Elsevier

Information Sciences

Volume 259, 20 February 2014, Pages 234-251
Information Sciences

Fault detection and isolation of a dual spool gas turbine engine using dynamic neural networks and multiple model approach

https://doi.org/10.1016/j.ins.2013.05.032Get rights and content

Abstract

In this paper, a fault detection and isolation (FDI) scheme for an aircraft jet engine is developed. The proposed FDI system is based on the multiple model approach and utilizes dynamic neural networks (DNNs) to accomplish this goal. Towards this end, multiple DNNs are constructed to learn the nonlinear dynamics of the aircraft jet engine. Each DNN corresponds to a specific operating mode of the healthy engine or the faulty condition of the jet engine. Using residuals obtained by comparing each network output with the measured jet engine output and by invoking a properly selected threshold for each network, reliable criteria are established for detecting and isolating faults in the jet engine components. The fault diagnosis task consists of determining the time as well as the location of a fault occurrence subject to presence of unmodeled dynamics, disturbances, and measurement noise. Simulation results presented demonstrate and illustrate the effectiveness of our proposed dynamic neural network-based FDI strategy.

Introduction

For aircraft jet engine fault diagnosis problem one utilizes knowledge of the measured variables taken along the engine’s gas path to determine how the engine performance differs from its desired state. Changes in the engine speed, temperature, pressure, fuel flow, etc., provide the required information for identifying the engine system malfunctions. The overall goal of a jet engine diagnostic system is to correctly detect, isolate and identify the changes in the engine parameters. This will prevent heavy economic losses due to stopped or aborted flights as well as the cost associated with untimely and unnecessary replacement of components and parts.

Fault diagnosis algorithms are mainly divided into two categories, namely model-based and data-driven (computational intelligence-based) techniques. Both techniques have been extensively studied in the literature for health monitoring of aircraft jet engines. In model-based approaches, Kalman filters are quite popular [5], [11], [10]. Although, model-based techniques have advantages in terms of on-board implementation considerations, their reliability often decreases as the system nonlinear complexities and modeling uncertainties increase.

Data-driven approaches, such as those based on neural networks, mostly rely on real-time or collected historical data from the engine sensors and do not require a detailed mathematical model of the system. Neural networks are a promising tool for fault diagnosis due to their proven success in system identification and strong capability in learning nonlinear transformations that map a set of inputs to a set of outputs.

Applications of neural networks in engine fault diagnosis have been widely presented in the literature. The authors in [17] have proposed a modular diagnostic system for a dual spool turbofan gas turbine. Multiple neural networks are proposed in [4] for fault diagnosis of a single shaft gas turbine. The author in [12] has further extended the multiple neural networks method to generate a cascade network to isolate component and sensor faults. The author in [3] has discussed the need to incorporate a neural network with other AI techniques to estimate the remaining useful life and incorporate prognosis capabilities of an engine. The authors in [14] have applied a probabilistic neural network (PNN) to diagnose faults on turbofan engines.

Dynamic multilayer perceptron or dynamic neural networks have recently been utilized for system identification problems due to their capabilities in modeling nonlinear dynamical systems. Such networks have a feedforward architecture and their dynamic properties are achieved by using dynamic neurons. Each neuron possesses dynamic characteristics that is achieved through a locally recurrent globally feedforward (LRGF) mechanism [16], [2].

Recently, dynamic neural networks have been utilized for fault diagnosis of nonlinear systems. The authors in [15] have used a multilayer perceptron network that is embedded with dynamic neurons for fault detection and isolation (FDI) of thrusters in the formation flight of satellites. A dynamic neural network is constructed in [1] for accomplishing the fault detection task. This is followed up by a static neural classifier using the learning vector quantization (LVQ) scheme for the fault isolation task. The authors in [9] have applied the dynamic neural network that was developed in [16] for fault detection of aircraft jet engines.

In the present work a dynamic neural network-based multiple model strategy is proposed and developed as a novel approach for fault detection and isolation of aircraft jet engines. It will be shown that by using a bank of dynamic neural networks the problem of fault detection and isolation of a dual spool jet engine can be addressed quite effectively. For comparison of our developed methodology with another and an alternative dynamic neural network-based approach that tackles the similar problem and application, the reader can refer to our work in [9]. These details are not included here for sake of brevity.

The remainder of this paper is organized as follows. In Section 2, the dynamic neuron model and the proposed dynamic neural network structure are described. The overall model of the considered dual spool jet engine is presented in Section 3. In Section 4 the problem of fault diagnosis, that is the fault detection and isolation strategy is proposed and described in detail. Simulation results are presented in Section 5 followed by the conclusions in Section 6.

