TY - BOOK

T1 - Modelling the causation of accidents: human performance separated system and human performance included system

AU - Wang, Yang

PY - 2023

Y1 - 2023

N2 - Millions of work-related accidents occur each year around the world, leading to a large number of deaths, injuries, and a huge economic cost. Previous studies on risk assessment have revealed that it is important to calculate the probabilities of factors that can contribute to the occurrence of accidents. Several methods, such as the Technique for Human Error Rate Prediction (THERP), have been proposed to evaluate potential risk factors and to improve system safety. However, these methods have some limitations, such as their low generalizability, treating accident causation and human factor as two separate research topics, requiring intensive data, or relying solely on expert judgement. To address these limitations, this dissertation 1) classifies systems into two types, Human Performance Separated System (HPSS) and Human Performance Included System (HPIS), depending on whether the system involves human performance; 2) develops accident causal models based on Bayesian Network (BN) that can be applied to both types of systems while examining the influence of human perception in HPSS and human errors in HPIS; 3) examines two methods for the analysis of human errors with the first method based on the cognitive view and the other method treating human errors as an essential part of the system; 4) proposes an innovative taxonomy as an example for HPIS, known as the Contributors Taxonomy for Construction Occupational Accidents (CTCOA), which not only targeting accident causation, but can also be used for tracking human error in construction; 5) builds example BN models in the different industrial sectors, including gas turbine failures as a typical example of HPSS machine failures, Multi-Attribute Technological Accidents Dataset (MATA-D) as simple HPIS failures, and Contributors to Construction Occupational Accidents Dataset (CCOAD) as complex HPIS failures. These three types of BN models demonstrate how our proposed methodology can be adapted to specific questions and how it can be applied in various industrial sectors. Our analysis demonstrates that it is efficient to combine expert judgement with mathematical independence analysis to identify the main dependency links for the BN structure in all models. The influence of human perception on model parameters should be measured before these parameters being identified based on expert judgement. Our proposed methodology can be used to study human errors by combining traditional human reliability analysis with statistical analysis.

AB - Millions of work-related accidents occur each year around the world, leading to a large number of deaths, injuries, and a huge economic cost. Previous studies on risk assessment have revealed that it is important to calculate the probabilities of factors that can contribute to the occurrence of accidents. Several methods, such as the Technique for Human Error Rate Prediction (THERP), have been proposed to evaluate potential risk factors and to improve system safety. However, these methods have some limitations, such as their low generalizability, treating accident causation and human factor as two separate research topics, requiring intensive data, or relying solely on expert judgement. To address these limitations, this dissertation 1) classifies systems into two types, Human Performance Separated System (HPSS) and Human Performance Included System (HPIS), depending on whether the system involves human performance; 2) develops accident causal models based on Bayesian Network (BN) that can be applied to both types of systems while examining the influence of human perception in HPSS and human errors in HPIS; 3) examines two methods for the analysis of human errors with the first method based on the cognitive view and the other method treating human errors as an essential part of the system; 4) proposes an innovative taxonomy as an example for HPIS, known as the Contributors Taxonomy for Construction Occupational Accidents (CTCOA), which not only targeting accident causation, but can also be used for tracking human error in construction; 5) builds example BN models in the different industrial sectors, including gas turbine failures as a typical example of HPSS machine failures, Multi-Attribute Technological Accidents Dataset (MATA-D) as simple HPIS failures, and Contributors to Construction Occupational Accidents Dataset (CCOAD) as complex HPIS failures. These three types of BN models demonstrate how our proposed methodology can be adapted to specific questions and how it can be applied in various industrial sectors. Our analysis demonstrates that it is efficient to combine expert judgement with mathematical independence analysis to identify the main dependency links for the BN structure in all models. The influence of human perception on model parameters should be measured before these parameters being identified based on expert judgement. Our proposed methodology can be used to study human errors by combining traditional human reliability analysis with statistical analysis.

U2 - 10.15488/13790

DO - 10.15488/13790

M3 - Doctoral thesis

CY - Hannover

ER -