[Previous]
[Module]
[Home]
[List of Projects]
[Next]
|
[Previous] |
[Module] |
[Home] |
[List of Projects] |
[Next] |
| Project No. | 5003-039047 |
| Subject | Wear diagnosis on axial face seals with neural networks |
| Authors | Urs Müller, Dr. Institut für Elektronik (IfE), Swiss Federal Institute of Technology at Zurich (ETH)
Bernhard Birchler, Dr.
|
| Start | 1.1.95 |
| End | 30.6.96 |
| Contents | The project is concerned with laying the foundations for a new type of monitoring and diagnostic system for axial face seals. Experiments show that casing noise includes components deriving from pump axial face seals. These seals work under conditions of heavy loading and are often the cause of damage. So far, no direct measurement method has allowed the early detection of wear on the seals. In the present project, diagnosis is based on measured structure-borne noise, with evaluation by a neural network. Analytic modelling has no chance of success in such cases because of the complexity involved; but the direct path via casing noise is very promising. The measurements can be made while the installation is running, without having to take the pumps out. The task is to find neural networks which meet the demands of the problem and to make available the necessary training data. Processing these very extensive data was only made possible in the first place thanks to the MUSIC parallel computer developed at IfE.
Diagnosing wear in machine components without interrupting operation is a major factor in cutting maintenance costs. For one thing, it means savings in cost-intensive downtimes; for another, machine components remain in service longer and unscheduled outages can be predicted in good time. In the long term, early diagnosis requires permanent monitoring and evaluation. For this, a monitoring system must be self-contained and equipped with its own intelligence. Neural networks represent a highly promising line of approach; they are now poised between theoretical research and practical implementation. However, successful industrial applications of this technology depend on interdisciplinary co-operation. |
| Co-operation with | Sulzer-Innotec company. Sulzer-Innotec provides the practical experience with problems and the necessary field data, while IfE contributes in-depth knowledge of neural networks and simulation on parallel computers. |
|
[Previous] |
[Module] |
[Home] |
[List of Projects] |
[Next] |