Investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis
Although the advancements in marine engines diagnosis technologies and systems, estimating faults combinations at the entire operating envelope is challenging. This study aims at first investigating the path logarithmic signatures (logS) method for information extraction and dimensions reduction fro...
| Main Authors: | , , , |
|---|---|
| Format: | Journal article |
| Language: | English |
| Published: |
Elsevier
2024
|
| _version_ | 1824458833454432256 |
|---|---|
| author | Patil, C Theotokatos, G Wu, Y Lyons, T |
| author_facet | Patil, C Theotokatos, G Wu, Y Lyons, T |
| author_sort | Patil, C |
| collection | OXFORD |
| description | Although the advancements in marine engines diagnosis technologies and systems, estimating faults combinations at the entire operating envelope is challenging. This study aims at first investigating the path logarithmic signatures (logS) method for information extraction and dimensions reduction from the in-cylinder pressure signals, and secondly, proposing the most effective data-driven hybrid approach employing logS as input to artificial neural networks (ANN) regression to estimate the severity of critical faults in marine engines. A large four-stroke marine diesel engine is considered and in-cylinder pressures are generated using a validated physics-based digital twin by simulating scenarios with four faults combinations of varying severity in the entire operating envelope. A parametric study is performed to quantify the logS number impact on the ANN regression models accuracy and training time. Four data pre-processing approaches, which consider elementary or high variance logS, without or with the use of Principal Components Analysis (PCA), are also comparatively assessed. The results demonstrate that the approach involving the use of eight elementary logS, Principal Components Analysis (PCA), standardisation and an ANN regression model comprising two hidden layers with ten neurons each is the most effective, as it exhibits the lowest values on both the root mean square and the standard error 95% confidence interval. This is the first study on logS application for marine engines faults severity estimation, and as such it impacts the development of future data-driven diagnostics methods. |
| first_indexed | 2025-02-19T04:32:10Z |
| format | Journal article |
| id | oxford-uuid:ce0ab053-99a5-4dc8-b1c4-3d320395f0cb |
| institution | University of Oxford |
| language | English |
| last_indexed | 2025-02-19T04:32:10Z |
| publishDate | 2024 |
| publisher | Elsevier |
| record_format | dspace |
| spelling | oxford-uuid:ce0ab053-99a5-4dc8-b1c4-3d320395f0cb2025-01-09T18:07:21ZInvestigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosisJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:ce0ab053-99a5-4dc8-b1c4-3d320395f0cbEnglishSymplectic ElementsElsevier2024Patil, CTheotokatos, GWu, YLyons, TAlthough the advancements in marine engines diagnosis technologies and systems, estimating faults combinations at the entire operating envelope is challenging. This study aims at first investigating the path logarithmic signatures (logS) method for information extraction and dimensions reduction from the in-cylinder pressure signals, and secondly, proposing the most effective data-driven hybrid approach employing logS as input to artificial neural networks (ANN) regression to estimate the severity of critical faults in marine engines. A large four-stroke marine diesel engine is considered and in-cylinder pressures are generated using a validated physics-based digital twin by simulating scenarios with four faults combinations of varying severity in the entire operating envelope. A parametric study is performed to quantify the logS number impact on the ANN regression models accuracy and training time. Four data pre-processing approaches, which consider elementary or high variance logS, without or with the use of Principal Components Analysis (PCA), are also comparatively assessed. The results demonstrate that the approach involving the use of eight elementary logS, Principal Components Analysis (PCA), standardisation and an ANN regression model comprising two hidden layers with ten neurons each is the most effective, as it exhibits the lowest values on both the root mean square and the standard error 95% confidence interval. This is the first study on logS application for marine engines faults severity estimation, and as such it impacts the development of future data-driven diagnostics methods. |
| spellingShingle | Patil, C Theotokatos, G Wu, Y Lyons, T Investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis |
| title | Investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis |
| title_full | Investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis |
| title_fullStr | Investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis |
| title_full_unstemmed | Investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis |
| title_short | Investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis |
| title_sort | investigation of logarithmic signatures for feature extraction and application to marine engine fault diagnosis |
| work_keys_str_mv | AT patilc investigationoflogarithmicsignaturesforfeatureextractionandapplicationtomarineenginefaultdiagnosis AT theotokatosg investigationoflogarithmicsignaturesforfeatureextractionandapplicationtomarineenginefaultdiagnosis AT wuy investigationoflogarithmicsignaturesforfeatureextractionandapplicationtomarineenginefaultdiagnosis AT lyonst investigationoflogarithmicsignaturesforfeatureextractionandapplicationtomarineenginefaultdiagnosis |