Mileage-Aware for Vehicle Maintenance Demand Prediction
Abstract
:1. Introduction
- Given that mileage is not in the same vector space as maintenance projects, we propose a mileage representation tailored to maintenance demand prediction.
- To fully leverage the impact of key temporal information on maintenance demand prediction, we introduce a key temporal information learning module that combines LSTM with attention mechanism.
- To enhance the deep integration of mileage and maintenance projects, we propose an information fusion module based on gated unit.
- Extensive experiments with real-world data demonstrate the superior performance of our model compared to contemporary baselines.
2. Related Work
2.1. Single Data-Based Vehicle Maintenance Demand Prediction
2.2. Combined Data-Based Vehicle Maintenance Demand Prediction
3. Method
3.1. Notations
3.2. Framework
Notation | Description |
---|---|
Set of all maintenance project codes | |
The maintenance projects for a vehicle | |
The maintenance projects for t-th maintenance | |
The maintenance mileages for a vehicle | |
The representation learning function of | |
Co-occurrence matrix for all maintenance projects | |
The correlation-aware learning function of | |
The correlation-aware learning result of and | |
The representation learning result of | |
The hidden state of after LSTM | |
The hidden state of after LSTM | |
The key temporal information learning result of | |
The key temporal information learning result of | |
B | The comprehensive key temporal information learning result of |
D | The comprehensive key temporal information learning result of |
J | The comprehensive key temporal information learning result of and |
The Hadamard product result of and | |
The concatenate result of and | |
The activation result of | |
The Hadamard product result of and | |
The fusion result of and | |
I | The comprehensive fusion result of and |
Y | The fusion result of I and J |
Maintenance projects for the ()-th maintenance |
3.3. Mileage Representation Learning
3.4. Key Temporal Information Learning
3.5. Multi-Source Information Fusion and Prediction
3.6. Model Optimization
4. Experiment Result and Analysis
4.1. Dataset Description
- Data cleaning: Removed duplicate records;
- Data normalization: Categorized the maintenance projects and coded them into a uniform format;
- Data filtering: Excluded vehicles with incorrect maintenance records and those with fewer than two maintenance records.
4.2. Inputs and Outputs
4.3. Baseline Models and Evaluation Metrics
4.4. Implementation Details
4.5. Prediction Performance
4.6. Performance Evaluation on Data Sufficiency
4.7. Ablation Study
- Mila-mileage: This variant attempts to elucidate the role of mileage in the model by omitting the input of mileage from the model such that , J = D.
- Mila-atten: This variant attempts to elucidate the role of the attention mechanism by omitting inputs to the attention mechanism from the learning module of key temporal information such that , .
- Mila-gated: This variant accounts for its role in the model by omitting the gated unit information fusion module, making .
- Overall performance: The complete Mila model demonstrates the highest performance across all evaluation metrics. This indicates that each component (mileage input, key temporal information learning, and gated unit information fusion) is crucial for improving overall performance.
- Component importance: The results for Mila-atten and Mila-gated are comparable; however, Mila-gated marginally outperforms the former in most metrics, suggesting that the attention mechanism might be slightly more crucial than the gated unit information fusion module. Moreover, the performance of Mila-mileage lags behind the other two ablation variants, indicating that mileage plays a significant role in forecasting vehicle maintenance demand.
4.8. New Maintenance Projects Prediction
- Predictive accuracy: The Mila model significantly outperforms all baseline models across all evaluation metrics, demonstrating superior accuracy in predicting new maintenance projects.
- Performance at higher k values: The Mila model exhibits an exceptional performance at higher k values, indicating its ability to more accurately predict new maintenance projects. This is crucial for improving service coverage and identifying potential vehicle issues.
