Risk Analysis of Autonomous Mobile Robots Using a Hybrid Fault Tree-Bayesian Network = Fault Tree - 베이지안 네트워크를 이용한 자율 이동 로봇의 실시간 위험도 분석

다국어 초록 (Multilingual Abstract)

With the recent advancements in artificial intelligence and sensor technology, Autonomous Mobile Robots (AMRs) are playing an increasingly pivotal role in environments that are difficult or hazardous for human intervention, such as reconnaissance, precision agriculture, logistics, and disaster response. As the missions of these robots become more sophisticated and critical, ensuring high levels of system safety and reliability to prevent mission failures or accidents from unforeseen faults has become a paramount challenge. Consequently, there is a pressing need for a systematic Prognostics and Health Management (PHM) framework capable of monitoring the robot's operational status in real time to predict and diagnose potential faults[1]. Traditional methodologies like Fault Tree Analysis (FTA) have been widely used for system reliability analysis, offering the advantage of clearly visualizing the logical relationships between a top-level system failure (Top Event) and its root causes (Basic Events). However, FTA is an inherently static model, which poses significant limitations in dynamically estimating reliability by reflecting real-time environmental changes and complex probabilistic interdependencies within the system, or in updating existing reliability assessments when new evidence (e.g., anomaly signals) emerges. To overcome these limitations, this paper proposes a hybrid Fault Tree-Bayesian Network (FT-BN) framework that integrates the structural analysis capability of FTA with the powerful probabilistic reasoning of Bayesian Networks (BN) to perform dynamic reliability analysis and real-time health monitoring for AMRs. The proposed framework follows a systematic procedure. First, a Software-In-the-Loop (SITL) environment, linking ArduPilot and the Gazebo simulator, is established to acquire diverse sensor log data (.bin) representing both normal and abnormal states through intentional fault injection. Second, the collected raw log data is parsed, and the Interquartile Range (IQR) technique is utilized to detect statistical anomalies in key sensor data streams, with each detected anomaly being defined as a Basic Event. Third, an FT is manually designed with 'Mission Failure' as the Top Event based on these Basic Events and expert knowledge. Finally, a BN model capable of dynamic inference is constructed by systematically mapping the logical structure of the FT into Conditional Probability Tables (CPTs). Experiments were conducted using log data from the simulation environment to validate the proposed framework. A sensitivity analysis, performed to assess the impact of each basic event on the top event, quantitatively identified that an anomaly in the magnetometer (MAG) sensor contributes the most significant risk to mission failure, making it the most critical single point of failure. This study presents a data-driven framework for real-time system health diagnosis and reliability assessment based on operational log data. Its academic significance lies in moving beyond the qualitative analysis of conventional FTA by probabilistically modeling system uncertainties and enabling dynamic belief updating based on new evidence. Furthermore, by quantitatively identifying the system's most critical vulnerabilities, this work provides practical guidelines for prioritizing fault-tolerant design and predictive maintenance strategies. Future work will focus on extending this research to a Dynamic Bayesian Network (DBN) to model fault transition processes over time and applying the framework to physical robot hardware for validation in real-world operational scenarios.

국문 초록 (Abstract)

자율 이동 로봇(AMR)의 운용 환경이 점차 복잡해짐에 따라, 예측 불가능한 고장을 사전에 방지하고 시스템의 안전성을 확보하는 것은 필수적인 과제가 되었다. 기존의 결함 트리 분석(FTA)은 시스템의 논리적 구조를 파악하는 데 유용하지만, 정적 모델의 한계로 인해 실시간 환경 변화와 구성 요소 간의 복잡한 확률적 상호의존성을 반영하는 데에는 어려움이 있다.
이에 본 연구에서는 FTA의 구조적 분석 능력과 베이지안 네트워크(BN)의 확률적 추론 기능을 결합하여, AMR의 실시간 신뢰성 평가를 위한 하이브리드 FT-BN 프레임워크를 제안하였다. 먼저, SITL(Software In the Loop) 시뮬레이션 환경을 구축하여 다양한 주행 시나리오의 로그 데이터를 수집하고, 사분위수 범위(IQR) 기법을 적용해 센서 데이터의 통계적 이상 징후를 탐지하여 이를 기본 이벤트로 정의하였다. 이후 설계된 FT를 BN으로 변환하여 실시간 데이터에 기반한 동적 확률 추론 모델을 구현하였다.
실험 결과, 제안된 모델은 시스템의 고장 확률을 효과적으로 갱신하였으며, 위험 달성 가치(RAW) 분석을 통해 임무 실패에 가장 치명적인 영향을 미치는 요인을 정량적으로 규명하였다. 본 연구는 데이터 기반의 동적 위험도 분석을 통해 시스템의 핵심 취약점을 식별하고, AMR의 결함 허용 설계를 위한 실질적인 방법론을 제시한다는 점에서 의의가 있다.

목차 (Table of Contents)

• Chapter 1. Introduction 1
• 1.1. Background and Motivation 1
• 1.2. Research Objectives 3
• 1.3. Research Methods and Scope 4
• 1.4. Thesis Outline 6
• Chapter 2. Theoretical Background and Related Work 8
• 2.1. Autonomous Mobile Robot System 8
• 2.2. System Reliability Analysis Methodologies 11
• 2.3. Review of Related Work on FT-BN Hybrid Models 21
• Chapter 3. Proposed Rover Monitoring Framework 24
• 3.1. Overall System Architecture 24
• 3.2. Data Acquisition and Simulation Environment Setup 27
• 3.3. Data Preprocessing and Anomaly Detection 30
• 3.4. Construction of the Rover Simulation FT-BN Model 35
• Chapter 4. Experimental Results and Analysis 44
• 4.1. Experimental Environment and Dataset 44
• 4.2. Analysis of BE Probability and System Failure Probability 45
• 4.3. Importance Analysis using Risk Achievement Worth (RAW) 46
• 4.4. Fault Injection Simulation and Validation 49
• Chapter 5. Conclusion 55
• 5.1. Summary of Research 55
• 5.2. Contributions 56