- Tytuł:
- Attempt to utilise histogram of vibration cepstrum of engine body for setting up the clearance model of the piston-cylinder assembly for PNN neural classifier
- Autorzy:
-
Madej, H.
Czech, P. - Powiązania:
- https://bibliotekanauki.pl/articles/243660.pdf
- Data publikacji:
- 2008
- Wydawca:
- Instytut Techniczny Wojsk Lotniczych
- Tematy:
-
diagnostics
combustion engines
artificial neural networks
vibration - Opis:
- The paper presents an attempt to evaluate the wear of piston-cylinder assembly with the aid of vibration signal recorded on spark ignition (SI) engine body. The subject of the study was a four-cylinder combustion engine 1.2 dm3. Diagnosing combustion engines with vibration methods is specifically difficult due to the presence of multiple sources of vibration interfering with the symptoms of damages. Diagnosing engines with vibroacustic methods is difficult also due to the necessity to analyse non-stationary and transient signals [5]. Various methods for selection of usable signal are utilised in the diagnosing process. Changes of the engine technical condition resulting from early stages of wear are difficult to detect for the effect of mechanical defect masking by adaptive engine control systems [3]. According to the studies carried out, it is possible to utilise artificial neural networks for the evaluation of the clearance in piston-cylinder assembly. It was proven that it is possible to set up a properly operating neural classifier able to identify the degree of wear in the piston-cylinder assembly, based on the signal of vibration acceleration in the engine body. Faultless classification was successfully obtained with the use of probabilistic neural network with properly selected value of y coefficient. At the same time, based on the experiments carried out, the crucial role was confirmed for the selection of proper method for pre-treatment of data intended for neural network teaching.
- Źródło:
-
Journal of KONES; 2008, 15, 3; 305-311
1231-4005
2354-0133 - Pojawia się w:
- Journal of KONES
- Dostawca treści:
- Biblioteka Nauki