From research laboratory to sugarcane factory – technology transfer challenges

Near Infrared Spectroscopy (NIRS) has been investigated as an alternative method for factory stream analysis in the South African sugarcane industry. Requirements of the technology include:
• Global calibration models that encompass processing streams originating from different geographical regions, extraction technologies and agricultural practices
• NIRS predictions giving equivalent results to conventional methods of analysis (results used in an industry-wide payment scheme)
• To be as robust as existing methods.

Transmission NIRS was successfully implemented in a research environment at the Sugar Milling Research Institute NPC (SMRI) for the prediction of up to seven analytes (brix, pol, sucrose, glucose, fructose, conductivity ash and dry solids) in nine factory stream samples for many years. However, transfer of the technology to sugar mill laboratories proved challenging due to apparent non-transferability of calibrations between instruments. Problem identification, resolution and development of new calibrations were required to enable the technology to be transferred.

Resolution was conducted in the following experimental areas:
1. Calibration non-transferability. Preliminary studies showed that the non transferability of calibrations between 16 Bruker MPA instruments was related to the type of cell used in the instruments. The background spectra of the original cells supplied to the mills and two alternative cell types were investigated.
2. Development of new calibrations. Replacement global calibrations were developed (nine process streams, seven analytes) using 16 instruments, 38 cells and a limited temperature range (18 24°C).
3. Validation of the calibrations using independent sets of samples not used in the development of the calibrations.
4. Development of robust sample introduction methods.
5. Parallel analysis. NIRS analysis of Mixed juice (MJ) and final molasses samples (1026 samples) were run in parallel to the conventional analysis methods for the 2014-15 season at SMRI. Similarly, MJ samples were run in parallel in 14 independent mill laboratories (69 000 samples) during the same period.

The original calibrations were developed using an OS cell which, on investigation, was found to contain residual water in the glass, the amount of which varied from cell to cell. The absorbance due to water in the original cell had been incorporated into the calibrations resulting in non reproducible analyte predictions when used with different cells. The residual water content of quartz based cell types were investigated and a suitable replacement cell identified. This necessitated the development of new calibrations allowing the incorporation of cell (38), instrument (16) and temperature variability to increase the robustness of the calibrations. Individual PLS calibrations were developed for each analyte within each factory stream. The calibrations typically provided RMSEP values of 0.06-0.09% with calibration R2 of 0.998. Validation of the models against independent sample sets showed no significant statistical difference between the NIRS predictions and conventional results. Introduction of the technology at the mills required the development of mass/mass dilutions procedures and robust sample introduction techniques cloned from liquid chromatography. NIRS has been run in parallel with conventional analysis for more than one year at both SMRI and 14 mills and has shown analytical equivalence.