Data quality assessment of routine operating data for process identification

• Routine operating data are readily available in many historians and can easily be extracted.
• At present, there are few methods that can be used to assess the quality of the routine operating data.
• This paper presents a novel data quality assessment method using the condition number of the Fisher information matrix.
• Using simulations and theoretical analysis, it is shown that this approach agrees with the previously published results.

In many chemical engineering plants, process identification is often performed de novo each time that it is needed. However, it is quite possible that sufficiently excited data regions, including routine operating regions, have already been collected and are available for identifying particular model structures. Therefore, there is a need to develop techniques for extracting these regions from the other uninformative regions. One potential approach to solving this problem is to consider the condition number of the Fisher information matrix for the desired model structure. The sensitivity of this approach to changes in sampling time, model structure, controller type, and number of data points is also examined. It is shown, through theoretical and simulation analysis that the proposed method determines data quality based on the situation. Practically, the proposed method can be used to determine the upper bound for the process model order that may be identified from the given data.