An interpretable data-driven approach for process flowsheet convergence troubleshooting

Practitioners typically alleviate the convergence problem of process flowsheet models through manual adjustment of the convergence-related flowsheet inputs, which is labor-intensive and relies heavily on expert experience. This paper aims to realize fast troubleshooting for process flowsheets with convergence problems and proposes an interpretable approach for the adjustment of the flowsheets to liberate the manpower for process model maintenance. Specifically, the flowsheet convergence problem is addressed from a data-driven perspective for the first time. The correlation between flowsheet inputs selected according to expert knowledge and convergence status is modeled utilizing the tree-based framework to capture the flowsheet convergence behavior. In addition, a novel interpretable adjustment procedure based on an adaptive minimum mean strategy is constructed to automatically identify strongly convergence-related flowsheet inputs and provide them with quantitative adjustment suggestions. The proposed approach shows effectiveness on non-convergence flowsheets with a success rate of up to 92.5%.