It is very important for the waste water industry to use modern condition monitoring technologies as cost effective pump maintenance strategies. Catastrophic failure of the pumps during flooding periods has always been a great threat to waste water companies and large numbers of stand by pumps are usually required to avoid this happening. The introduction of an on-line vibration based condition monitoring system will enable the maintenance work to be planned in advance and completed at the right time. this will ensure that damaged pump parts are changed when needed and will reduce the need for unnecessary quantities of pumps to be held on site. Vibration monitoring as a mature technique has numerous applications in condition monitoring and a vibration based system was successfully developed in this project for the pumps in one of Southern Water's pumping stations. A vibration database has subsequently been developed, using conventional signal processing techniques. However, a major challenge for this project was to use the vibration monitoring system to predict pump conditions automatically instead of analysing the data off-line. Artificial intelligence technology is the solution proposed in this work. Artificial neural networks (ANNs) are one of the fastest developing areas in artificial intelligence. This is a powerful technique for pattern recognition, which can be used for pump faults classification. Suitably designed neural networked are able to 'think' like a human being and can 'tell' when the machine has problems and also identify the type of problems present. Due to the long period of time required to obtain essential information from the pumping station for neural network training, a test-rig pump was established in the laboratory to simulate the common pump faults, including impeller imbalance, impeller wear problems and typical bearing defects. Although the application of ANNs has been investigated in a wide range of research, most of the applications in pattern recognition to date have tended to use a particular network model without sufficient justification. A systematic way of finding the best performance network for a pattern recognition problem would be very useful for neural network applications. During this study, some investigations in this respect have been carried out for pump fault classification. the best performance networks for the classification of the simulated pump faults on the test-rig were successfully designed based on the effectively developed feature extraction methods. The other area studied here was Wavelet Transforms (WTs). Being a new signal processing technique. It has been used as an alternative to the fast Fourier transform with its unique advantages. By using WTs not only frequency information can be retrieved from the original signal, but also time information. The potential of more applications of WTs in pump fault diagnosis is expected based on the research results from this study. Several novel developments have emerged from this study. Condition monitoring is introduced to the waste water industry as a cost effective maintenance strategy. Artificial neural networks are used for pump fault diagnosis, which will provide the basis for a future on-line condition monitoring systems. A methodology of systematically finding the best performance neural networks was also proposed.
| Date of Award | 2002 |
|---|
| Original language | English |
|---|
| Awarding Institution | - Nottingham Trent University
|
|---|