstrated using principal component analysis (PCA). This method has more power in discerning between operational changes and abnormal conditions which have potential to cause accidents.

Complex technical systems, e. g. chemical plants, are prone to equipment failures. To ensure the safe operation of such systems, the occurrence of a fault has to be reliably detected. Set-based validation and identification methods are well suited for this problem as they are flexible with respect to modeling uncertainties and as they can provide guaranteed results. One of the main challenges of set-based approaches is, however, the complexity of underlying computations. Simplifying the problem formulation via a suitable approximation of the model is one way to reduce the computational effort. However, to retain the ability to diagnose faults, the underlying structure of the model has to be taken into account. We present a method to reduce the problem formulation based on causal reasoning and lifting technique that orders the system states according to the effects of occurring faults. We present an approach to derive such a reduction and illustrate its application considering two 5-tank configurations.

To ensure safe operation of technical processes, faults have to be reliably detected and isolated to provide information for process maintenance, shutdown, or reconfiguration. Fault detection and isolation can be achieved by invalidation of fault candidates, i.e. models of the system in fault-free and faulty condition. In order to enhance the performance of fault detection and isolation, so-called active approaches use input signals with the objective that the resulting system outputs are consistent with at most one fault candidate. Guaranteeing or analyzing robustness of active fault diagnosis with respect to input, output, and process uncertainties and nonlinearities is challenging. This paper provides certificates of robustness of input sequences with respect to the aforementioned uncertainties and nonlinearities. The certificates enable the determination of input and output uncertainties for which unique fault diagnosis results can still be guaranteed. In addition, a method is presented to select a minimal number of outputs that still guarantee robust fault diagnosis, thus reducing the measurement setup and cost. The approach employs nonlinear mixed-integer feasibility problems and a relaxation framework and does not require the explicit computation of reachable sets. The approach is applicable to polynomial discrete-time systems and is demonstrated for a numerical example.

For process fault detection and diagnosis, a real time hybrid method based on Principle component analysis (PCA) and Bayesian belief network (BBN) is described. Upon successful identification of fault from PCA residual plot and Q statistics, information from the PCA contribution of each variable is passed to the BBN for root cause analysis. Pearl's message passing algorithm is used for belief updating. Early detection of fault, makes the methodology more reliable and robust during the process fault occurrence. The aim of this monitoring tool is to incorporate prior process knowledge along with the present observed evidence to come up with most plausible explanation of how the process is behaving. The effectiveness of the proposed method is demonstrated for a Dissolution tank model for different simulated scenarios by detecting and diagnosing the fault accurately.

In this article, a novel dynamic Bayesian network based networked process monitoring approach is proposed for fault detection, propagation pathway identification and root cause diagnosis. First, process network structure is designed according to the prior process knowledge including process flow sheets and used to characterize the causal relationships among different measurement variables. Then, the dynamic Bayesian network model parameters including the conditional probability density functions of different nodes are learned from historical process data to quantify the causality among those variables. Further, the new monitoring index is derived from the likelihoods of the entire process network for detecting abnormal operating events. With the captured process abnormality, the novel probabilistic contribution indices within Bayesian network are proposed to identify the major fault effect variables. Subsequently, the fault propagation pathways from the downstream backwards to upstream process are isolated through the variable contribution indices and hence the ending nodes of the identified pathways are determined as the root-cause variables of the abnormal events. The proposed approach is applied to the Tennessee Eastman Chemical process and the results show that the presented method can accurately detect abnormal events, identify fault propagation pathways, and diagnose the root-cause variables.

This paper describes the application of batch trajectory alignment, outlier detection, and multiblock multiway principal component analysis (MPCA) to data from an industrial active pharmaceutical ingredient manufacturing process. The process data routinely collected from historical batches, including temperatures, pressures, and controller outputs, has been used to improve process operation and understanding. MPCA highlighted questionable batches from which plant issues were identified. Variable contributions to the MPCA scores were used to identify the process variables potentially causing the variation in batch drying time.

A model based fault detection and diagnosis scheme is proposed in this paper and a lumped parameter MIMO model has been considered. The characteristics of single as well as multiple faults using residuals are investigated in air-conditioning (AC) system provided in passenger coach of an Indian Railways. The residuals of the fault are generated by using real life simulator and model. The faults are diagnosed by classifying residual patterns. The classifications of residual patterns have been done by using three approaches: hexadecimal decision table, least square support vector machine (LS-SVM) and a novel approach named as hybrid classification approach.

To achieve worry-free production and prevent unexpected downtime, predictive maintenance has become a prevailing strategy, supported by prognostics and health management techniques. In this paper, a systematic data-driven approach for intelligent maintenance is discussed, as well as an embedded system architecture for fault detection and prediction. Two case studies are presented, including performance prediction for wire-twisting machine based on temperature data, and an embedded solution for electric valve actuator monitoring.

