Front Matter

Centrifugal pumps are rotating machines that are widely used in process operations and other applications. Efficient and failure-free operation of these pumps is important for effective plant operation and productivity. However, continuous operation can lead to failure and maintenance requirements. Therefore, a centrifugal pump is considered in this book for fault diagnosis and classification using Artificial Intelligence (AI) methods combined with Wavelet Transform (WT) and Genetic Algorithm (GA).

This book shows that the proposed approach is implemented in three stages: data acquisition, in which the vibration signals of different pump conditions, namely, a healthy condition, five mechanical faulty conditions, and one hydraulic faulty condition, are acquired from a centrifugal pump rig using appropriate instrumentations and software, including an accelerometer, a data acquisition device, and LabVIEW; pre-processing and feature extraction, where the signals of the different pump conditions are pre-processed and the features are extracted using WT methods; and pump condition classification and diagnosis, in which two AI classifiers are harnessed with the involvement of GA for the optimization and training.

In this work, three types of wavelet transform are used: Continuous Wavelet Transform (CWT), Discrete Wavelet Transform (DWT), and Wavelet Packet Transform (WPT) are tested and applied as pre-processing methods to extract significant wavelet transform features to be used after that as input vectors for the artificial neural network system with the proper learning algorithm in order to approach an automatic detection and diagnosis of centrifugal pump faults. Two artificial intelligence systems are investigated for this book, namely, Multilayer Perceptron (MLP) and Support Vector Machine (SVM). MLP is trained using its traditional learning algorithm [Back Propagation (BP)] and also with a hybrid training method using a Genetic Algorithm (GA) combined with BP. Furthermore, GA is also proposed as an optimization method for the network architecture of MLP and, particularly, for the selection of hidden layers and neurons.

This approach has novel contributions starting from the centrifugal pump rig, which is specifically designed and built for this book; testing and investigating three different WT methods in which WPT is applied in a centrifugal pump fault detection and applying GA in optimization and training of MLP are also a new application for centrifugal pumps. Furthermore, a new integrated diagnostic system is proposed for the centrifugal pumps, which can be further developed and commercialized.

This book provides a background on the research area, including the motivation, problem definition, an overview of machinery condition monitoring, the methods applied, and benefits that may arise from this research and areas of consideration for the literature review.

This book reviews and discusses previous works in the areas of rotating machinery and centrifugal pump fault diagnosis methods, including conventional and automatic ones and those that apply artificial intelligence. Furthermore, a critical discussion is provided based on the advantages and disadvantages of the applied feature processing, extraction, and selection methods. Finally, limitations of existing methods are identified.

Moreso, experimental work based on centrifugal pump vibration tests, including a comprehensive description of the apparatus and installation, the data acquisition system, and the procedures for creating and monitoring the faults, was investigated in this book. A total of seven cases have been considered: a healthy condition, five mechanical faults, and a hydraulic fault.

This book illustrates the implementation of pre-processing and feature extraction for the acquired vibration signals from the centrifugal pump. A Wavelet Transform (WT) is applied for this stage using three methods, namely, Continuous CWT, DWT, and WPT. The number of features and the selection of mother wavelet function are considered to investigate the classifiers’ performance.

Furthermore, this book presents the classification results obtained from a multilayer perceptron neural network, which is trained with MLP-BP and also using hybrid training of GA combined with BP (MLP-GABP). GA is also used to optimize the number of hidden layers and neurons of MLP. The results comprise the classification accuracy rates based on the applied WT method, manual selection of hidden layers and neurons, GA based selection, training algorithm, and number of features.

In this book, the obtained classification results of SVM are presented. These results are represented by plotting the hyper-plane, which separates two different conditions, and the error rate (classification rate). The results comprise the classification accuracy rates based on the applied WT methods, selection of the best kernel type, and number of features.

Besides, a discussion on the obtained results of a multilayer perceptron neural network, which is trained with MLP-BP and also using hybrid training of GA combined with BP (MLP-GABP), and SVM classifiers to classify the fault conditions of a centrifugal pump is presented in this book. The discussion includes a comparison on the classification results based on WT feature extraction methods, CWT, DWT, and WPT, GA based hidden layers and neuron selection, training algorithm, and number of features.

