NIR Spectroscopy in sugar mill laboratories: An adoption issue

Research in the Australian sugar industry is focussed on improving the economic environment through productivity gains; the adoption of successful research outcomes is highlighted as a key factor in achieving this. NIR spectroscopy, while prevalent on-line, is under-utilised in the mill laboratory. Of the 24 mills in Australia, one uses a dispersive system (Foss InfraXact) and a second uses a diode array system (Foss DA1650). The cause for low industry adoption is two-fold: the high capital cost of benchtop instruments and the excessive resource investment required to develop local calibrations at mill level. The advent of diode array systems has solved the first problem. The second is more challenging.
To improve adoption, the industry requires a turn-key solution that demonstrates real value to the mill the moment it’s installed. However, every mill requires a unique solution due to the extreme variability in cane supply, mill processing and reference methods. Past mill trials utilising simple transfer calibrations as a look-see exercise has had damaging consequences.
Global calibrations are proposed as a potential bridging solution and will likely provide acceptable interim results, until local calibrations can be developed. Generating calibrations within the resource restraints of the industry required leniency with best-practice and an acceptance that this will be corrected during localisation. Sampling capitalised on the mills’ standard daily activities by collecting sample residues from routine testing in the 2014 season. Samples and the associated reference data were sent to a central research facility for NIR analysis. Data was also sourced from the InfraXact and DA1650 instruments already stationed in mills.
Spectra were treated with a derivative (1st, G1S6; 2nd G1S9) and SNV transform; PLS was used to develop calibrations for bagasse (pol and moisture), molasses (sucrose and dry substance) and raw sugar (pol and moisture). The value of the large InfraXact dataset was evaluated by comparing validation statistics on calibrations generated with both InfraXact and DA1650 data and only DA1650 data from two instruments. Validation was blind (new season or mill not included in calibration).
Overall, the conditions that provided the best validation statistics were unique for each product and constituent and identified that some validation is required to identify each mills’ response to the global, prior to installation. Some constituents, such as sucrose for molasses produced exceptional calibrations exclusively with DA1650 data (RMSEC 1.45, SEP 1.34, Bias -0.106). Others, such as moisture (0.04, 0.04, 0.01) and pol (0.13, 0.25, 0.07) in raw sugar benefitted from the increased sample representation of the combined calibration set. Moisture for bagasse showed a significant but consistent bias (0.65, 0.57, -2.44), which would be easily remedied in-mill with an adjustment. All calibrations produced errors that approach those of the reference methods.
Bagasse, molasses and raw sugar are key control points for the optimisation of the sugar factory to minimise losses and impurities. It has been identified that, while not perfect, global calibrations will be a suitable bridging procedure to increase the number of NIR instruments in mill laboratories, allowing subsequent high-value, local calibrations to be developed.