【作者】 郭伟良；

【导师】 滕利荣；

【作者基本信息】 吉林大学 ， 微生物与生化药学， 2010， 博士

【摘要】 近红外(NIR)光谱分析技术由于具有无前期样品预处理程序、无需损坏样品、无需大量有机化学试剂、分析速度快、成本低、可多组分同时测定、在光纤探头的辅助应用下可实现在线检测和反馈控制等众多的优点,其伴随着计算机科学的快速发展而逐渐得到完善和广泛应用的绿色环保分析技术。NIR技术具有众多优点的同时,也存在固有的弱点,由于其主要检测的是分子振动倍频,信号比较弱,在涉及到水溶液的检测过程受水的严重干扰,同时NIR光谱技术是一种间接的分析技术,建立稳健性好,泛化能力强和预测能力强的校正模型是关键,本文考察了不同的光谱预处理方法,包括卷积平滑(Savitzky-Golay)、快速傅立叶变换光滑(Fast Fourier Transform Smoothing)、一阶导数(First Order Derivative)、二阶导数处理(Second Order Derivative),标准正态变换(Standard Normal Variate)和小波变换(Wavelet Transform)等光谱预处理方法对样品的NIR光谱进行预处理时,去除光谱中随机噪声和提取有效信息的效果。采用蒙特卡罗偏最小二乘法(MCPLS)识别异常样本。分别尝试采用偏最小二乘法(PLS)和径向基神经网络(RBFNN)法建立模型,采用可移动窗口法筛选两种模型的波长变量,并以逼近度(Da)为指标,对两种模型的建模参数进行了优化。本文将NIR分析技术应用到发酵工艺分析领域。主要包括三方面的内容:(1)采用NIR光谱技术同时快速无损检测蛹虫草发酵产品蛹虫草菌丝体粉末样品中腺苷、蛋白质、多糖和虫草酸含量,以有效评价蛹虫草发酵产品质量,可推广应用于蛹虫草优良菌种的筛选和发酵条件优化等领域。(2)采用NIR光谱技术实时监测蛹虫草发酵过程中发酵液中生物量、胞内多糖、胞外多糖、虫草酸、腺苷和葡萄糖含量。为蛹虫草发酵工艺优化,参数自动控制和提高蛹虫草发酵水平做铺垫,同时也为NIR光谱技术在真菌发酵工艺实时监测中的应用进行可行性探索。(3)采用NIR光谱技术实时监测乳酸乳球菌发酵过程中发酵液中乳链菌肽效价、葡萄糖含量、pH和生物量,为NIR光谱技术在细菌发酵工艺优化和自动控制做铺垫。更多还原