Model | w-F1 | R@5 | R@10 | R@15 | R@20 | R@25 | R@30 | R@35 |
---|---|---|---|---|---|---|---|---|
SLFN | 13.08 ± 0.16 | 22.12 ± 0.13 | 33.63 ± 0.12 | 36.31 ± 0.13 | 46.25 ± 0.11 | 52.11 ± 0.12 | 53.75 ± 0.22 | 55.53 ± 0.32 |
MLP | 13.06 ± 0.29 | 22.03 ± 0.19 | 33.29 ± 0.42 | 36.96 ± 0.11 | 46.19 ± 0.36 | 52.13 ± 0.23 | 53.46 ± 0.19 | 55.96 ± 0.24 |
DBN | 13.04 ± 0.21 | 22.21 ± 0.23 | 33.68 ± 0.12 | 36.61 ± 0.22 | 46.49 ± 0.25 | 52.23 ± 0.27 | 53.86 ± 0.21 | 55.66 ± 0.13 |
CNN | 17.98 ± 0.54 | 23.96 ± 0.46 | 33.76 ± 0.55 | 40.68 ± 0.50 | 46.23 ± 0.29 | 51.09 ± 0.27 | 55.70 ± 0.33 | 58.99 ± 0.10 |
GRU | 15.52 ± 0.41 | 21.88 ± 0.45 | 31.68 ± 0.19 | 36.81 ± 0.56 | 43.20 ± 0.64 | 50.28 ± 0.31 | 56.29 ± 0.66 | 58.78 ± 0.48 |
RNN | 16.85 ± 0.28 | 21.92 ± 0.07 | 29.97 ± 0.16 | 36.86 ± 0.19 | 44.34 ± 0.24 | 51.40 ± 0.39 | 54.02 ± 0.31 | 58.79 ± 0.17 |
LSTM | 13.35 ± 0.17 | 23.94 ± 0.18 | 30.96 ± 0.20 | 38.27 ± 0.14 | 45.15 ± 0.20 | 49.99 ± 0.13 | 53.82 ± 0.32 | 56.04 ± 0.34 |
EFMSAE-LSTM | 13.74 ± 0.21 | 23.13 ± 0.12 | 30.68 ± 0.12 | 38.24 ± 0.26 | 45.65 ± 0.23 | 49.36 ± 0.14 | 53.54 ± 0.13 | 56.62 ± 0.22 |
Dipole | 14.89 ± 0.31 | 23.43 ± 0.00 | 32.94 ± 0.28 | 37.35 ± 0.18 | 45.83 ± 0.17 | 49.21 ± 0.48 | 53.80 ± 0.41 | 59.71 ± 0.31 |
RETAIN | 18.91 ± 0.50 | 25.02 ± 0.57 | 34.97 ± 0.24 | 41.81 ± 0.37 | 46.10 ± 0.50 | 53.75 ± 0.39 | 56.00 ± 0.42 | 59.73 ± 0.63 |
Chet | 19.24 ± 0.33 | 26.06 ± 0.33 | 34.24 ± 0.26 | 42.11 ± 0.21 | 46.69 ± 0.21 | 55.23 ± 0.34 | 55.54 ± 0.41 | 62.96 ± 0.56 |
Mila | 21.97 ± 0.04 | 27.36 ± 0.11 | 37.31 ± 0.06 | 42.97 ± 0.09 | 49.13 ± 0.15 | 56.05 ± 0.22 | 59.52 ± 0.15 | 63.41 ± 0.20 |
4.9. Parameter Sensitivity Analysis
5. Conclusions
Author Contributions
Funding
Data Availability Statement
Conflicts of Interest
References
- Chen, C.; Liu, Y.; Sun, X.; Di Cairano-Gilfedder, C.; Titmus, S. Vehicle maintenance prediction using deep learning with GIS data. In Procedia CIRP; Elsevier: Amsterdam, The Netherlands, 2019; Volume 81, pp. 447–452. [Google Scholar]
- Hu, H.; Luo, H.; Deng, X. Health monitoring of automotive suspensions: A lstm network approach. Shock Vib. 2021, 2021, 6626024. [Google Scholar] [CrossRef]
- Begni, A.; Dini, P.; Saponara, S. Design and test of an lstm-based algorithm for li-ion batteries remaining useful life estimation. In Proceedings of the International Conference on Applications in Electronics Pervading Industry, Environment and Society, Genoa, Italy, 26–27 September 2022; Springer: Berlin/Heidelberg, Germany, 2022; pp. 373–379. [Google Scholar]
- Dini, P.; Basso, G.; Saponara, S.; Romano, C. Real-time monitoring and ageing detection algorithm design with application on SiC-based automotive power drive system. IET Power Electron. 2024, 17, 690–710. [Google Scholar] [CrossRef]
- Zhao, Y.; Liu, P.; Wang, Z.; Hong, J. Electric vehicle battery fault diagnosis based on statistical method. Energy Procedia 2017, 105, 2366–2371. [Google Scholar] [CrossRef]
- Wang, M.-H.; Chao, K.-H.; Sung, W.-T.; Huang, G.-J. Using ENN-1 for fault recognition of automotive engine. In Expert Systems with Applications; Elsevier: Amsterdam, The Netherlands, 2010; Volume 37, pp. 2943–2947. [Google Scholar]
- Jeong, K.; Choi, S. Model-based Sensor Fault Diagnosis of Vehicle Suspensions with a Support Vector Machine. Int. J. Automot. Technol. 2019, 20, 961–970. [Google Scholar] [CrossRef]
- Vasavi, S.; Aswarth, K.; Pavan, T.S.D.; Gokhale, A.A. Predictive analytics as a service for vehicle health monitoring using edge computing and AK-NN algorithm. Mater. Today Proc. 2021, 46, 8645–8654. [Google Scholar]
- Ma, M.; Wang, Y.; Duan, Q.; Wu, T.; Wang, Q. Fault detection of the connection of lithium-ion power batteries in series for electric vehicles based on statistical analysis. Energy 2018, 164, 745–756. [Google Scholar] [CrossRef]
