Fonterra

Background

Fonterra measures composition (fat, moisture, protein) of most of its products with a secondary method, near infrared (NIR) spectroscopy that is calibrated against reference laboratory results (the primary method). These instruments are networked and share a calibration across product classes that allows them to predict product composition. The stability of the instruments is monitored via regular ring lab trials where a representative bulk sample of product is subdivided, and a subsample is put through all the NIR instruments along with the reference labs and the results are compared.

Fonterra is interested in understanding the differences in performance between the NIR instruments (both raw spectral data and predicted results) for each ring lab trial, for multiple ring lab trials when compared to each other as well as the performance over time. We seek better definition of the source of errors and also assess the relative stability of the NIRs and effectiveness of our monitoring programme.

Details

Fonterra will provide data that includes NIR predictions, raw NIR spectral data, and reference lab info from a number of ring lab trials from a specific period with results for whole milk powder, protein powders and cheese from a single class of instruments. We’d like to understand the relative variation from the results (both predictions and raw spectral data, do we go to lab results too?) from a single ring lab trial, and then compare a number of ring lab trials across the three products groups and determine if there are any trends over time and between product groups.

To extend we’d also like to see if there are broader trends across time with the ring lab trial results, although we acknowledge that each week a different group of products is used in the ring trial. We will also present a second data set derived from a different newer group of NIR instruments with a different calibration to validate the findings from the previous work.

Outcomes

The outcomes for Fonterra could include an assessment of the relative stability of our NIRs over time and across different products, recommendations regards the frequency and nature of our ring lab trials and an understanding of the error sources in our compositional testing process.