Fault detection for time-variant avionics systems based on a new data-driven time-varying approach

One of the greatest drawbacks in wind energy generation are the high maintenance costs associated to mechanical faults. In order to reduce these impacts have been integrated fault detection system in wind turbines, known as FDD's (‘Fault detection and Diagnosis System’). The approach to the development of FDD systems presented is known as ‘Data-Driven’ (FDD-DD) which involves the use of collections of data from a monitoring system for building models of classification/regression. The aim of this paper is to perform a comparative analysis of different techniques: decision trees, bayesian classification, neural networks and support vector machines applied to fault detection systems in wind turbines. The results indicate that support vector machines bi-class gets a fairly high level of accuracy like Bayesian classifiers.

Problems related to the integrated design of robust fault detection (FD) systems are studied. Firstly, a problem of the 2-norm based approaches is discussed, i.e. due to the necessary length of the time window introduced to realize the 2-norm based residual evaluation function of a FD system, the expected optimal performance of the FD system may not be achieved when the system is realized in real applications. Then a new kind of evaluation function based on the absolute value of the wavelet transform (WT) of the residual signal and the corresponding integrated design approach for the FD systems are proposed. Different from the usual 2-norm based approaches whose mathematical basis is the relationship between the whole energy of the output and input signal of a dynamic system, a relationship between the instant power of the WT of the output signal and the energy of the past input signal of a dynamic system is established and further used for the FD system design. It solves the problem in the 2-norm based method, and the resulting FD system is easy to implement, and can detect faults very rapidly with a low miss detection ratio.

Plant Fault Diagnosis System using Negative Selection Algorithm

In this article, an integrated design diagram for a stable kernel representation (SKR)-based data-driven fault detection (FD) system and performance criteria is proposed for stochastic dynamic systems in the probabilistic sense. A new distributionally robust FD system is developed using input and output data in the absence of a system model and perfect probability distributions for noises and random faults. To be specific, an SKR-based data-driven primary residual generator is first constructed. By introducing the so-called mean-covariance based ambiguity sets, families of probability distributions of the primary residual in fault-free and the concerned multiple faulty cases are characterized. The FD system design is then formulated as a distributionally robust optimization problem in the sense of minimizing the missed detection rate (MDR) with a predefined upper bound of false alarm rate (FAR). With the aid of worst-case conditional value-at-risk, a matrix-valued distribution independent solution to the targeting FD problem is derived without posing specific distribution assumptions. The developed FD system is, thus, robust against the distributional uncertainties of noises and random faults. Simultaneously, a tighter upper bound of MDR for an identical FAR criterion is achieved in comparison with the vector-valued distributionally robust FD method. An experimental study on a laboratory setup of a three-tank system shows the applicability of the proposed method.

International audience

In this paper, a technique of merging typical process data with variables containing fast periodic oscillations is proposed for the purpose of detecting faults in industrial systems working under variable operating conditions. Analysing windows of the fast-oscillating signals allowed key features to be extracted from the data at the same rate at which the process variables are sampled. This allows the fusion of both types of data acquired at different sampling rates in a single data matrix. The data is then analysed using canonical variate analysis (CVA) looking for deviations in any parameter that can point at a fault in the system. The dynamic characteristics of CVA allow the detection and diagnosis of faults in systems working under variable operating conditions. This approach was tested using experimental data acquired from a compressor test rig where the compressor surge process fault. Results suggest that the combination of both types of data can effectively improve the detectability of faults in systems working under variable operating conditions.

At the era of big data, industrial plants generate significant scale of data which contain plenty information of operation. Data-driven approach is becoming increasingly popular in the research of fault diagnosis. In this paper, a novel data-driven fault detection approach named implicit model approach (IMA) is proposed for typical nonlinear system. This approach is inspired by the subspace relevant method which is widely applied in system identification. Residual estimator, which is the key for fault detection, is generated by subspace relevant method in IMA without any prior knowledge of system. Nonlinear system in this paper is formulated with Takagi-Sugeno (T-S) fuzzy model. Every subsystem of T-S fuzzy model is presented as linear timeinvariant model. IMA is used to obtain sub-residual estimator for each subsystem. The overall residual estimator is constructed with a kind of method named parallel distributed compensation (PDC), based on which the fault detection algorithm for typical nonlinear system is generated. The proposed algorithm is validated on a simulating platform of a compact reactor. The simulated results show that the algorithm based on IMA is effective for fault detection of the typical nonlinear system in nuclear plant.

This paper presents an algorithm for fault detection of a sewage heat pump system by designing multi-mode principal component analysis with Gaussian mixture model. If the heat pump system fails, the loss of energy and time is enormous, therefore the fault detection of the system is important. For this purpose, this study proposes a fault detection method using multi-mode principal component analysis with Gaussian mixture model. The data were clustered into multi-mode of Gaussian on principal component subspace. Based on the multi-model, the values of Hotelling’s T2 and SPE were calculated and used for the fault detection as indexes that are compared performance with clustering model using k-means and k-medoids algorithm as well as conventional PCA. Actual data of the sewage heat pump were used to verify the proposed method. The results of the fault detection performance show that the proposed model shows the best performance of fault detection among the conventional, k-means, and k-medoids PCA models.

