| Due to the restrictions of characteristics of solid-state fermentation (SSF) and the lack of related testing equipment, time-consuming and labor-consuming traditional offline physical and chemical detection method was the major measurements, it is difficult to realize rapidly detecting of process parameters, thus influence the efficiency of detection and control in SSF. Near infrared reflectance spectroscopy (NIRS) is widely used as an important process analysis detection technology because of its fast and accurate, pollution-free, etc. advantages.With SSF of Monascus as the research object in this dissertation, the feasibility of rapidly detection of key parameters including moisture content, pH value, glycerol and biomass by NIRS was investigated. The main resluts of this dissertation are summarized as follows:(1) The relation between biomass and glucosamine content was investigated; the relation in wheat bran liquid medium(WPL), wheat bran agar medium(WPA) and liquid medium(LM) was investigated, the results showed that the correlation of these two parameters was linear, the glucosamine content of per gram of mycelium was an important correlation coefficient to detect biomass by glucosamine method and subjected to the influence of the cultivation time, cultivation method and the medium composition. On the base of the relation of biomass and glucosamine content, calculat were built for biomass detection in SSF of M. purpureus zh and9901, which used wheat bran and millet as substrate. It laid the foundation for biomass detection by NIRS via glucosamine content measurements.(2) For the SSF process of Monascus when non-sticky wheat bran used as substrate, the feasibility of application of NIRS to detect moisture content, pH value and biomass was studied. Different spectra pretreatment methods were tested firstly, and then based on the best pretreatment methods, moisture content, pH value and glucosamine PLS models were established by three wavelength selection methods including correlation coefficient method, iPLS algorithm and siPLS algorithm. The result showed that Normalization, Savitaky-Golay smoothing were the best pretreatment for moisture content, pH value respectively, and the best pretreatment of glucosamine was the combination of Normalization and Savitaky-Golay smoothing. Comparison with correlation coefficient method and Ipls, siPLS algorithm exhibited better performance in wavelength selection and most significantly improved the PLS calibration models. RMSECV of moisture content, pH value and glucosamine content models was reduced by13.5%,7.12%'24.8%, respectively. To further improve the glucosamine calibration model, the samples were pretreated by dried and crushed method, and then use siPLS algorithm to select wavelength for model establishment, RMSECV was reduced to0.1973. The desired results were achieved by using calibration models and the built biomass calculation to predict moisture content, pH value and biomass of the separate batch samples, the mean relative error of predicted and actual values of was1.857%、2.735%、5.090%, which illustrated that the application of NIRS was feasible for detecting the parameters in SSF of Monascus.(3) On the basis of the method for NIRS models eatablished in wheat bran substrate,application of near infrared spectroscopy for rapid detection moisture content, glycerol andbiomass in SSF of Monascus using sticky particles millet as substrate was studied. Aftereliminating abnormal samples and the comparison of different pretreatment methods, PLScalibration models were established based on siPLS algorithm respectively. RMSECV ofmoisture content, glycerol and glucosamine models were2.1543,8.5921,0.3292, res-pecpectively. The predicted values of moisture content, glycerol and biomass could reflect thechanges of the parameters in external samples prediction, which indicated that the modelshave good predictive performance. |
PDF Full Text Request