Research On The Application Of Near Infrared Reflectance Spectroscopy In Microbial Fermentation Process

Solid-state fermentation(SSF) is a form of microbial fermentation, and the emergence ofenergy crisis and environmental crisis makes it to get people’s attention again in recent years.To increase the production of SSF, it is necessary that understanding the change informationsof some important process parameters in the fermentation system in real time. In actualproduction, however, due to the restrictions of characteristics of SSF and the lack of relatedtesting equipment, combined with the time-consuming disadvantage of traditional offlinephysical and chemical detection method, makes it difficult to detect rapidly of these keyprocess parameters in real time. Near infrared reflectance spectroscopy(NIRS) is widely used as animportant process analysis technology and progressive replacement of complex physical andchemical method, so which is also one of the effective ways to solve the above problembecause of its advantages in the detection of hydrogen organic compounds as well as fast and efficient,pollution-free, etc.With SSF of Monascus as the research object in this dissertation, mainly application ofNIRS rapid detection of the process variables including moisture content, pH value, glycerinand biomass. The main research works of this dissertation are summarized as follows:(1) For the SSF process of Monascus which used nonviscous bran substrate, thisdissertation mainly studied the feasibility of application of NIRS determination moisturecontent, pH value and biomass in the fermentation. On the case of using NIRS to detectmoisture content and pH value, as the disadvantages of the traditional wavelength selectionmethods used interval strategy ignoring the nonlinear factors, this paper adopted a newwavelength selection algorithm based on LS-SVM nonlinear model which names synergyinterval least squares support vector machines(siLS-SVM), and the new algorithm iscompared with correlation coefficient method, iPLS algorithm and siPLS algorithm. Studiesshowed that the combination of siLS-SVM and LS-SVM model achieved the best predictionresult. The Rp of moisture content and pH were0.9621and0.9761, respectively. TheirRMSEP were0.0129and0.1452, respectively. The obtained results demonstrated that thefitting and the predictive accuracy were satisfactory. This article also explain the wavebandsobtained by siLS-SVM algorithm, the results showed that the wavebands selected by thenew method have great consistency with near infrared spectrum characteristic absorption bandwhich generated by corresponding component’s molecular bonds. In addition, usingindependent batch samples verify the prediction accuracy of the model, also achieved greatresults. The results show that application of NIRS to detect moisture content and pH value isfeasible.(2) The feasibility for detection of biomass which has a indirect correlation with the nearinfrared spectrum was studied by use of NIRS. Glucosamine method is used to measurebiomass, by applying genetic algorithm (GA) to optimal spectral bands and establishing thePLS model to predict biomass in the SSF process of Monascus which used nonviscous bransubstrate. To illustrate the feasibility of GA to optimize spectral variables, partial least squaresregression (PLSR) was constructed with full-spectrum and the wavelengths were selected by the correlation coefficient method, respectively. the prediction ability of the three models werecomparatively analyzed, The results show that the PLS calibration model established by usingthe wavebands selected by GA has optimal predictive ability, with Rc=0.9983,RMSECV=3.5802, Rp=0.9931, RMSEP=3.6431, data points participate in modeling from theoriginal1557to585, and the model prediction accuracy improve by11.55%compared withthe full spectrum’s PLS model. This article also explained the wavebands obtained by GAalgorithm, the results show that the wavebands selected by the GA method have greatconsistency with near infrared spectrum characteristic absorption band which generated byhydrogen groups of bacteria’s composition. The NIRS method used to rapid determinatebimass was proved to be repeatable and reproducible by experimental verification. Besidesindependent batch samples was used to validate the generalization ability of the predictionmodel, good results were achieved too. The results show that the PLS model built by usingNIRS and GA algorithm could realize the rapid detection of biomass in SSF of Monascus.(3) For the SSF process of Monascus which used viscous millet9901substrate, thisdissertation mainly studied the feasibility of application of NIRS determination moisturecontent, glycerin and biomass in the fermentation. After eliminating abnormal samples,121,84,83samples were used for near infrared spectral analysis about moisture content, glycerinand biomass, respectively. Different pretreatment methods were tried for spectra pretreatment,and the best pretreatment method was selected. This chapter continued to use the KS methodto select a representative calibration samples, use GA method to select wavebands and establishedthe PLS calibration models, respectively. The study found that GA method is still suitable forwavebands selection, the selected wavebands have a good correspondence with the absorptionbands generated by corresponding covalent bonds of component’s molecules, the predictionaccuracy of moisture content’s PLS model is more ideal, with Rc=0.9894, RMSECV=0.0132,Rp=0.9827, RMSEP=0.0114, the prediction errors of glycerol’s model and biomass‘s modelare big, but the correlation coefficients between the model’s predicted values and measuredvalues are all above0.95, which show that the model can establish the mathematicalrelationship between the spectral variables and the corresponding measured values.