A causal model of performance shaping factors for human reliability analysis in manufacturing.

Human reliability analysis is a critical component of probabilistic risk assessment, aimed at predicting and mitigating human errors in complex systems. This dissertation develops a novel approach to human reliability analysis in manufacturing by integrating structural equation modeling and Bayesian networks to improve the estimation of human error probabilities. Traditional human reliability assessment methods, such as the Standardized Plant Analysis Risk-Human Reliability Analysis (SPAR-H) and the Technique for Human Error Rate Prediction (THERP), provide structured techniques for estimating human error probabilities. However, these methods often fail to capture the complex interdependencies among performance shaping factors (PSFs), limiting their applicability in dynamic manufacturing environments. To address this limitation, this study constructs a hierarchical causal model of PSFs, categorizing them into internal factors—such as stress, experience and training, and physical and cognitive abilities—and external factors, including task complexity, human-machine interface, and work processes. The research begins by identifying inconsistencies in existing PSF taxonomies and proposes an improved framework that captures the interrelationships among PSFs. A survey-based data collection process was conducted to validate the framework and gather expert assessments. Structural equation modeling, using a multiple indicators multiple causes (MIMIC) approach, is applied to confirm causal relationships within the PSF model, demonstrating that external factors significantly influence internal factors, which subsequently affect human reliability. A Bayesian network model is developed to estimate human error probabilities and is validated through a case study in ceramic filter manufacturing. The analytic hierarchy process is used to derive expert-based weightings, while sensitivity analysis identifies the most influential PSFs. Comparative evaluation with SPAR-H illustrates the enhanced predictive accuracy of the Bayesian model, especially for complex tasks involving multiple interacting factors. The findings provide valuable insights for improving human reliability in manufacturing and bridge theoretical contributions with practical applications. The proposed Bayesian model supports a probabilistic, data-driven approach to error estimation, making it adaptable to evolving industrial contexts. Future research directions include the integration of real-time operational data, the incorporation of artificial intelligence and machine learning for adaptive modeling, and broader validation across diverse industrial sectors.