【Abstract】 NIR technology is an non-preprocessing, non-destructive, non-pollution analysis technology. It can simultaneously determine multi-ingredients and it is a fast, low cost analysis technology. It is great interested in chemical analysis regions. Since the NIR spectrum is molecular vibrational spectrum which is weak and much complex. It is difficult to comprehensive parse. This problem can be solved since the compute technology and chemometrics are greatly developed. This technology can be applied to deal with the numerous NIR spectrum data and extract the effective messages. NIR technology was applied in agriculture, chemical, medicine, pharmaceuticals, food and life science and so on. There are several literatures on the application of NIR spectroscopy on fermentation processes especially on at line/on line monitoring key parameters. There are few of them in china. There are several reasons obstruct apply NIR technology to fermentation. Firstly there is much water in the broth, which seriously interfere the messages. Secondly, the morphology of the bacterial, the viscosity and the pH of the broth, the components of broth are changing during the fermentation processes. It is difficult to extract the effective messages in these spectra. There are several reports on successfully applying the NIR technology on monitoring key parameters during fermentation processes. However, most of them were only limited in the same batch, the defined media, the same bacterial. The generalization, the stability and the predictive ability of the calibration models is not good enough. In this paper, the application of NIR analysis technology in examining the quality of the fermentation production, real time monitoring the key parameters during fungi fermentation and bacterial fermentation.In this paper, NIR analysis method was applied to examining the quality of the Cordyceps militaris mycelia. Adenosine, protein, intracellular polysaccharide and Cordyceps acid can be simultaneously determined using this method, while only one integrate in the production can be determined using traditional methods. The calibration models for determining the contents of adenosine, protein, intracellular polysaccharide and cordyceps acid in mycelia have been developed using NIR spectra. The samples for modeling were collected by mutants’fermentation under various fermentation conditions. These samples were representative. MCPLS was employed to examine the outliers, which can enhance the stability of model. Several preprocessing methods including Savitzky-Golay smoothing method, FFT, SNV, first order derivative method, second order derivative method and WT were applied to preprocess the NIR spectra. The effect of preprocessing windows’size was investigated. PLS and RBFNN were employed to develop the models respectively. MWPLS was applied to select the wavelengths for developing the PLS model. The number of latent variables of PLS model was selected depending on Da. MWRBFNN was employed to select the wavelengths for developing RBFNN model. The number of hidden nodes and the spread constant of the RBFNN models were selected depending on Da. A comparing study between the optimized PLS models and RBFNN models has been done and the optimum models for determining the contents of adenosine, protein, intracellular polysaccharide and Cordyceps acid in Cordyceps militaris mycelia was obtained. Using these models for determining the contents of adenosine, protein, intracellular polysaccharide and Cordyceps acid in the samples, Rc were 0.9436、0.9884、0.9079 and 0.8848 respectively, which indicated that the fit of models was satisfied. The RMSEP of there models were 0.6225、0.0179、0.0113 and 0.0102 respectively, which indicated that the predictive abilities of these model were satisfied. The generalization of these models was fine and they can be applied to screen the high production mutants and optimize the fermentation conditions.NIR analysis technology was applied to real time monitoring the biomass, intracellular polysaccharide, extracellular polysaccharide, Cordyceps acid, adenosine and glucose during Cordyceps militaris fermentation. The samples for calibration modeling were collected from 39 batches of Cordyceps militaris fermentation at various fermentation conditions. These samples were representative. To enhance the stability of models, MCPLS was employed to examine the outliers. Several preprocessing methods including Savitzky-Golay smoothing method, FFT, SNV, first order derivative method, second order derivative method were applied to preprocess the NIR spectra. The effect of preprocessing windows’size was investigated. PLS was employed to develop the models. MWPLS was applied to select the wavelengths for developing the PLS model. The number of latent variables of PLS model was selected depending on Da. The optimum PLS models for determining biomass, intracellular polysaccharide, extracellular polysaccharide, Cordyceps acid, adenosine and glucose during Cordyceps militaris fermentation were obtained. The Rc of the models for determining biomass and glucose were 0.9114 and 0.9185 respectively and the RMSEC of them were 1.5230 and 0.0171. These results demonstrated that the fit and the predictive ability of these models were satisfied. The Rc and RMSEP of the model for determining extracellular polysaccharide concentration were 0.6875 and 0.6016, it was greatly interference with water and other factors. The Rc of these models for determining the intracellular polysaccharide, adenosine and cordyceps acid contents in the mycelia were 0.7632, 0.7252 and 0.7786, and their RMSEP were 0.3193、0.3341 and 11.4215 respectively. These results demonstrated that it was feasible to apply NIR analysis technology in real time monitoring the components in mycelia.NIR analysis technology was applied to real time monitoring nisin titer, glucose, pH and biomass during Lactococcus lactis subsp. lactis fermentation. The samples for calibration modeling were collected from 15 batches of Lactococcus lactis subsp. lactis fermentation in 3 different 5 l fermentors at various fermentation conditions. These samples were representative. To enhance the stability of models, MCPLS was employed to examine the outliers. Several preprocessing methods including Savitzky-Golay smoothing method, FFT, SNV, first order derivative method, second order derivative method were applied to preprocess the NIR spectra. The effect of preprocessing windows’size was investigated. PLS and RBFNN were employed to develop the models respectively. MWPLS was applied to select the wavelengths for developing the PLS model. The number of latent variables of PLS model was selected depending on Da. MWRBFNN was employed to select the wavelengths for developing RBFNN model. The number of hidden nodes and the spread constant of the RBFNN models were selected depending on Da. A comparing study between the optimized PLS models and RBFNN models has been done and the optimum models for monitoring nisin titer, glucose, pH and biomass during Lactococcus lactis subsp. lactis fermentation was obtained. The Rc of these models were 0.8649、0.9491、0.9390 and 0.9914 respectively, which indicated that the fit of models was satisfied. The RMSEP of there models were 2865.05、1.3076、0.2471 and 0.1414 respectively, which indicated that the predictive abilities of these model were satisfied. These methods should be popular in at line monitoring the key parameters during bacterial fermentation processes.更多还原