Case-based expert system using wavelet packet transform and kernel-based feature manipulation for engine ignition system diagnosis

Abstract

Whenever there is any fault in an automotive engine ignition system or changes of an engine condition, an automotive mechanic can conventionally perform an analysis on the ignition pattern of the engine to examine symptoms, based on specific domain knowledge (domain features of an ignition pattern). In this paper, case-based reasoning (CBR) approach is presented to help solve human diagnosis problem using not only the domain features but also the extracted features of signals captured using a computer-linked automotive scope meter. CBR expert system has the advantage that it provides user with multiple possible diagnoses, instead of a single most probable diagnosis provided by traditional network-based classifiers such as multi-layer perceptions (MLP) and support vector machines (SVM). In addition, CBR overcomes the problem of incremental and decremental knowledge update as required by both MLP and SVM. Although CBR is effective, its application for high dimensional domains is inefficient because every instance in a case library must be compared during reasoning. To overcome this inefficiency, a combination of preprocessing methods, such as wavelet packet transforms (WPT), kernel principal component analysis (KPCA) and kernel K-means (KKM) is proposed. Considering the ignition signals captured by a scope meter are very similar, WPT is used for feature extraction so that the ignition signals can be compared with the extracted features. However, there exist many redundant points in the extracted features, which may degrade the diagnosis performance. Therefore, KPCA is employed to perform a dimension reduction. In addition, the number of cases in a case library can be controlled through clustering; KKM is adopted for this purpose. In this paper, several diagnosis methods are also used for comparison including MLP, SVM and CBR. Experimental results showed that CBR using WPT and KKM generated the highest accuracy and fitted better the requirements of the expert system.

Introduction

Automotive engine ignition systems vary in construction, but are similar in basic operation. All have a primary circuit that causes a spark in the secondary circuit. This spark must then be delivered to the correct spark plug at the proper time. Conditions in the ignition system and in the cylinder affect the ignition pattern (i.e. scope pattern) in the secondary circuit. An automotive oscilloscope, also known as a scope meter, is considered a valuable tool for detecting engine and ignition problems by displaying the scope patterns for analysis of the operation of an ignition system. The scope patterns can reflect the conditions within the ignition system and help pinpoint their causes of failure, such as narrow spark-plug gaps, open spark-plug cables, a shorted ignition coil, etc. In a typical diagnosis, the scope patterns are usually matched against a normal pattern. Fig. 1 shows some examples of scope patterns and their corresponding engine faults.

A pattern pickup-clamp for an automotive scope meter is necessary when an automotive scope meter for engine trouble diagnosis is used. The pickup-clamp is usually connected to the ignition system to capture its spark ignition patterns. Capturing ignition pattern usually requires intervention of the mechanic to indicate the start and end points of the pattern. The captured pattern would then be compared with samples from the handbooks (Liu and Chu, 2005, Crouse and Anglin, 1993) for diagnosis. The diagnosis is always based on domain knowledge and user experience because the samples from the handbooks are for reference only. However, the ignition patterns are time-varying and non-stationary. Different engine models produce various amplitude and duration for the same ignition system trouble. Even on the same engine, different sizes of ignition patterns under different engine operating conditions may be produced, which increases the difficulty for human diagnosis. Moreover, the ignition patterns of many engine faults are very similar, leading to increased difficulty in identifying the patterns correctly. After diagnosis, the corresponding parts in the ignition system will be disassembled for intensive investigation (this is called a trial). In view of the inaccuracy of human diagnosis, identifying a fault based on ignition patterns, several trials for disassembling and assembling of engine parts are necessary, which incurs a large amount of time and effort by the mechanic. A computer-based pattern classification system is proposed to aid an automotive mechanic for this problem.

Currently, there is very little research in the literature (Vong and Wong, 2011) on computer-aided ignition pattern analysis for automotive engine ignition system trouble-diagnosis. The development of an intelligent system for engine ignition system diagnosis is a promising area of research.

Machine learning methods (Xue and Zhu, 2009, Kotsiantis, 2007, Whiteson and Whiteson, 2009) play an important role in the pattern recognition, such as multi-layer perceptions (MLP) and support vector machines (SVM). A major focus of machine learning research is to automatically learn recognition of complex patterns and make intelligent decisions based on sampled data. In this research, if the computer-aided diagnostic system is designed merely based on captured signals, the accuracy will likely be poor due to the excessive similarity between the captured signals. Applying this rationale, machine learning methods combined with domain knowledge are proposed in this research. However, most of the machine learning methods can only provide a user with a single solution. To increase the chance of locating the faults, a case-based reasoning (CBR) expert system is proposed, which can generate multiple diagnoses, instead of the single most probable one generated by traditional network-based classifiers such as MLP and SVM (Fig. 2). This kind of CBR expert system helps a mechanic make a final decision. From this viewpoint, CBR paradigm fits the requirements of the expert system better. In addition, CBR can overcome the problem of incremental and decremental knowledge update as compared to inductive learning (Smyth and Cunningham, 1995) such as MLP and SVM. Moreover, when adding or removing a case, the maintenance of CBR classifier is relatively easy.

One of the main difficulties of building a CBR diagnostic system is to define the clear and effective case representation in the case library. In the current research, it is unwise to directly put the captured signals in the case representation. The ignition patterns captured by the scope meter are very similar and their differences can only be compared under extracted features. Similar work extracting effective representation from sensor signals can be found in (Funk and Xiong, 2008) and hence wavelet packet transform (WPT) is proposed for the purpose of feature extraction. However, the number of the extracted WPT features for an ignition signal is very large (about 18,000 points) so that the computational cost of building and running the diagnostic system will be very high. For this reason, kernel principal component analysis (KPCA) is proposed to perform a dimension reduction while retaining most of the pattern information of the ignition signals. Finally, the set of features processed by WPT and KPCA along with the domain knowledge of the signals constitutes the case representation of the CBR diagnostic system. CBR is effective but mostly inefficient in computational time because every instance in a case library must be compared during its reasoning, i.e., retrieval. To overcome this inefficiency, a clustering method of kernel k-means (KKM) was employed to compress the case library, where similar cases were clustered into a fixed number of representative cases. Using this set of representative cases, retrieval time can be further reduced significantly. In the following section, the employed techniques are briefly discussed.