Skip to main content
main-content

Über dieses Buch

Condition Monitoring Using Computational Intelligence Methods promotes the various approaches gathered under the umbrella of computational intelligence to show how condition monitoring can be used to avoid equipment failures and lengthen its useful life, minimize downtime and reduce maintenance costs. The text introduces various signal-processing and pre-processing techniques, wavelets and principal component analysis, for example, together with their uses in condition monitoring and details the development of effective feature extraction techniques classified into frequency-, time-frequency- and time-domain analysis. Data generated by these techniques can then be used for condition classification employing tools such as:

fuzzy systems; rough and neuro-rough sets; neural and Bayesian networks;hidden Markov and Gaussian mixture models; and support vector machines.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Introduction to Condition Monitoring in Mechanical and Electrical Systems

Abstract
This chapter reviews condition monitoring techniques in mechanical and electrical systems. The condition monitoring domain in which the data is visualized is discussed and in particular the time, modal, frequency, and time-frequency domains. The generalized condition monitoring framework which includes the data acquisition device, data analysis device, feature selection device, and decision making device is also presented. Techniques for using these decision making devices are introduced. These are the finite element models, correlation based methods, and computational intelligence methods.
Tshilidzi Marwala

Chapter 2. Data Processing Techniques for Condition Monitoring

Abstract
This chapter reviews data processing techniques for condition monitoring in mechanical and electrical systems. Methods for acquiring data are described and methods for analyzing data are explained. In particular, modal properties, pseudo-modal energies, wavelet and mel-frequency cepstral coefficients techniques are described. In addition, the principal component analysis method is described. Finally, examples that are followed in this book are also described. These examples are gearbox data, the population of cylindrical shells data and transformer bushing data.
Tshilidzi Marwala

Chapter 3. Multi-Layer Perceptron for Condition Monitoring in a Mechanical System

Abstract
This chapter introduces the use of a Multi-Layer Perceptron (MLP) neural network to the condition monitoring of a population of cylindrical shells. The MLP technique is explained in detail with a review of relevant literature and then the MLP is used to identify faults in a population of cylindrical shells. For this chapter, the modal properties and the pseudo-modal energy data were used to classify faults. A principal component analysis was undertaken to reduce the dimensions of the input data. The multifold cross validation technique was applied to choose the optimal number of hidden units amongst the 20 trained pseudo-modal-energy-networks and the 20 trained modal-property-networks. The pseudo-modal-energy-network and the modal-property-network were found to offer similar levels of accuracy in classifying faults.
Tshilidzi Marwala

Chapter 4. Bayesian Approaches to Condition Monitoring

Abstract
Two Bayesian multi-layer perceptron neural networks were developed by applying the hybrid Monte Carlo method with one trained using Pseudo-Modal Energies while the other was trained using modal properties. They were applied to the condition monitoring of a population of cylindrical shells. The Pseudo-Modal-Energy network was found to perform better than the modal property network.
Tshilidzi Marwala

Chapter 5. The Committee of Networks Approach to Condition Monitoring

Abstract
This chapter presents the committee of neural networks method, to which was applied pseudo modal energies, modal properties (natural frequencies and mode shapes), and wavelet transform data simultaneously to identify faults in cylindrical shells. The method was tested to identify faults in a population of ten steel seam-welded cylindrical shells. The committee technique identified faults better than the three individual techniques.
Tshilidzi Marwala

Chapter 6. Gaussian Mixture Models and Hidden Markov Models for Condition Monitoring

Abstract
Bearing vibration signals features were extracted using the time-domain fractal-based feature-extraction technique. This technique used the Multi-scale Fractal Dimension (MFD) which was estimated using the Box-Counting Dimension. The extracted features were then used to classify faults using Gaussian Mixture Models (GMM) and Hidden Markov Models (HMM). The results showed that the presented feature extraction technique did indeed extract fault-specific information. Furthermore, the experiment demonstrated that HMM outperformed GMM. Nevertheless, the disadvantage of HMM was that it was more computationally expensive to train than GMM. It was therefore concluded that the presented framework gives enormous performance improvement for bearing fault detection and diagnosis. However it is recommended the GMM classifier be used when computational effort is a major consideration.
Tshilidzi Marwala

