Doctor of Philosophy (PhD)
Process operations in chemical industries are complicated, where abnormal behaviors cannot be perfectly prevented even in the best-designed plants. Once faults occur, human operators have to make rapid decisions to return the process to normal operation, often relying on heuristics, experience, and instinct. Failure to correct the faults could result in a poor quality of the product, equipment damage, or catastrophic consequences such as loss of equipment and personnel injuries.
In the past decade, process monitoring and fault detection methods have been developed based on first principle models and expert systems. Nevertheless, these approaches suffer from major drawbacks such as the high cost of model building and poor transferability of the developed expert systems, which limit their further applications. Recently, data-driven techniques such as multivariate statistics and machine learning approaches have achieved great success in different aspects including pattern recognition, regression, and classification problems. They brought huge opportunities in intelligent system development for the chemical industry, particularly for process operations and early detection of abnormal behaviors.
Therefore, this work aims at formulating and designing smart systems that integrate data-driven approaches into the chemical industry for process monitoring purposes. The proposed monitoring system includes four parts: unsupervised learning for historical data analysis, adaptive fault detection approach for drifting processes, online visualization for multiple variables, and deep learning-based infrared sensor development. Besides that, this work also addresses the generalization issue for current data-driven methods, where a general feature extraction method was discussed. A pyrolysis reactor that cracks naphtha into ethylene was used as the case study in this dissertation. The proposed framework illustrates the effectiveness of novel data-driven approaches to be applied in solving complicated decision-making problems for chemical processes. As data analytics becomes more important in many plants today, this work presents solid results on the pyrolysis reactor in the current big data context of Industry 4.0 and Smart Manufacturing.
Zhu, Wenbo, "An Integrated Framework for Chemical Process Monitoring Towards Smart Manufacturing" (2020). LSU Doctoral Dissertations. 5372.
Romagnoli, Jose A.
Available for download on Monday, October 11, 2027