Deploying a calibration based on near-infrared NIR spectroscopy in a factory setting often includes first creating a calibration based on laboratory data and then a calibration based on data from the factory. This has been done in this work, where the aim was to establish in-line monitoring of fat content in drycured sausages directly after the sausage stuffer. The laboratory data was also used to create an industrial calibration, with factory data used for tuning and testing. This approach has the potential to reduce the required factory samples in similar scenarios if the performance of a model based on laboratory data is sufficient. The root mean square error of prediction of the final factory validation set was 0.9% for the laboratory model and 0.8% for the factory model. The calibrated NIR spectroscopy instrument was used in process to monitor the fat variation in the sausages over two periods of 2 days each. The in-line measurements revealed that there was 0.5% variation in fat content between different batches using the same raw materials, and 0.6% variation within batches. The strategy of deploying an in-line NIR spectroscopy instrument gives efficient insight into the process variation and clues about how it can be improved for better raw material utilization and more reliable end-product quality. In this case, there was a factory upgrade between the two measurement trials. The measurements showed that the factory upgrade resulted in a reduction of the standard deviation of the within-batch fat content from 0.60% to 0.32% (after taking the measurement error into account).
Tengstrand et al. (Tue,) studied this question.