A comparative study of two artificial intelligent systems, namely, MLP and SVM, to classify six fault conditions and the normal (non-faulty) condition of a centrifugal pump, is investigated in this book. A hybrid training method for MLP is proposed for this work based on the combination of BP and GA. The two training algorithms are tested and compared separately as well. Features are extracted using DWT; both approximations, details, and two mother wavelets are used to investigate their effectiveness on feature extraction. GA is also used to optimize the number of hidden layers and neurons of MLP. In this book, the feature extraction, GA based hidden layers, neuron selection, training algorithm, and classification performance, based on the strengths and weaknesses of each method, are discussed. From the results obtained, it is observed that the DWT with both MLP-BP and SVM produces better classification rates and performance.

In this book, a hybrid training method for MLP is proposed based on combining BP and GA. The proposed scheme is compared with the SVM approach to classify six fault conditions and the normal condition of a centrifugal pump. Two training algorithms are tested and compared. Features are extracted using WPT with three levels of decomposition, and two mother wavelets are used to investigate their effectiveness on feature extraction. Furthermore, GA is also used to optimize the number of hidden layers and neurons of MLP. The results obtained show improved performance on the feature extraction, GA based hidden layers, neuron selection, training algorithm, and classification performance using the proposed scheme.

Finally, this book presents major and overall conclusions that are extracted from the implemented work based on the obtained results. Further possible implementations and investigations are also presented as future work. Readers can refer to Appendix A for the codes to generate some of the results obtained in this book.

The authors gratefully acknowledge Dr. Peter Wallace (Department of Applied Science, Glasgow Caledonian University, United Kingdom); Dr. David Harrison (Department of Mechanical Engineering, Glasgow Caledonian University, United Kingdom); and Dr. Khalid Abdulraheem (Canadore Stanford College, Canada).

Dr. Maamar Ali Saud Al Tobi

Assistant Professor, National University of Science and Technology, Muscat, Oman

Dr. Geraint Bevan

Senior Lecturer, Glasgow Caledonian University, United Kingdom

Dr. Kenneth Okedu

Visiting Professor, National University of Science and Technology, Muscat, Oman

Adjunct Professor, Nisantasi University, Istanbul, Turkey

AI

Artificial Intelligence

ANFS

Adaptive Network Fuzzy System

ANN

Artificial Neural Network

BP

Back Propagation

BPN

Back Propagation Network

CDET

Compensation Distance Evaluation Technique

CM

Condition Monitoring

CP

Centrifugal Pump

CWT

Continuous Wavelet Transform

D/AC

Digital/Analog Converter Card

DAQ

Data Acquisition Device

DWT

Discrete Wavelet Transform

ELM

Extreme Learning Machine

FFT

Fast Fourier Transform

FP

Forward Propagation

GA

Genetic Algorithm

GABP-ANN

Genetic Algorithm Back Propagation-Artificial Neural Network

GEP

Gene Expression Programming

IDE

Improved Distance Evaluation

KF

Kurtosis Factor

KNN

K-Nearest-Neighbor

LM

Levenberg–Marquardt

MLP

Multilayer Perceptron

MLP-ANN

Multilayer Perceptron-Artificial Neural Network

MLP-BP

Multilayer Perceptron-Back Propagation

MLP-GABP

Multilayer Perceptron-Genetic Algorithm Back Propagation

MSE

Mean Square Error

NSPs

Non-Dimensional Symptom Parameters

PCA

Principal Component Analysis

PNN

Probabilistic Neural Network

PSO

Particle Swarm Optimization

PSVM

Proximal Support Vector Machine

RBF

Radial Basis Function

RMS

Root Mean Square

SD

Standard Deviation

SDWPC

Standard Deviations and Wavelet Packet Coefficients

SOM

Self-Organizing Maps

STFT

Short Time Fourier Transform

SVM

Support Vector Machine

VFD

Variable Frequency Drive

WPS

Wavelet-Scale Power Spectrum

WPT

Wavelet Packet Transform

WT

Wavelet Transform