- Zhou, Y.; Zhu, L.; Yi, J.; Luan, T.H.; Li, C. On Vehicle Fault Diagnosis: A Low Complexity Onboard Method. In Proceedings of the GLOBECOM 2020—2020 IEEE Global Communications Conference, Taipei, Taiwan, 7–11 December 2020; IEEE: Piscataway, NJ, USA, 2020. [Google Scholar]
- Shen, D.; Wu, L.; Kang, G.; Guan, Y.; Peng, Z. A novel online method for predicting the remaining useful life of lithium-ion batteries considering random variable discharge current. Energy 2021, 218, 119490. [Google Scholar] [CrossRef]
- Aye, S.A.; Heyns, P.S. An integrated Gaussian process regression for prediction of remaining useful life of slow speed bearings based on acoustic emission. Mech. Syst. Signal Process. 2017, 84 Pt A, 485–498. [Google Scholar] [CrossRef]
- Revanur, V.; Ayibiowu, A.; Rahat, M.; Khoshkangini, R. Embeddings based parallel stacked autoencoder approach for dimensionality reduction and predictive maintenance of vehicles. In IoT Streams for Data-Driven Predictive Maintenance and IoT, Edge, and Mobile for Embedded Machine Learning: Second International Workshop, IoT Streams 2020, and First International Workshop, ITEM 2020, Co-Located with ECML/PKDD 2020, Ghent, Belgium, September 14–18, 2020, Revised Selected Papers 2; Springer: Berlin/Heidelberg, Germany, 2020; pp. 127–141. [Google Scholar]
- Khoshkangini, R.; Sheikholharam Mashhadi, P.; Berck, P.; Gholami Shahbandi, S.; Pashami, S.; Nowaczyk, S.; Niklasson, T. Early prediction of quality issues in automotive modern industry. Information 2020, 11, 354. [Google Scholar] [CrossRef]
- Tessaro, I.; Mariani, V.C.; Coelho, L.S. Machine learning models applied to predictive maintenance in automotive engine components. Proceedings 2020, 64, 26. [Google Scholar] [CrossRef]
- Biddle, L.; Fallah, S. A novel fault detection, identification and prediction approach for autonomous vehicle controllers using SVM. In Automotive Innovation; Springer: Berlin/Heidelberg, Germany, 2021; Volume 4, pp. 301–314. [Google Scholar]
- Al-Zeyadi, M.; Andreu-Perez, J.; Hagras, H.; Royce, C.; Smith, D.; Rzonsowski, P.; Malik, A. Deep learning towards intelligent vehicle fault diagnosis. In Proceedings of the 2020 International Joint Conference on Neural Networks (IJCNN), Glasgow, UK, 19–24 July 2020; IEEE: Piscataway, NJ, USA, 2020; pp. 1–7. [Google Scholar]
- Guo, J.; Lao, Z.; Hou, M.; Li, C.; Zhang, S. Mechanical fault time series prediction by using EFMSAE-LSTM neural network. Measurement 2021, 173, 108566. [Google Scholar] [CrossRef]
- Safavi, S.; Safavi, M.A.; Hamid, H.; Fallah, S. Multi-sensor fault detection, identification, isolation and health forecasting for autonomous vehicles. Sensors 2021, 21, 2547. [Google Scholar] [CrossRef] [PubMed]
- Xu, P.; Wei, G.; Song, K.; Chen, Y. High-accuracy health prediction of sensor systems using improved relevant vector-machine ensemble regression. Knowl.-Based Syst. 2021, 212, 106555. [Google Scholar]
- Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A.N.; Kaiser, L.; Polosukhin, I. Attention is all you need. In Proceedings of the Advances in Neural Information Processing Systems 30 (NIPS 2017), Long Beach, CA, USA, 4–9 December 2017; pp. 6000–6010. [Google Scholar]
- Dauphin, Y.N.; Fan, A.; Auli, M.; Grangier, D. Language modeling with gated convolutional networks. In Proceedings of the 34th International Conference on Machine Learning, Sydney, Australia, 6–11 August 2017; pp. 933–941. [Google Scholar]
- Ministry of Transport of the People’s Republic of China. Decision on Amending the Provisions on the Administration of Motor Vehicle Maintenance. Order No. 14 [2023], 10 November 2023. Available online: https://www.gov.cn/gongbao/2024/issue_11146/202402/content_6930538.html (accessed on 20 February 2024).
