Intelligent Method for Condition Diagnosis of Pump System Using Discrete Wavelet Transform, Rough Sets and Neural Network

Abstract

An intelligent method for condition diagnosis of a pump system is proposed using the discrete wavelet transform (DWT), rough sets (RS), and a neural network to detect faults and distinguish fault types at an early stage. The Daubechies wavelet function is used to extract fault features from measured vibration signals and to capture hidden fault information across an optimum frequency region. We also propose a new diagnosis method based on a fuzzy neural network realized by the partially-linearized neural network (PNN), one which can automatically distinguish fault types on the basis of the probability distributions of symptom parameters. The diagnosis knowledge for the training of the PNN can be acquired by using the RS. The PNN can deal with the ambiguity problem of diagnosis, and is always convergent. Practical examples of diagnosis for a centrifugal pump system are shown in order to verify the efficiency of the method.

Fault Diagnosis and State Detection in Centrifugal Pumps—A Review of Applications

“…Appropriate models are Fourier transforms regarding harmonic signals as in [8,9] for instance or statistical representations concerning random signals. Other signal model-based feature generators are correlation analysis and wavelet analysis as used in [10]. An example for recent techniques based on Empirical Mode Decomposition is given in [11].…”

“…With a decision tree classifier, the fault states of gas in the fluid, cavitation, and blockade are detected. In contrast to previous studies, Discrete Wavelet Transforms are used in [10]. The identification of signal models from vibration signals acquired with accelerometers generates features for classification with a Neuro-Fuzzy classifier based on rough set theory.…”

“…Furthermore, inference methods as decision making by fuzzy logic (FUZ) are utilized [24,25]. To define the rules, rough set theory can be used as in [26] and [10]. Those use a lower and an upper approximation for the boundaries of a cluster based on probabilities and show several similarities to other methods like fuzzy sets or Bayesian inference [27].…”

Ant colony optimization of rough set for HV bushings fault detection

“…Rough set is suitable in modeling methods that are used for both classification and interpretation due to its rule base nature. Rough set is usually used as a feature extraction and dimension reduction tool in conjunction with another computational intelligence method [8,9]. It has been used successfully in fault detection of bushings [10].…”

A Transfer-Based Convolutional Neural Network Model with Multi-Signal Fusion and Hyperparameter Optimization for Pump Fault Diagnosis

“…The signal types required for fault diagnosis of the pump based on the signal processing method include vibration signals [6] and sound signals [7]. Usually, the feature extraction methods employ wavelet transformation [8] and fast Fourier transformation [9].…”