A methodology to diagnose process/model mismatch in first-principles models

A methodology is proposed to diagnose the root cause of the process/model mismatch (PMM) that may arise when a first-principles (FP) process model is challenged against a set of historical experimental data. The objective is to identify which model equations or model parameters most contribute to the observed mismatch, without carrying out any additional experiment. The methodology exploits the available historical dataset and a simulated dataset, generated by the FP model using the same inputs as those of the historical dataset. A data-based model (namely, a multivariate statistical model) is used to analyze the correlation structure of the historical and simulated datasets, and information on where the PMM originates from is obtained using diagnostic indices and engineering judgment. The methodology is tested on two simulated systems of increasing complexity: a jacket-cooled continuous stirred reactor and a solids milling unit. It is shown that the proposed methodology is able to discriminate between parametric and structural mismatch, pinpointing at the model equations or model parameters that originate the mismatch.