This paper proposes a Distributed Fault-Detection System for fault detection and prediction in electrical actuators used in pipelines for oil and gas transportation. The proposed system incorporates a signal processing flow that requires low complex mathematical operations using C-ANSI language. For the embedded Fault-Detection System was created a network of real-time on the RS-485 physical layer to application in electric actuators. The system is composed of Master and Slave modules using the Modbus protocol in order to communicate with other devices. To prove its functionalities a testbench was developed, which aims to reproduce in a lab some common faults and degradation processes that may occur in real world field applications. The monitoring of the electric actuators operation through this system can identify problems and faults before they become severe, providing greater efficiencies on the production line and avoid expressive financial losses. The prototype developed in this paper is based on software and hardware platforms flexible and open-source. The slave module equipment was used to collect the sensors information from specific points of the actuator. The sensor data collected and emulation were used to validate the propose fault detection methodology.

In this paper a method for designing the input of a state affine system in order to guarantee a pre-specified degree of observability is proposed. It consists of an off-line algorithm to build an optimal, and so-called regularly persistent, input for the system. Optimality is defined with respect to the minimization of the input energy, and observability through a lower bound for the observability Gramian of the system, which is known to allow for an exponential observer design. The approach is presented for discrete-time systems, and is illustrated in simulation by an application example corresponding to a problem of fault detection in pipelines.

A space mission has to be primarily economical, save and reliable. To ensure these aspects, the system has to react in a timely manner on occurring faults or unforeseen problems. In many cases the operators on the ground can only react delayed in time due to great distances and resulting signal delays and therefore, the satellite has to enter into a predefined, safe configuration and looses valuable operational time. This forces the developer to place more autonomy on-board. This paper uses a cognitive network to describe the satellite for the innovative Fault Detection and Isolation and Recovery program (iFDIR). It digitizes the physical relations of a satellite's systems to find faults in sensors and actuators. The fault detection and isolation should be improved by combining already working algorithms with the physical and logical regularity. Furthermore, the algorithm shall be able to adapt itself to new circumstances by adapting the network and the system parameters.

Systems, such as nuclear power plants, chemical processes and means of transportation, are typically seen as critical, as their failure may endanger lives and assets. The safety of such systems is, therefore, rigorously considered and established during the design and operational stages. In the design stage, an off-line safety analysis investigates, retrofits and affixes whenever necessary fault-tolerant means and reliable components. In the operational stage, the functionality of systems is monitored through three safety tasks: fault detection and diagnosis, alarm annunciation and fault controlling. However, systems still showing malfunctions and hazardous failures continue to be recorded. To address this issue, this paper develops a distributed on-line safety monitor. The monitor aims to achieve an effective integration among the delivery of the three safety tasks through the exploitation of a thorough and cost-effective off-line safety analysis model and the distributed reasoning of a multi-agent system.

This paper presents a fault detection H ∞ filter design for uncertain linear systems in the presence of time-varying delays. The methodology is based on the computation of a reference threshold from the residue generated between the states of the system and the filter estimates such that when the residue takes values higher than this reference threshold the uncertain system is considered in fault. The main difference between the proposed method and other ones in the literature is that the filter design considered in this paper takes into account the average of the time-varying delay estimation. Therefore, the proposed methodology does not require online measurement of the time-delay, avoiding increase the implementation difficulty and the filter cost. Numerical simulations illustrate the effectiveness of the proposed strategy.

This paper deals with the problem of observer-based fault diagnosis for a Simulated Moving Bed (SMB) chromatographic process. A fault detection and isolation (FDI) system is designed such that the sensor faults are detected and located. The residual generation is based on a nonlinear varying-parameters observer scheme. The question of stability is addressed in terms of Lyapunov quadratic stability and sufficient conditions are obtained through strict linear matrix inequalities formulation (LMIs).

The conventional statistical process monitoring methods usually require that the measurement data on all variables must be present at all times. In practice, however, most large process data sets contain missing data. In this paper, a new method called robust non-negative matrix projection (RNMP) based on the concept of incomplete process data, is proposed to detect and diagnose the faults in industrial process. Non-negative matrix factorization (NMF) is a widely-used method for low-rank approximation where the non-negativity constraints are imposed on factor matrices in the decomposition. The proposed RNMP method is a new variant of NMF based on positively constrained projections, which can obtain an accurate prediction of the missing values in the data set. Then, we use RNMP to extract the latent variables from incomplete data that drive a process and to combine them with process monitoring techniques for fault detection and isolation. Afterwards, the proposed method is applied to the continuous-stirred tank reactor (CSTR) process to evaluate the monitoring performance. The experiment results clearly illustrate the feasibility of the proposed method.

This paper concerns fault detection and diagnosis for discrete event systems modeled with partially observed Petri nets. The proposed method provides a decision via the analysis of observation sequences that include some observable events and the partial measurement of the successive states visited by the system. To this end, the observation sequences are decomposed into elementary observation sequences, linear matrix inequalities are used to compute the firing sequences consistent with each elementary observation sequence and an algorithm of linear complexity with respect to the length of the observation sequences is proposed to return on-line diagnosis decisions.