Causation in chemical engineering education: Application of machine learning in fault diagnosis

This study integrates the design and preassessment of an exercise, incorporating a causation modeling approach into the Tennessee Eastman Process (TEP) dataset to enhance engineering students' understanding of process monitoring and fault diagnosis. The dataset, which contains 41 measured and 11 manipulated variables under normal and faulty conditions, was used to illustrate the application of the machine learning algorithm causal random forests (CRF) and treatment effect estimation in chemical process analysis. This approach differs from traditional ways of teaching/learning complex chemical engineering phenomena through governing equations, heuristics, and lab experiments. Three learning outcomes were developed for this exercise: understanding the impact of dataset composition on model interpretation, understanding how the model performance metrics differ when applied to regression and classification tasks, and understanding causality using different treatment variables. These learning outcomes were proposed to provide students with strong foundations in data integrity, model evaluation, and causal inference. In the context of engineering education, our preassessment of the effectiveness of applying this exercise to a course cohort was conducted by a summer student and closely supervised by the instructor. While the 3-hour session proved valuable and somehow feasible, some logistic challenges were gathered from this preassessment, mainly regarding time constraints and the complexity of the dataset, suggesting that breaking the exercise into smaller sessions and offering additional resources would enhance student understanding, as well as providing students with clearer explanations of technical concepts, and interactive feedback to increase engagement in future implementations.