Por favor, use este identificador para citar o enlazar este ítem: http://dspace.azti.es/handle/24689/1567
Ficheros en este ítem:
No hay ficheros asociados a este ítem.
Título : Marine diesel engine ANN modelling with multiple output for complete engine performance map
Autor : Castresana, Joseba; Gabina, Gorka; Martin, Leopoldo; Basterretxea, Aingeru; Uriondo, Zigor
Resumen : Marine incidents given in recent years have been in part caused by propulsion issues. In this context, incipient propulsion faults may be identified by deviations between real values and healthy engine values provided by an accurate model. Engine modelling techniques have thus become a topic of interest in the last decade. On this basis, Machine learning approaches such as Artificial Neural Networks (ANN) have proved to be accurate and fast in terms of calculation times. However, up to now most research work has focused on predicting a few parameters for specific operation points. In order to analyse the generalization capability of ANN when predicting multiple outputs in real engine conditions, 35 different performance and emission parameters were simultaneously predicted in this study with an ANN. To do so, different engine operation points were tested in a sixcylinder marine diesel engine, characterizing the whole engine performance map. Additionally, some points from random regions throughout the entire engine performance map were tested to later analyse ANN performance on them. After defining network optimum structure and training and validating the Artificial Neural Network with 1000 data samples, the ANN was tested with data extracted from unseen random regions of the performance map. Mean Absolute Percentage Errors obtained for testing samples from random points of the engine performance map remained below 8.5\% for all parameters with the exception of CO and NO2 emissions predictions. For low temperature and high temperature cooling systems, oil system and exhaust gas system, MAPE values obtained were below 4.3\%. Calculation time for 24 testing samples containing 35 parameters was 0.109 s, which along with the high accuracy level obtained demonstrated that ANN can predict multiple outputs throughout the whole engine performance map.
Palabras clave : ANN; Diesel engine modelling; Performance prediction; Emission prediction; Performance map modelling; Multiple output prediction; ARTIFICIAL NEURAL-NETWORK; EMISSION CHARACTERISTICS; FAULT-DETECTION; PREDICTION; FUEL; SIMULATION; MAINTENANCE; COMBUSTION; PRESSURE; SYSTEM
Fecha de publicación : 2022
Editorial : ELSEVIER SCI LTD
Tipo de documento: Article
Idioma: 
DOI: 10.1016/j.fuel.2022.123873
URI : http://dspace.azti.es/handle/24689/1567
ISSN : 0016-2361
E-ISSN: 1873-7153
Patrocinador: Basque Government
Department of Economic Development and Infrastructures
Aparece en las tipos de publicación: Artículos científicos



Los ítems de DSpace están protegidos por copyright, con todos los derechos reservados, a menos que se indique lo contrario.