Light detection and ranging(LiDAR) is the core 3-D perception sensor for autonomous driving perception systems. Its housing is susceptible to contamination by dew, dirt, oil, foam, water, and other contaminants in complex road environments, leading to signal attenuation, optical path distortion, and point cloud anomalies, directly reducing the perception reliability and driving safety of autonomous driving systems. Current research mostly focuses on qualitative analysis of single contaminant and lacks quantitative modeling of multi-type contamination, correlation mechanisms between fault features and data anomalies, and reproducible simulation–measurement verification systems. To address the above issues, this study focuses on the requirements of autonomous driving functional safety and sensor health management, conducts research on modeling and quantitative analysis of performance degradation induced by LiDAR housing contamination, and establishes a quantitative relationship between contamination and performance degradation, aiming to provide support for fault diagnosis, performance evaluation, and safety design of autonomous driving sensors.

The study established a closed-loop “simulation–measurement” verification framework. An OS1-64 LiDAR was used to build a pixel-level simulation model of contamination-induced performance degradation. Based on Mie scattering theory, Snell’s law, Fresnel equations, and ray-tracing method, the model simulated the absorption attenuation, optical path distortion, and occlusion effects of six typical contaminants (dew, dirt, artificial contamination, foam, water, and oil). Simulation parameter configuration and process design were completed in accordance with the IEC 62906-5 standard (Fig.1, Fig.2). In the measurement phase, a Velodyne-16 LiDAR was used. Under controlled indoor conditions, a transparent acrylic plate was used to simulate the LiDAR housing, and uniform contaminants were applied for comparative testing. The experiments complied with IEC 60825-1 laser safety and ISO 13228:2022 performance test specifications (Fig.6, Fig.7, Fig.8). The study proposed an anomaly detection and error analysis method based on point cloud feature quantification and clustering. Using point cloud density, reflectivity distribution, and distance deviation as multi-scale evaluation indicators, the method established a normalized contamination impact coefficient model and a comprehensive point cloud quality evaluation indicator to quantitatively characterize and differentiate contamination effects (Table 1, Table 2).

Experimental results revealed that dew contamination had the most significant impact on LiDAR performance, reducing the point cloud count by more than 95% and significantly increasing distance deviation, causing the radar to almost completely lose its effective detection capability (Fig.3, Fig.4, Fig.5). Artificial contamination and dirt caused obvious point cloud loss and ranging bias, with performance degradation becoming more severe as the occluded area increased. Foam and water contamination led to moderate performance degradation. Oil contamination had the smallest influence, causing only about 5% performance degradation. The order of contamination impact from strongest to weakest was: dew > artificial contamination > dirt > foam > water > oil. Simulation and measurement results were highly consistent, verifying the accuracy and reliability of the model (Fig.10, Fig.11, Fig.12, Fig.13). The proposed multi-scale point cloud quantification method effectively distinguished different contamination types and degradation levels, providing a reliable feature basis for sensor anomaly detection.

This study establishes a complete quantitative analysis system for performance degradation induced by LiDAR housing contamination, clarifies the mechanisms and impact differences of the six typical contaminants, and provides reproducible simulation and measurement methods. The research findings can provide quantitative data and methodological support for LiDAR fault diagnosis, health management, functional safety design, and FDIIR system development, and have significant theoretical and engineering value for improving the robustness and safety of autonomous driving perception systems in complex environments. Future efforts can focus on conducting further research on the coupled effects of mixed contamination, the influence of dynamic operating conditions, and machine learning-based intelligent contamination identification to improve the LiDAR reliability evaluation system.