Abstract
As the automotive market matures, after-sales service has become a primary profit driver. However, traditional "passive" service models—relying on fixed schedules or reactive repairs—often fail to meet individual needs. This paper proposes a data-driven inference method combining Rough Set theory and Evidential Reasoning (ER) to provide personalized maintenance recommendations. By analyzing driving behavior data from FAW-Volkswagen, the study demonstrates that this hybrid approach effectively identifies critical maintenance needs with high statistical confidence.
Introduction: Moving Beyond Passive Service
In modern automotive markets, after-sales services account for over half of industry profits. Despite this, many providers still operate under a passive service model. This approach typically follows rigid, time-based or mileage-based schedules that ignore the nuances of individual driving habits. A city commuter and a long-distance mountain driver wear down components at vastly different rates; a one-size-fits-all schedule leads to either premature replacements or missed maintenance windows, ultimately lowering customer trust and service quality.
The Proposed Framework
To address these challenges, the researchers developed a three-stage inference model designed to handle the inherent uncertainty and "fuzziness" of driving data.
1. Attribute Reduction via Rough Set Theory
Driving behavior involves numerous variables, many of which are redundant. Using an information entropy reduction algorithm based on Rough Set theory, the model identifies "core attributes"—the specific factors that most significantly impact the health of vehicle components. For the brake system, analysis revealed that mileage interval, average driving speed, and road conditions were the most critical predictors.
2. Weighting Evidence
Rather than relying on subjective expert opinions, the system employs an attribute importance algorithm. This mathematically weights each core attribute based on its objective influence on maintenance outcomes, ensuring the model remains data-driven.
3. Evidential Reasoning Synthesis
The final stage uses an Evidential Reasoning (ER) synthesis algorithm to combine weighted evidence. This recursive process calculates a "belief degree" for various service categories. A recommendation is only considered valid if the top choice exceeds the runner-up by a confidence margin ($\epsilon = 0.2$).
Case Study: FAW-Volkswagen Brake Systems
The model's effectiveness was validated using real-world data from 60 customers driving FAW-Volkswagen SAGITAR models. The system categorized service needs into three levels:
- Simple maintenance (cleaning/fastening).
- Inspection and brake oil replacement.
- Major component replacement (brake pads/blocks).
Key Results
The method demonstrated high precision in identifying personalized service requirements. In test cases, the model consistently produced a clear "winner" among service options.
As shown in the figure above, the belief degrees are clearly differentiated. In all validated samples, the difference between the maximum belief degree and the second-largest was greater than 0.2, meeting the system's strict confidence threshold. For example, in one specific test case, the belief degree for "Inspection and oil replacement" reached 0.716, significantly outperforming alternative recommendations.
Conclusion
By integrating Rough Set theory for feature selection and Evidential Reasoning for decision-making, this method provides a robust pathway for personalized automotive maintenance. While future work is needed to address conflicting evidence and expand the model to other vehicle subsystems like engines, this framework represents a significant step toward proactive, data-driven automotive care.
Summary updated