- Lu, C.; Han, T.; Ning, Y. Context-aware Health Event Prediction via Transition Functions on Dynamic Disease Graphs. In Proceedings of the AAAI Conference on Artificial Intelligence, Online, 22 February–1 March 2022; pp. 2982–2989. [Google Scholar]
- Rumelhart, D.E.; Hinton, G.E.; Williams, R.J. Learning representations by back-propagating errors. Nature 1986, 323, 533–536. [Google Scholar] [CrossRef]
- Krizhevsky, A.; Sutskever, I.; Hinton, G.E. ImageNet classification with deep convolutional neural networks. In Proceedings of the Advances in Neural Information Processing Systems 25 (NIPS 2012), Lake Tahoe, NV, USA, 3–6 December 2012; pp. 1097–1105. [Google Scholar]
- Elman, J.L. Finding structure in time. Cogn. Sci. 1990, 14, 179–211. [Google Scholar] [CrossRef]
- Cho, K.; Van Merriënboer, B.; Gulcehre, C.; Bahdanau, D.; Bougares, F.; Schwenk, H.; Bengio, Y. Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv 2014, arXiv:1406.1078. [Google Scholar]
- Choi, E.; Bahadori, M.T.; Kulas, J.A.; Schuetz, A.; Stewart, W.F.; Sun, J. RETAIN: An interpretable predictive model for healthcare using reverse time attention mechanism. In Proceedings of the 30th International Conference on Neural Information Processing Systems, Barcelona, Spain, 5–10 December 2016; pp. 3504–3512. [Google Scholar]
- Ma, F.; Chitta, R.; Zhou, J.; You, Q.; Sun, T.; Gao, J. Dipole: Diagnosis prediction in healthcare via attention-based bidirectional recurrent neural networks. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Halifax, NS, Canada, 13–17August 2017; pp. 1903–1911. [Google Scholar]
Statistic | Value |
---|---|
Number of vehicle | 10,252 |
Maximum number of maintenance projects | 69 |
Average number of maintenance projects | 3.92 |
Number of maintenance project codes | 2427 |
Maximum number of project codes at one maintenance projects | 51 |
Average number of project codes per maintenance | project 5.21 |
Model | w-F1 | R@5 | R@10 | R@15 | R@20 | R@25 | R@30 | R@35 |
---|---|---|---|---|---|---|---|---|
SLFN | 31.05 ± 0.06 | 53.20 ± 0.11 | 59.26 ± 0.17 | 64.41 ± 0.16 | 68.59 ± 0.21 | 72.16 ± 0.22 | 75.38 ± 0.21 | 77.36 ± 0.13 |
MLP | 31.25 ± 0.16 | 53.23 ± 0.01 | 59.76 ± 0.17 | 64.93 ± 0.26 | 68.95 ± 0.27 | 72.46 ± 0.15 | 75.67 ± 0.16 | 78.05 ± 0.16 |
DBN | 31.14 ± 0.13 | 53.21 ± 0.21 | 59.88 ± 0.21 | 65.11 ± 0.32 | 68.45 ± 0.15 | 72.13 ± 0.23 | 75.86 ± 0.22 | 78.06 ± 0.31 |
