以資料為依據飛航安全風險評估模型之建構與應用

Abstract

與飛航安全有關的風險管理一直是大眾關注的焦點，近年來國際上針對該項風險管理的研究重心逐漸從層級較低的風險因素識別轉往層級較高的資源有效配置，而在配置過程中，如何量化不同事故及不同狀態的風險成為準確決策的一項關鍵議題。就此，本研究針對各類飛航事故發生率及其損失建構分析模型，該模型除可分析既有資料數據，藉以識別高風險事故外，也可依貝葉斯網路(Bayesian network)，透過新收集到的數據，更新模型參數。該模型主要係以狄利克雷分布(Dirichlet distribution)描述各類事故發生機率，並用以評估事故損失；兩者可供風險矩陣的繪製，並可藉以建構反帕雷托前沿，做為高風險事故之識別準則。經應用該模型分析所收集到我國、美國及歐盟的飛航事故資料，本研究發現，整體而言，我國近年來的飛航安全略優於美國、但不如歐盟；就事故識別而言，影響我國飛航安全之高風險事故類型包含偏出跑道(Runway Excursion, RE)、可控飛行撞地(Controlled Flight Into Terrain, CFIT)、非動力相關的零件組失效(System/Component Failure or Malfunction (non-powerplant), SCF-NP)、在飛行中失控(Loss of Control – Inflight, LOC-I)等，這些高風險事故值得管理當局配置有效資源予以防範。

Risk management related to aviation safety, which is sometimes called “aviation risk management” in brief, is important to the public. Its research has shifted gradually from risk identification to the more ambitious resource allocation in recent years. The purpose is to allocate resources appropriately for preventing various kinds of risk-event based on the result of quantitative risk assessment of those events. To that end, a quantitative risk assessment model is proposed in this thesis. The model employs Dirichlet distribution to describe probabilities of various risk events and evaluate their potential losses. The result is then used for constructing a risk matrix and an Anti-Pareto frontier. The former can be used for risk management including resource allocation and the latter can be used for setting criteria of high-risk events. Bayesian network is also employed in the model for updating model parameters based on available data. Aviation accident data of Taiwan, the United States and European Union are collected and analyzed based on the proposed model. It is found that, in terms of overall aviation safety, the situation of Taiwan is slightly better than that of the United States but slightly worse than that of the European Union. It is identified that high-risk aviation events in Taiwan are runway excursion (RE), controlled flight into/toward terrain (CFIT), system/component failure or malfunction (non-power plant) (SCF-NP) and loss of control–inflight (LOC-I), listed in sequence. These high-risk events deserve the authority to allocate appropriate resources for improvement.