基于混合专家零样本异常检测模型的机器人化飞行器起飞前检查系统及方法
A Robotic System for Automated Aircraft Pre-Flight Inspection Using a Mixture-of-Experts Zero-Shot Anomaly Detection Model

针对传统飞行器人工检查效率低下，以及现有自动化系统存在检测盲区和难以识别未知缺陷等问题，本文提出了一种新型自主移动机器人检查系统。该系统利用搭载多模态传感器的移动机器人进行全方位数据采集，核心算法采用本文提出的混合专家零样本异常检测(MoE-ZSAD)模型。该模型通过将多维图像特征与正常状态的文本描述进行语义对齐，在无需缺陷样本训练的情况下，实现了对各类已知及潜在未知缺陷的精准识别。实验结果表明，该模型在复杂场景下的缺陷检测中表现优异，实现了超过0.96的像素级平均受试者工作特征曲线下面积($Pixel\text{-}AUROC$ )，在检测精度和泛化能力上均显著优于当前的基准模型。分析显示，该方法有效克服了传统视觉检测对异常样本的依赖局限，验证了利用跨模态技术解决工业检测难题的可行性。该研究不仅为飞行前自动化检查提供了一种高效、客观且具备强适应性的解决方案，显著提升了航空维修保障的安全性与运营效率，也为未来大型装备的智能化运维提供了新的技术路径。
To address the low efficiency of traditional manual aircraft inspection and the issues of detection blind spots and difficulty in identifying unknown defects in existing automated systems, this paper proposes a novel autonomous mobile robotic inspection system. The system employs a mobile robot equipped with multi-modal sensors for comprehensive data acquisition, with the proposed Mixture-of-Experts Zero-Shot Anomaly Detection (MoE-ZSAD) model serving as the core algorithm. By semantically aligning multi-dimensional image features with textual descriptions of normal states, the model achieves precise identification of various known and potential unknown defects without requiring training on defect samples. Experimental results demonstrate that the model performs exceptionally well in defect detection within complex scenarios, achieving a Pixel-Level Area Under the Receiver Operating Characteristic Curve ($Pixel\text{-}AUROC$ ) exceeding 0.96. It significantly outperforms current benchmark models in terms of both detection accuracy and generalization ability. Analysis indicates that this method effectively overcomes the limitations of traditional visual detection regarding dependency on anomalous samples and validates the feasibility of utilizing cross-modal technology for industrial inspection. This study not only provides an efficient, objective, and highly adaptive solution for automated pre-flight inspection, significantly enhancing aviation maintenance safety and operational efficiency, but