This approach allows you to build a predictive model relating raw material selection, blending ratios, analytical test results (CFC, GPC, DSC, etc.) and operating conditions to product quality in a single model.
Modelling for Polyolefin Manufacturers
Such a model has many applications for polyolefin manufacturers aiming to develop new and improved products. ProSensus has recently worked with leading resin manufacturers to develop such models and tackle some common manufacturing challenges including:
1. Making reliable predictions of key quality variables
Making reliable predictions of key quality variables such as Secant modulus, Elongation, Dart, Tear, etc. Upon validation, the multi-block PLS model can be utilized in optimization.
Each graph above plots one specific product quality parameter. The “observed” value corresponds to the actual (measured) value of that product quality parameter, while “predicted” indicates the value obtained from the MVA multi-block. A perfect model would result in an R2 = 1.0, RMSEE = 0.0, and all data points lying directly on the identity line (y=x).
2. Assessing the impact of different blocks for predicting product quality
This study is useful for highlighting which data block is most correlated with final product quality and can highlight where the manufacturer should focus their development efforts. ProSensus can also evaluate where there may be redundant testing being performed on the resins.
The coefficients plot above indicates which X-variables have the most overall impact on the Y-variables (product quality). Each bar represents one particular X-variable. The X-variables are colour-coded according to 4 defined blocks of data.
Here, it is readily observed that the orange block (Raw Materials & Blending Ratios) and green block (Properties of Raw Materials) have the most significant impact on product quality in the MVA multi-block model.
Further, the purple block (Analytical Tests) provide little value, suggesting that consideration could be made to eliminate these tests and their associated costs.
3. Diagnosing issues that could cause product variability
Pointing out any equipment/line/instrument differences that might lead to product variability. This is relevant to the process improvement team since it directs their troubleshooting effort to the process unit leading to inconsistent quality.
As evident in the score plot above, clustering of the datapoints forms a clear distinction between each of the 3 reactors used in this plant. In the observed versus predicted plot, it becomes clear that the product quality obtained from each reactor is significantly different.
4. Identifying materials & operating conditions
Identifying the resins, blending ratios, and operating conditions required to formulate targeted products for specific customer needs.
5. Guiding product development
Guiding the product development team to an efficient design space for future runs depending on the desired properties of the final product.
In the score plot above, a region of interest (green dashed rectangle) is readily established to target specific product quality characteristics (high dart, high tear, and low haze).
6. Narrowing down the analysis
Additionally, ProSensus can narrow down the analysis to one or two process units for any focused study of interest. To learn more about how we can help you predict quality, reduce product variability, and formulate new products, contact us.