Pumping costs within British industry are enormous, with the potential for considerable financial savings through fault diagnosis and condition-based maintenance. Accurate condition monitoring data interpretation is a key requirement in pump fault diagnosis. However, the human skills required to transform monitored data into maintenance information are often unavailable. Artificial neural networks (ANNs) are proposed for automation of this skill in the development of a pumping system decision support tool, the key requirement of which is accurate pump fault diagnosis. The cumulative sum charting procedure was used to establish a knowledge base of fault data for ANN implementation based on historical parameter measurements. Various preprocessing techniques were investigated in relation to generalization ability and convergence rates during the learning phase. Preprocessing predominantly affected ANN convergence rate, with the quality of training data crucial to generalization ability. ANNs could provide accurate, incipient fault diagnosis of pumping machinery based on real industrial data corresponding to historical pump faults.

Fault diagnosis of pumping machinery using artificial neural networks

Automatic Language Identification involves analyzing language-specific features in speech to determine the language of an utterance without regard to topic, speaker or length of speech. Although much progress has been made in recent years, language identification systems have not been built on detailed underlying theory or linguistically meaningful design criteria. This thesis is motivated by the belief that features used to discriminate between languages should be linguistically sound; the result is a unique combination of design, theory and implementation.
In this thesis a "word-spotting" algorithm is introduced motivated by a perceptual study (82) reporting that human subjects use language-dependent phonemes and short sequences to identify languages. In order to find an optimal set of phoneme-like tokens to represent speech in a linguistically meaningful way, a mathematical model of the discrimination between two languages is developed. This model permits the automatic design of a token representation of speech by selecting a list of discriminating "words" in a data-driven manner. The resulting system has the flexibility to automatically take into account the inherent structure of the Languages to be discriminated. A second mathematical model is developed to measure the impact of inaccurate automatic alignment of tokens on language discrimination. This model indicates why some algorithms aiming to compensate for these inaccuracies have not been successful. The theoretical models and the "word"-spotting algorithm have been implemented and validated on both generated and real-world speech data.
This dissertation makes several significant contributions: the design of a simple and linguistically sound language-identification module; a flexible automatic feature extraction algorithm; a mathematical model to estimate the discriminability of two languages; and a mathematical model to capture the impact of inaccurate alignment on the discriminability of two languages.

Automatic language identification with sequences of language-independent phoneme clusters

A practical system, based on acoustic emission techniques, for detection of partial discharge activity in cable terminations is presented. Analysis of the received signal is carried out in both the time domain and in the frequency domain using a neural network approach. Laboratory synthesized samples were used with controlled discharges. Signals received by the sensor were recorded by a digital oscilloscope and then analyzed by computer. Time domain analysis provided an indication of the size of the discharge. The relationship between the acoustic signal and the discharge size is demonstrated. The frequency content of the signal is used to differentiate between the signals generated by different discharge sources. Variation of the acoustic signal with applied voltage has shown that the discharge may be detected at its inception. Practical samples of cable terminations containing known discharges were also tested. The system could detect these discharges and give an upper bound on their energy dissipation. The system is acoustic in nature and thus virtually free from electrical interference. An optical fibre coupling between the acoustic transducer and the test equipment protects the system from flashover and allows for on-line testing of cable terminations. >

Partial discharge detection in cable termination using acoustic emission techniques and adaptive signal processing

We propose a new structural learning algorithm for organizing the structure of the multi-layered neural networks. The proposed pruning algorithm consists of two already known algorithms, the structural learning algorithm with forgetting and the optimal brain damage algorithm using the second derivatives of the assessment. After the network is slimmed by the structural learning algorithm with forgetting, unimportant weights are pruned from the network using the second derivatives. The simulations are performed for the Boolean function and the acoustic diagnosis of compressors. The results show that the proposed algorithm is effective for eliminating the unimportant weights.

A structural learning algorithm for multi-layered neural networks

The detection efficiency of the continuous dynode electron multiplier, Ceratron, for positrons has been obtained as a function of the incident energy Ein for a slow positron beam measurement. For comparison, the efficiency curve for a channel electron multiplier (CEM) was also obtained under the same conditions. In the use of a Ceratron, the counting rate in conventional use is low compared with that under the best conditions, obtained using a new method. The dependence of the efficiency on Ein is very large for the Ceratron, but rather small for the CEM. The dependences in the low energy region for the Ceratron and for the CEM show that calibration of the efficiency on Ein for each detector is necessary in the conventional use for electrons. The detection efficiency of a Ceratron for positrons under the best conditions is almost the same as that for a CEM.

Detection Efficiency of a Continuous Dynode Electron Multiplier (Ceratron) for Positrons

Describes an acoustic diagnosis technique for a compressor using a hybrid neural network (HNN). The HNN is composed of two neural networks: an acoustic feature extraction network, and a fault discrimination network. The acoustic feature extraction network uses an auto-associative neural network (ANN) whose target patterns are the same as the input patterns. The five-layered neural network is composed of two three-layered neural networks to compress the input information and to restore the compressed information. The authors examine the architecture of the ANN for acoustic diagnosis, the proper form of the activation function in the output layer and the proper number of hidden layers. The fault discrimination network uses a multilayered neural network whose input patterns are the output values of the hidden layer in the ANN. The authors examine the possibility of discriminating between eight types of compressor faults with high accuracy by using an HNN. >

Acoustic diagnosis for compressor with hybrid neural network

Hybrid Neural Networks as a Tool for the Compressor Diagnosis

It is important for this diagnosis to detect the surging phenomena which lead to the destruction of the blower. Since the surging sound is a non-stationary signal, the wavelet transform is more suitable for the pre-processing method than FFT transform. The dyadic wavelet transform is used as the pre-processing method. The multi-layered neural network is used as the discrimination method. The results show that the neural network with the wavelet transform can detect the surging sound perfectly.

Acoustic diagnosis for blower with wavelet transform and neural networks

Describes the acoustic diagnosis technique for a compressor using a hybrid neural network (HNN). The HNN is composed of two neural networks, an acoustic feature extraction network using a backpropagation network (BPN) and a fault discrimination network using a Gaussian potential function network (GPFN). The BPN is composed of five layers and the number of the middle hidden units is smaller than the others. The target patterns for the output layer are the same as the input patterns. After the learning of the network, the middle hidden layer acquires the compressed input information. The input patterns of the GPFN are the output values of the middle hidden layer in the BPN. The task of the HNN is to discriminate four conditions of the valve under various experimental conditions. As a result, 93.6% discrimination accuracy is obtained in this experiment. This suggests that the proposed model is effective for the acoustic diagnosis.

Haruya Matsumoto

Papers

Acoustic diagnosis for compressor with hybrid neural network

Hybrid Neural Networks as a Tool for the Compressor Diagnosis

Acoustic diagnosis for blower with wavelet transform and neural networks

Hybrid neural networks for acoustic diagnosis

Acoustic diagnosis with neural networks.