(127a) Performance Monitoring of Global Chemometric Models in Manufacturing Plants: New Approaches for Efficient Model Maintenance

Online FTIR is commonly employed for physical property and composition analysis in manufacturing plants. In general, the most common approach involves utilization of local, site-specific chemometric models to predict properties that would otherwise require labor intensive offline analysis. This approach can only be efficiently used for a few products. Performance monitoring in this type of setting is routinely carried out by periodically collecting audit samples and comparing the model prediction to a primary analysis. The choice of products to be monitored is a straight forward process due to the limited scope. Results of such auditing processes are key for making maintenance decisions from a hardware or chemometric model perspective.

Dow’s Polyethylene business aims for a consistent global quality assurance approach with globally harmonized analytical tools and capabilities. Part of this strategy approach requires the creation of chemometric models that can be successfully employed for continuous analysis of multiple properties and products on all production trains globally. In such a scenario, traditional model maintenance approaches fail to produce an accurate maintenance plan that can facilitate proactive and targeted use of resources for improved future model performance. In order to overcome these shortcomings, various tools focused on quality, equivalency and capability have been developed. This presentation will discuss the designed audit program system that allows alignment of information collected using the traditional approach with process data that can provide additional insight regarding model prediction performance. Detailed analysis enables early identification of the biggest opportunities for improvement and maintenance.