Scholarlink

<p>Data-driven safety monitoring in general aviation (GA) faces a critical bottleneck: the scarcity of high-quality flight data containing verified anomalies, particularly across heterogeneous aircraft fleets. Existing datasets often lack physical realism or are limited to single aircraft models. Here, we present a comprehensive benchmark dataset derived from 120 real flight sorties (approximately 1 million data points) collected from Cessna 172 (R/S) and Cirrus SR20 (G1/G6) training aircraft. We developed a physics-based synthetic injection framework to generate four types of distinct anomalies: throttle surge, flight path deviation, cylinder misfire, and pitch excursion. The dataset is structured hierarchically by anomaly type and aircraft model, providing paired normal and abnormal samples for direct counterfactual analysis. We validate the dataset’s utility through statistical distribution analysis and baseline anomaly detection benchmarks. This resource bridges the gap between simulation and reality, enabling the development of robust, model-agnostic monitoring algorithms for aviation safety.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A High-Fidelity Multi-Model Benchmark Dataset for General Aviation Anomaly Detection Generated via Physics-Based Fault Injection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月25日

查看全文
  • Lu Jing,
  • Yali Fang,
  • ZiYi Huang,
  • Yuan Zhang,
  • Songhan Fan,
  • Xuan Wu

摘要

Data-driven safety monitoring in general aviation (GA) faces a critical bottleneck: the scarcity of high-quality flight data containing verified anomalies, particularly across heterogeneous aircraft fleets. Existing datasets often lack physical realism or are limited to single aircraft models. Here, we present a comprehensive benchmark dataset derived from 120 real flight sorties (approximately 1 million data points) collected from Cessna 172 (R/S) and Cirrus SR20 (G1/G6) training aircraft. We developed a physics-based synthetic injection framework to generate four types of distinct anomalies: throttle surge, flight path deviation, cylinder misfire, and pitch excursion. The dataset is structured hierarchically by anomaly type and aircraft model, providing paired normal and abnormal samples for direct counterfactual analysis. We validate the dataset’s utility through statistical distribution analysis and baseline anomaly detection benchmarks. This resource bridges the gap between simulation and reality, enabling the development of robust, model-agnostic monitoring algorithms for aviation safety.