Learning from the Dark Side: a parallel time series modelling framework for forecasting and fault detection on intelligent vehicles
| dc.contributor.author | Xing, Yang | |
| dc.contributor.author | Hu, Zhongxu | |
| dc.contributor.author | Hang, Peng | |
| dc.contributor.author | Lv, Chen | |
| dc.date.accessioned | 2024-03-07T15:36:31Z | |
| dc.date.available | 2024-03-07T15:36:31Z | |
| dc.date.issued | 2023-12-13 | |
| dc.description.abstract | Time series vehicle state modelling is crucial in various real-world applications, such as fault detection, fault tolerance, optimization, and cyber security for intelligent vehicles (IVs). In this study, we propose a novel parallel time series modeling framework (PTSM) to forecast and detect vehicle braking cylinder pressure states, thereby enhancing the safety of the braking system. Specifically, the PTSM consists of two branches: LightNet and DarkNet. The LightNet learns time-series (TS) representations of real-world signals to forecasts and identifies vehicle states. On the other hand, the DarkNet employs a novel multi-task learning and dual Relativistic Generative Adversarial Network (dual-RaGAN) framework to reconstructs healthy sequential states, detects faults, and forecasts future vehicle states using synthesized faulty sequences. To develop the PTSM framework, we introduce a novel data processing and random fault synthesizing method. We evaluate the performance of the dual-RaGAN model using real-world data and compare it with non-adversarial approaches, demonstrating the efficiency of the multi-task generative sequential representation. Extensive experimental results show that by integrating knowledge from the dark side, real-world time-series modelling (TSM) for forecasting and fault detection can be significantly improved, with a 34.7% enhancement in forecasting and an 11% improvement in fault recognition. The results indicate that signal reconstruction leads to more accurate sequence forecasting and fault recognition in both the dark and light sides. This proposed study not only introduces a novel time-series modelling framework but also establishes a new approach for vehicle testing, fault detection, and cyber security research for intelligent vehicles. Data and Codes are available at: https://github.com/YXING-CC/Dark-Light. | en_UK |
| dc.identifier.citation | Xing Y, Hu Z, Hang P, Lv C. (2024) Learning from the Dark Side: a parallel time series modelling framework for forecasting and fault detection on intelligent vehicles. IEEE Transactions on Intelligent Vehicles, Volume 9, Issue 2, February 2024, pp. 3205-3219 | en_UK |
| dc.identifier.eissn | 2379-8904 | |
| dc.identifier.issn | 2379-8858 | |
| dc.identifier.uri | https://doi.org/10.1109/TIV.2023.3342648 | |
| dc.identifier.uri | https://dspace.lib.cranfield.ac.uk/handle/1826/20949 | |
| dc.language.iso | en | en_UK |
| dc.publisher | IEEE | en_UK |
| dc.rights | Attribution-NonCommercial 4.0 International | * |
| dc.rights.uri | http://creativecommons.org/licenses/by-nc/4.0/ | * |
| dc.subject | Time-series modelling | en_UK |
| dc.subject | generative time-series model | en_UK |
| dc.subject | relativistic GAN | en_UK |
| dc.subject | deep learning | en_UK |
| dc.subject | fault detection | en_UK |
| dc.subject | intelligent vehicles | en_UK |
| dc.title | Learning from the Dark Side: a parallel time series modelling framework for forecasting and fault detection on intelligent vehicles | en_UK |
| dc.type | Article | en_UK |