Modelling and visualising causal relations in process chains: How do multi-linear, systemic, and combined methods affect human fault diagnosis?

• Applicants: Dr. Romy Müller; Dr.-Ing. Lukas Oehm
• Subject Area: Human Factors, Ergonomics, Human-Machine Systems
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 530537983

Machine faults in the processing and packaging industry often result from previous production steps. Such causes usually are unknown to operators, making it hard for them to correctly diagnose faults. Operator support systems could foster an understanding of causal relations in process chains, for instance by presenting causal diagrams. However, such multi-linear fault models do not adequately represent the complex interactions in a system. This can be achieved by systemic modelling methods. They reveal how different system functions interact and how this can give rise to undesirable, emergent effects. However, these models might be hard to understand and use for fault diagnosis due to their visual complexity and their lack of diagnostically relevant information about observable symptoms. Therefore, method combinations are needed, but it is unclear what information they should represent in what ways, and how this would affect human fault diagnosis. These issues are investigated in the present project. In the modelling part of the project, we model the causal relations in a process chain, using a multi-linear, a systemic, and two combined methods. The combinations are either based on causal diagrams or on networks of functional relations, in which they integrate information from the respective other method. Moreover, they provide information on different levels of abstraction, thus supporting the recognition of general principles as well as the integration of functional knowledge with concrete observations of symptoms. Subsequently, we test the generalisability of the models by transferring them to three other packaging lines that vary in their similarity to the first one. Finally, we formalise the models in a domain-specific ontology of causal relations. In the empirical part, we investigate how visualisations of the four models affect human diagnostic processes and performance, as well as their learning and understanding of causal relations. Four experiments use one of the model visualisations each, and vary two situational factors that might affect the impacts of model visualisations (i.e., complexity of faults and availability of visual highlighting). A fifth experiment compares the models in one and the same experiment. We measure participants’ speed and accuracy of selecting fault causes, their examination of relevant process parameters, and their recall and inference of causal information. The following hypotheses are tested: (1) for simple faults, causal diagrams are useful but the systemic model is not, (2) for interactions, both basic models are deficient (because causal diagrams do not adequately represent them, while systemic models are too complex and lack diagnostically relevant information), (3) the combined visualisations retain the benefits of both methods and mitigate the problems. The results serve to adapt our models and integrate the best one in a concept for operator support, which we evaluate with machine operators.

• DFG Programme: Research Grants