Software-Defined Vehicle (SDV) employs centralized high-performance controllers and advanced E/E architectures capable of acquiring and fusing large volumes of heterogeneous sensor data in real time. This enables virtual sensing and the application...
Software-Defined Vehicle (SDV) employs centralized high-performance controllers and advanced E/E architectures capable of acquiring and fusing large volumes of heterogeneous sensor data in real time. This enables virtual sensing and the application of AI-based anomaly detection and predictive maintenance techniques, supporting more precise and proactive diagnostic strategies. However, most existing AI-driven diagnostic studies require large-scale training datasets and substantial computational resources, presenting considerable challenges for practical deployment in terms of Domain Adaptation. This study focuses on chassis components—specifically shock absorbers and wheel bearings—as diagnostic targets. Physics-informed modeling and major fault-mode analyses were conducted using domain knowledge, and experimental data were collected from both healthy and fault-replicated components using high-precision external sensors and production-grade low precision sensors. Based on these two sensor modalities, the expressiveness of fault signatures was examined, and optimal signal-processing techniques were selected to extract key fault features. A Knowledge Distillation–based diagnostic framework was then developed to transfer the representational capability of high-precision sensor data to models operating solely on low-precision sensor inputs. This approach improves diagnostic accuracy under production-sensor constraints and enables a deployable chassis fault-diagnosis framework suitable for real-vehicle implementation. Additionally, the study evaluates the performance gains achieved by incorporating domain-knowledge-derived fault features into the model training process. The proposed physics-informed Knowledge Distillation model demonstrates practical Domain-Adaptation capability and embodies a paradigm shift from default threshold-based diagnosis to deployable, anomaly-detection–centric chassis fault diagnosis for next-generation SDV.