Generation of causal networks and model of influence propagation

• Autor(a) principal: Azevedo, Inês Enes
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/39496
• topic: Bayesian networks; Directed Acyclic Graphs; Root cause analysis; Fault detection and identification; Industrial processes; Influence propagation; Causal analysis; Random graph generation

The general propagation of causal influence poses significant challenges in complex systems. In particular, in industrial processes, fault occurrence and propagation results in economic losses. This thesis focuses on the identification, analysis and prediction of faults and anomalies. To study these problems effectively, it proposes and characterizes methods for generating random Directed Acyclic Graphs (DAGs) with multiple paths from source nodes to a single objective node, representing the final product state in a production line. The DAG generation algorithm that was ultimately defined controls the number of source nodes and can accommodate arbitrary degree distributions. Additionally, a parameterized model of probabilities and dependencies based on an artificial Bayesian network is introduced. On this framework, a study on the influence propagation throughout the network was performed. It was found that the system undergoes a phase transition similar to percolation. The critical point is reached when the influence of the nodes’ state takes values equal to the inverse of the mean degree of the network. Below this threshold, the propagation of influence is limited to a short distance within the DAG; above it, the source nodes have the potential to influence the activation of nodes located at any given distance.