Flexible software for condition monitoring, incorporating novelty detection and diagnostics

Abstract

Condition monitoring and machinery fault diagnosis are central to the implementation of efficient maintenance management strategies. They can be based on empirical modelling, which aims at associating measured data to machine conditions. Arguably, different monitoring tasks present different challenges to the maintenance engineer. This paper presents the development of a flexible software solution for condition monitoring, novelty identification and machinery diagnostics, which can easily be customised to a wide range of monitoring scenarios. Its main constituents are a number of independent software modules, such as the fault and symptom tree, the fuzzy classification module, the novelty detection and the neural network diagnostics sub-systems. It is implemented on two different applications, namely machine tool monitoring and gearbox monitoring.

Introduction

Condition-based maintenance (CBM) seeks to implement a policy wherein maintenance management decisions are based on the identification of the current condition of monitored machinery. The implementation of efficient maintenance management strategies based on CBM presupposes that adequate condition monitoring, as well as machinery fault diagnostics and prognostics are in place [1]. Reliable prognostics itself relies on condition monitoring and fault diagnostics, as prediction of the future state of machinery implies that the current state of machinery is known. Therefore, condition monitoring and fault diagnosis are central to the implementation of efficient maintenance management strategies. While their importance is evident, reliable automated identification of machinery condition is not straightforward, as machinery malfunction demonstrates itself with different signal patterns, depending on the particular kind of problem encountered. These patterns of behaviour can vary considerably even for the same fault types on similar equipment [2], [3], [12].

Diagnostic systems for industrial machinery can benefit from employing empirical models, such as statistical (Bayesian, nearest neighbours, time series), polynomial, experts systems, neural network, evolutionary computation, fuzzy and neurofuzzy classification methods [1], [4], [5], [6], [7], [8]. Despite past successes, diagnostic systems utilising empirical modelling are usually tailored to very specific tasks and cannot easily be customised for other applications. Part of the difficulty is that the challenges posed by different monitoring tasks may require very different modelling approaches. A potential solution would be to have a decision support system making the choice of the most adequate diagnostic technique to employ for each different application [9]. Alternatively, it would be desirable to have diagnostic systems, which can easily be customised to serve different tasks.

In this paper the development of a flexible and generic system for novelty detection and diagnostics, which can be customised so that it can be applied to a wide range of monitored machinery, is reported. The objective was to develop a system capable of performing novelty detection and fault diagnosis, even in cases when very little a priori knowledge exists about the task in hand. In doing so, it was important to recognise that the aim is not only to be able to diagnose faults when they have already occurred but also to have the capability to provide warnings when a fault is still at its early stages. Central to this development has been the acoustic emission technique, as it has a number of advantages in terms of its high signal to noise ratio and wide applicability, while in many cases it can exhibit sensitivity to early progression of faults. Other sensorial information is also incorporated in the cases of gearbox and machine tool monitoring. Nonetheless, the developed system is not restricted to these particular set-ups and with adequate customisation it can be tailored to other kind of monitoring tasks.

This paper is structured as follows. Section 2 presents the architecture of the developed system and outlines the functionality of the main system modules. Section 3 presents in more detail the concept and functionality of the fault tree module. Next, in Section 4, the fuzzy classification module is discussed. The novelty detection and neural network diagnostics principle of operation and functionality is provided in Section 5. Data handling is based on a central database structure, described in Section 6. The system was tested on some real life condition monitoring test cases and indicative examples are discussed in Section 7. The main conclusions drawn from the work described in this paper are summarised in the last section, including a brief discussion on potential future extensions of the work.