Chapter 7. Fuzzy Systems for Condition Monitoring

Abstract
The chapter presents the application of Fuzzy Set Theory (FST) and fuzzy ARTMAP (Adaptive Resonance Theory Mapping) to diagnose the condition of high voltage bushings. The diagnosis uses Dissolved Gas Analysis (DGA) data from bushings based on IEC60599, IEEE C57-104, and California State University Sacramento (CSUS) criteria for Oil Impregnated Paper (OIP) bushings. FST and fuzzy ARTMAP are compared in terms of accuracy. Both FST and fuzzy ARTMAP could diagnose the bushings condition with accuracy of 98% and 97.5% respectively.
Tshilidzi Marwala

Chapter 8. Rough Sets for Condition Monitoring

Abstract
This chapter applies the rough set technique and the ant colony optimization method for the condition monitoring of transformer bushings. The theories of rough set and ant colony optimization method are described and the method was tested for the condition monitoring of transformer bushings. The rough set optimized using the ant colony optimization method gave 96.1% accuracy using 45 rules while the equal-frequency-bin partition model gave 96.4% accuracy using 206 rules. This, therefore, implies that the ant colony optimization gives marginally better results than the equal-frequency-bin partition method.
Tshilidzi Marwala

Chapter 9. Condition Monitoring with Incomplete Information

Abstract
This chapter introduces a method for fault classification in mechanical systems in the presence of missing data entries. The method is based on auto-associative neural networks where the network is trained to recall the input data through some non-linear neural network mapping. An error equation with missing inputs as design variables is constructed from the trained network. The genetic algorithm was used to solve for the missing input values. The presented method is tested on a fault classification problem for a population of cylindrical shells. It was found that the method could estimate single-missing-entries to an accuracy of 93% and two-missing-entries to an accuracy of 91%. The estimated values were then used in the classification of faults and a fault classification accuracy of 94% was observed for single-missing-entry cases and 91% for two-missing-entry cases while the full database set gave a classification accuracy of 96%.
Tshilidzi Marwala

Chapter 10. Condition Monitoring Using Support Vector Machines and Extension Neural Networks Classifiers

Abstract
Feature extraction and condition classification are considered in this chapter. The Feature extraction approaches applied in this chapter are fractals, Kurtosis and Mel-frequency Cepstral Coefficients. The classification approaches applied in this chapter are Support Vector Machines (SVMs) and Extension Neural Networks (ENNs). The usefulness of these features were tested with SVMs and ENNs for the condition monitoring of bearings and were found to give good results.
Tshilidzi Marwala

Chapter 11. On-line Condition Monitoring Using Ensemble Learning

Abstract
This chapter introduces an on-line bushing condition monitoring approach, which can adapt to newly acquired data. This approach can accommodate new classes that are introduced by incoming data and it was implemented using an incremental learning algorithm that uses the multi-layered perceptron. The results improved from 67.5% to 95.8% as new data were introduced and the results improved from 60% to 95.3% as new conditions were introduced. On average, the confidence value of the framework about its decision was 0.92.
Tshilidzi Marwala

Chapter 12. Conclusion

Abstract
In Chap. 1, condition monitoring methods in mechanical and electrical systems were reviewed. A condition monitoring framework was discussed which entailed the domain in which the data were visualized and in particular the time, modal, frequency and time-frequency domains. A generalized condition monitoring framework which encompasses the data acquisition device, data analysis device, feature selection device, and decision making device was presented. Techniques for decision making devices were introduced: finite element models, correlation based methods and computational intelligence techniques.
Tshilidzi Marwala

Backmatter

Weitere Informationen

Premium Partner

    Marktübersichten

    Die im Laufe eines Jahres in der „adhäsion“ veröffentlichten Marktübersichten helfen Anwendern verschiedenster Branchen, sich einen gezielten Überblick über Lieferantenangebote zu verschaffen. 

    Bildnachweise