CNN | 39.13 ± 1.15 | 56.01 ± 0.75 | 62.92 ± 0.79 | 67.71 ± 0.81 | 71.20 ± 0.72 | 74.02 ± 0.32 | 76.08 ± 0.34 | 77.88 ± 0.44 |
GRU | 31.44 ± 0.03 | 53.21 ± 0.03 | 59.71 ± 0.11 | 65.01 ± 0.04 | 68.56 ± 0.05 | 72.30 ± 0.03 | 75.84 ± 0.06 | 78.15 ± 0.01 |
RNN | 31.39 ± 0.07 | 53.24 ± 0.00 | 59.70 ± 0.08 | 65.02 ± 0.05 | 68.59 ± 0.03 | 72.41 ± 0.12 | 75.78 ± 0.09 | 78.14 ± 0.03 |
LSTM | 31.48 ± 0.02 | 53.23 ± 0.02 | 59.56 ± 0.08 | 65.04 ± 0.02 | 68.61 ± 0.03 | 72.30 ± 0.04 | 75.89 ± 0.03 | 78.14 ± 0.02 |
EFMSAE-LSTM | 31.62 ± 0.11 | 53.35 ± 0.23 | 59.96 ± 0.21 | 65.24 ± 0.16 | 68.66 ± 0.13 | 72.32 ± 0.24 | 75.94 ± 0.31 | 78.44 ± 0.12 |
Dipole | 31.39 ± 0.06 | 53.23 ± 0.02 | 59.88 ± 0.18 | 65.03 ± 0.03 | 68.67 ± 0.18 | 72.36 ± 0.14 | 75.84 ± 0.13 | 78.25 ± 0.09 |
RETAIN | 40.83 ± 0.42 | 56.61 ± 0.30 | 63.93 ± 0.20 | 68.58 ± 0.26 | 72.19 ± 0.25 | 75.20 ± 0.39 | 77.65 ± 0.39 | 79.47 ± 0.45 |
Chet | 39.53 ± 0.26 | 56.45 ± 0.14 | 64.20 ± 0.13 | 68.82 ± 0.12 | 72.60 ± 0.17 | 75.65 ± 0.20 | 78.23 ± 0.25 | 80.36 ± 0.19 |
Mila | 40.99 ± 0.21 | 57.25 ± 0.06 | 64.70 ± 0.15 | 69.75 ± 0.10 | 73.54 ± 0.11 | 76.23 ± 0.16 | 78.65 ± 0.07 | 80.77 ± 0.06 |
Model | w-F1 | R@5 | R@10 | R@15 | R@20 | R@25 | R@30 | R@35 |
---|---|---|---|---|---|---|---|---|
Mila-mileage | 40.25 ± 0.12 | 56.56 ± 0.15 | 64.04 ± 0.13 | 69.23 ± 0.12 | 73.05 ± 0.05 | 75.70 ± 0.15 | 78.00 ± 0.23 | 80.22 ± 0.14 |
Mila-atten | 40.53 ± 0.21 | 56.92 ± 0.19 | 64.40 ± 0.20 | 69.58 ± 0.15 | 73.31 ± 0.10 | 76.08 ± 0.15 | 78.35 ± 0.06 | 80.49 ± 0.02 |
Mila-gated | 40.56 ± 0.21 | 57.22 ± 0.11 | 64.67 ± 0.07 | 69.61 ± 0.10 | 73.31 ± 0.11 | 76.02 ± 0.14 | 78.44 ± 0.07 | 80.44 ± 0.11 |
Mila | 40.99 ± 0.21 | 57.25 ± 0.06 | 64.70 ± 0.15 | 69.75 ± 0.10 | 73.54 ± 0.11 | 76.23 ± 0.16 | 78.65 ± 0.07 | 80.77 ± 0.06 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Chen, F.; Shang, D.; Zhou, G.; Ye, K.; Ren, F. Mileage-Aware for Vehicle Maintenance Demand Prediction. Appl. Sci. 2024, 14, 7341. https://doi.org/10.3390/app14167341
Chen F, Shang D, Zhou G, Ye K, Ren F. Mileage-Aware for Vehicle Maintenance Demand Prediction. Applied Sciences. 2024; 14(16):7341. https://doi.org/10.3390/app14167341
Chicago/Turabian StyleChen, Fanghua, Deguang Shang, Gang Zhou, Ke Ye, and Fujie Ren. 2024. "Mileage-Aware for Vehicle Maintenance Demand Prediction" Applied Sciences 14, no. 16: 7341. https://doi.org/10.3390/app14167341