Please use this identifier to cite or link to this item:
Files in This Item:
There are no files associated with this item.
Title: Healthy marine diesel engine threshold characterisation with probability density functions and ANNs
Authors: Castresana, Joseba; Gabina, Gorka; Quincoces, Inaki; Uriondo, Zigor
Abstract: The condition-based maintenance of marine propulsion systems is attracting increasing interest in safety-related, financial, and environmental terms. Many researchers have studied different marine diesel engine models and fault identification techniques. However, the thresholds between a healthy and a faulty engine have not been thoroughly analysed. Thus, this study aims to determine healthy engine threshold values for multiple parameters of a marine diesel engine. To this end, an operative commercial fishing vessel was considered, and multiple engine performance variables were measured through 2020 and the first half of 2021, totalling 5181 operating hours of the main engine without any fault occurrence. Preliminary correlation and relative deviation studies suggested the analysis of some constant trend parameters with alternative modelling techniques. Hence, probability density functions (PDFs) were used to establish confidence intervals for such parameters with data from the entire year of 2020. The parameters with the highest correlation and deviation were alternatively modelled using artificial neural networks (ANNs). Four different ANNs were trained, validated, and tested with data from 2020, calculating the mean absolute percentage errors for all the predicted parameters. Finally, data from 2021 were used to validate both the PDF and ANN modelled parameter thresholds set in 2020. For the 2021 data, the confidence interval set with the PDF showed a maximum failure rate of 1.21\%. Alternatively, the ANN model parameters exhibited maximum percentage errors of 1.1\%, 1.22\%, and 1.95\% for the engine performance, cooling, and cylinder subsystems, respectively. Finally, all the obtained thresholds were summarised, providing a good source for establishing faulty engine threshold values in future fault detection studies.
Keywords: Artificial neural networks; Probability density functions; Diesel engine modelling; Ship propulsion characterisation; Threshold characterisation; Onboard model validation; ARTIFICIAL NEURAL-NETWORK; FAULT-DETECTION; PERFORMANCE; PREDICTION; DIAGNOSIS; ALGORITHMS; IMPROVE; SYSTEMS; MODEL; RISK
Issue Date: 2023
Type: Article; Early Access
DOI: 10.1016/j.ress.2023.109466
ISSN: 0951-8320
E-ISSN: 1879-0836
Funder: Basque Government
Department of Economic Development and Infrastructure
European Union's Horizon 2020 research and innovation programme (SusTunTech) [869353]
Appears in Publication types:Artículos científicos

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.