The Identification of Forest Bio-Product Process Components through Near-Infrared Spectroscopy
The Identification of Forest Bio-Product Process Components through Near-Infrared Spectroscopy
Abby Hamilton
Advisor: Dr. Darrell Donahue
Near-infrared spectroscopy (NIRS) has the potential to advance the productivity of the forest bio-refinery process by rapid identification of material components comprising of liquid extract and woody biomass. The potential exists for composition identification via NIRS to be performed as an in-line process control operation. Before this technology is applied to the forest bio-refinery process, a NIR spectral database of solid wood chips and liquid extract solutions must be developed and analyzed. Model liquid extracts with known compositions were generated in the laboratory while wood chips pre- and post-extraction were acquired from a laboratory-scale bio-refinery process. After developing the database from collected extract and wood chip spectra, partial least squares (PLS) techniques were used in combination with selected pretreatments to develop regression models. Three data pretreatments including standard normal variate (SNV), first derivative and second derivative were completed separately and then compared. The best fit models were then validated by comparing them to spectra of other wood chips and actual liquid extracts removed during a laboratory-scale bio-refining process. Pre-extracted wood chip spectra had a greater magnitude of reflectance than the post-extracted wood chip spectra. Significant differences were seen when a water spectrum was subtracted from liquid extract spectra. First derivative models based on known woody biomass components indicate positive validation results. In order to improve these PLS models, a narrower wavelength range will be used to attempt to optimize regression values. A subtraction data pretreatment may be used to remove the water signal from all of the liquid extract spectra. With the improved PLS models, the components of known composition can hopefully be predicted more accurately. These models may also help predict the composition of actual liquid extract to see if the composition is comparable to the model liquid extract composition. The results to date support the potential for advancement in the identification of extract components via NIRS. With further development of the spectral database and with additional improvements of the PLS models, the identification technique could become more practical for use in industry.
Interview with Abby on July, 11, 2007
Abby was mentoring an Old Town High School student at the time of this interview
