City Research Online

Abstract

Over the last few years, vehicular predictive maintenance has witnessed a shift from utilizing raw sensor reading directly to using fault events registered in On-Board Diagnostic systems (OBDs). Instead of providing raw sensory data, OBDs equip drivers and technicians with diagnostic information coming from different Electric Control Units (ECUs) in the vehicles, usually indicated as Diagnostic Trouble Codes (DTCs). These DTCs are categorical (non-numeric) or alphanumeric, and relate to different problems within the vehicle. Having many categories and multiple attributes has previously restricted researchers to analyzing a few DTCs at a time, with a limited set of machine learning algorithms. This has recently changed with the advent of the self-supervised next DTC approach, which ranges from an LSTM-based multivariate next-prediction model to an attention mechanism and transformerdecoder model. These models reframe the problem of predictive maintenance as the next fault event prediction task and use metrics like top-3 and top-5 accuracy to evaluate the predictive capabilities of the model. We propose a new architecture for the next DTC prediction task, DTC-TranGru, which combines the benefits of transformer and GRU models and shows that it outperforms them with around 2% increase in the top-5 accuracy benchmark for a next-DTC prediction task.

Publication Type:	Conference or Workshop Item (Paper)
Additional Information:	© 2024 Association for Computing Machinery. This is the author's version of the work. It is posted here for your personal use. Not for redistribution. The definitive Version of Record will be published in https://dl.acm.org/.
Publisher Keywords:	Predictive maintenance, Diagnostic Trouble Code, GRU, Transformers, Attention Mechanism, Embeddings
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science T Technology > TK Electrical engineering. Electronics Nuclear engineering T Technology > TL Motor vehicles. Aeronautics. Astronautics
Departments:	School of Science & Technology > Computer Science

Export

Downloads