This paper considers fault detection (FD) for large-scale systems with many nearly identical units operating in a shared environment. A special class of hybrid system mathematical models is introduced to describe such multi-unit systems, and a general approach for estimation and change detection is proposed. A novel FD algorithm is developed based on estimating a common Gaussian-mixture (GM) distribution for unit parameters whereby observations are mapped into a common parameter-space and clusters are then identified corresponding to different modes of operation via the expectation-maximization (EM) algorithm. The estimated common distribution incorporates information from all units and is utilized for FD in each individual unit. The proposed algorithm takes into account unit mode switching and parameter drift and can handle sudden, incipient, and preexisting faults. It can be applied to FD in various industrial, chemical, or manufacturing processes, sensor networks, and others. The second part of the paper is focused on the application of the new technique to practical heating, ventilation, and air-conditioning (HVAC) systems. Reliable and timely FD is a significant and still open practical problem in the HVAC industry, and, as such, the first application of this approach is aimed at this industry. It addresses important details of the algorithm's implementation and presents results from an extensive performance study based on both Monte Carlo simulations and real data collected from three operational large HVAC systems. The results demonstrate the capabilities of the new technique in a more realistic setting and provide insights that can facilitate the design and implementation of practical FD for systems of similar type in other industrial applications.

Problems related to the integrated design of robust fault detection (FD) systems are studied. First, it is revealed that due to the time window introduced to realize the 2-norm based evaluation function, an optimal design of a FD system with the 2-norm based evaluation function may not ensure the expected optimal performance when the system is realized in real applications. To solve this problem, an integrated design method of FD systems using the absolute value of residual signal as evaluation function is then proposed. It leads to a residual generator which is much easier to be realized. Different from the usual 2-norm based approaches whose mathematical basis is the relationship between the energy of the output and input signals of a dynamic system, a relationship between the instant power of the output signal and the energy of the past input signal of a dynamic system is established and further used for FD system design. Another new kind of evaluation function based on the absolute value of wavelet transform of residual signal and the corresponding integrated design approach for FD systems are further proposed.

A review of computing-based automated fault detection and diagnosis of heating, ventilation and air conditioning systems

The air conditioning (AC) system is the primary building end-use contributor to the peak demand for energy. The energy consumed by this system has grown as fast as it has in the last few decades, not only in the residential section but also in the industry and transport sectors. Therefore, to combat energy crises, urgent actions on energy efficiency should be taken to support energy security. Consequently, the faults in AC system components increase energy consumption due to the degradation of the system’s performance and the losses in the energy conversion procedure. In this work, AC system fault detection and diagnosis (FDD) methods are investigated to propose analytic tools to identify faults and provide solutions to those problems. The analysis of existing work shows that data-driven approaches are more accurate for both soft and hard fault detection and diagnosis in AC systems. Therefore, the proposed methods are not accurate for simultaneous fault detection, while in some works, authors tested the method with several faults separately without investigating scenarios that combine more than one fault. Moreover, this study shows that integrating data-driven approaches requires deploying an optimal sensing and measurement architecture that can detect a maximum number of faults with minimally deployed sensors. The new sensing, information, and communication technologies are discussed for their integration in AC system monitoring in order to optimize system operation and detect faults.

This article is concerned with data-driven realization of fault detection (FD) for nonlinear dynamic systems. In order to identify and parameterize nonlinear Hammerstein models using dynamic input and output data, a stacked neural network-aided canonical variate analysis (SNNCVA) method is proposed, based on which a data-driven residual generator is formed. Then, the threshold used for FD purposes is obtained via quantiles-based learning, where both estimation errors and approximation errors are considered. Compared with the existing work, the main novelties of this study include: 1) SNNCVA provides a new parameterization strategy for nonlinear Hammerstein systems by utilizing input and output data only; 2) the associated residual generator can ensure FD performance where both the system model and its nonlinearity are unknown; and 3) with consideration of modeling-induced errors, the quantiles are invoked and used to provide a reliable FD threshold in situations where only limited samples are available. Studies on a nonlinear hot rolling mill process demonstrate the effectiveness of the proposed method.

This paper deals with the problem of robust fault detection for a class of linear discrete time-varying (LDTV) systems subject to multiple packet dropouts. Under the assumption of data packets being time-stamped, the system under consideration can be described by an augmented LDTV systems without packet dropouts and, based on this, a parity space approach to fault detection of linear time invariant systems is extended to handle the problem of fault detection of the LDTV systems. Recursive algorithms are provided to reduce the computations of parameter matrices. A numerical example is given to show the effectiveness of the proposed method.

In this paper we study fault detection in systems that can be modeled as finite state machines (FSMs). We aim at detecting faults that manifest themselves as permanent changes in the next-state transition functionality of the FSM. Fault diagnosis is performed by an external observer/diagnoser that functions as an FSM and which has access to the input sequence applied to the system but has only limited access to the system state or output. In particular, we assume that the observer/diagnoser is only able to obtain partial information regarding the state of the system at irregular time intervals that are determined by certain synchronizing conditions between the system and the observer/diagnoser. By adopting a probabilistic framework, we analyze ways to optimally choose these synchronizing conditions and develop adaptive strategies that achieve a low probability of aliasing, i.e., a low probability that the external observer/diagnoser incorrectly declares the system as fault-free.