Section snippets

Dynamic neural network (DNN) architecture

Dynamic neural networks (DNNs) presented in [16], [2] have great capabilities in learning the dynamics of complex nonlinear systems, where conventional static neural networks cannot yield an acceptable representation. Dynamic neural networks or dynamic multilayer perceptron networks (MLP) represents an extension of static neural networks by embedding discrete or continuous time dynamics to the neuron model. Such an extension enhances the capability of the resulting neural networks to

Aircraft jet engine model

Based on the available literature [9] on modeling of aircraft jet engines, a Matlab Simulink model for a dual spool engine is developed. Rotor and volume dynamics are considered in order to obtain a nonlinear dynamics for the system. In order to take into account the volume dynamics, the engine components are considered to model an imbalance mass flow rate. This results in elimination of large algebraic loops in the model. The detailed mathematical expressions corresponding to the overall

Dynamic neural network FDI methodology

In this section, a fault diagnosis scheme, that is a fault detection and isolation methodology for a dual spool gas turbine engine is developed. Towards this end, a dynamic neural network-based multiple model scheme is proposed in which a bank of dynamic neural networks acts as an estimator or identifier of different engine operating conditions corresponding to the various but limited number of faulty modes that are of most interest or possible in the jet engine.

The dynamic neural network-based

Simulation and FDI developed results

The implementation of our proposed dynamic neural network-based fault detection and isolation scheme consists of three main tasks, namely (i) System Identification, (ii) Fault Detection, and (iii) Fault Isolation. The implementation of these steps and the case studies considered for validation of our claims are explained and illustrated in detail below.

Conclusions

In this paper, a dynamic neural network-based multiple model scheme is proposed for achieving fault diagnosis, that is fault detection and fault isolation of aircraft jet engines. Several banks of dynamic neural networks are trained, where each network corresponds to and is associated with a specific faulty and/or healthy mode of the aircraft jet engine. Our fault detection and isolation results are summarized in Table 5, Table 6. One of the future directions of research will be on the

Acknowledgement

This work was made possible by the NPRP Grant # 4-195-2-065 from Qatar National Research Fund (a member of Qatar Foundation). The statements made herein are solely the responsibility of the authors.

References (17)

  • A. Yazdizadeh et al.

    Adaptive time delay neural network structures for nonlinear system identification

    Neurocomputing

    (2002)
  • I.A.-D. Al-Zyoud, K. Khorasani, Detection of actuator faults using a dynamic neural network for the attitude control...
  • M. Ayoubi, Fault diagnosis with dynamic neural structure and application to a turbocharger, in: Proc. Int. Symp. Fault...
  • A. Green, D. Allen, Artificial-intelligence for real-time diagnostics and prognostics of gas-turbine engines, in: 33rd...
  • K. Kanelopoulos, A. Stamatis, K. Mathioudakis, Incorporating neural-networks into gas-turbine performance diagnostics,...
  • T. Kobayashi, D.L. Simon, Application of a bank of kalman filters for aircraft engine fault diagnostics, in: Proceeding...
  • J. Korbicz et al.

    Dynamic neural networks for process modelling in fault detection and isolation systems

    International Journal of Applied Mathematics and Computer Science

    (1999)
  • K. Mathioudakis et al.

    Probabilistic methods for gas turbine fault diagnostics

There are more references available in the full text version of this article.

Cited by (148)

  • Data-driven fault detection and isolation of nonlinear systems using deep learning for Koopman operator

    2023, ISA Transactions
    Citation Excerpt :

    Various data-driven methods have been presented in the literature investigating fault detection and isolation. Among those, statistical methods such as Principle Component Analysis (PCA) [6] and intelligent computational methods such as Artificial Neural Networks (ANN) [7] have been the most well-known approaches for their efficacy and have been widely used in many applications [8]. A reduced kernel PCA method is presented in [9] for fault detection and diagnosis in cement rotary kiln.

  • Protocol-based fault detection for discrete-time memristive neural networks with quantization effect

    2022, Information Sciences
    Citation Excerpt :

    For example, a dc FD has been utilized for high voltage direct current grids in [28], and a decentralized FD method has been applied to detect single phase for power distribution systems [29]. By generating a residual signal to detect the fault, FD is seen as an effective technique in improving the reliability of systems and reducing operational risk [30]. Note that in recent decades, FD problem has received ever-increasing attention in various control communities, such as Markov jump systems [31], fuzzy systems [32], linear-parameter-variable systems [33], and nonlinear systems [34].

View all citing articles on Scopus
View full text