【作者】 李波；

【导师】 张新申；

【作者基本信息】 四川大学 ， 皮革化学与工程， 2005， 博士

【摘要】 本文主要围绕两方面进行研究:第一方面为低压离子色谱单柱法测定阴离子,第二方面为利用人工神经网络建立离子色谱保留模型。论文分为两大部分,从不同的角度对离子色谱进行研究。 一、有关低压离子色谱单柱法测定阴离子的研究 自1979年Fritz提出单柱离子色谱法以来,已取得了极大的成功,能够实现无机阴离子、有机阴离子、有机酸和无机阴阳离子的同时测定,应用领域也非常广。但一直采用高压泵输送流动相,使得仪器的价格居高不下。 低压离子色谱法,因在低压下工作而极大地降低了费用,减少了泄漏的可能性和工作噪声。低压离子色谱的应用领域已深入到环境保护、皮革工业、石油化工、医疗保健和农业等行业。低压离子色谱在测定无机阴离子时大多采用双柱法,虽然能够满足各种样品的测定需要,但在实际使用中需要经常再生抑制柱。 通过对低压离子色谱单柱法测定阴离子进行深入的研究,取得了一定的进展,主要研究内容如下: 1.洗脱体系的选择对降低本底电导具有非常重要的意义,好的缓冲体系可以起到明显降低本底电导的作用,实验表明选择有机碱调节洗脱液pH值的效果更多还原

【Abstract】 This paper is divided into two parts to study on ion chromatography from two sides. The first part is determination of anion by single-column LPIC. The second part is modeling of ion chromatography retention usingANN.Ⅰ、 Study on determination of anion by single-column LPICSingle-column ion chromatography has been widely used for determination of inorganic anion, organic anion and simultaneous determination of organic acid, inorganic anion and cation successfully since 1979. For the using of high-pressure pump, the price of the equipment is high.The introduction of low-pressure working greatly reduces the cost and the possibility for leakage in high-pressure operation also removing the noise of the high-pressure pump thus improving the operator’s working condition. LPIC is applied in many fields such as environmental protection, leather, earth oil chemical engineering, medicaments and agriculture, etc. In determining of inorganic anion by LPIC, two-columnion chromatography with conductance detector are used. The first column serves to separate the ion of interest while the second one, the suppressor, serves to lower the conductance of the eluent. The suppressor column must be periodically regenerated.By studying on the determination of anion using single-column LPIC systematically, some evolvements have been made. The main contents are as follows:1. The Selection of eluting system is very important for reducing the background conductance of the eluent, and a good buffer system can reduce the background conductance distinctly. Experiments showed that the effect of adjusting the eluent pH with organic alkali is better than that with inorganic alkali.2. The exchange capacity of separation column is very important for improving the separation resolution of weakly retained ions. The retention time of injection peak has relationship with the length of the column. When the eluting flow rate is constant, the longer of the length of the column, and the longer of the retention time of the injection peak. The retention time of the injection peak has no relationship with the exchange capacity of the column. When other conditions are constant, increasing the exchange capacity of the column will increase the retention times of ions peak but the retention time of injection peak will be constant. The separation resolution between the early eluting ion peak and the injection peak can be improved.3. The analytical conditions of determination of Cl" , N03~and SO/" by single-column LPIC of conductance detection were selected through experiments. The exchange capacity of the separation column ( cj> 5 X 50mm)is 0. 08mmol/g. The eluent is 0. 15mmol/l phthalic acid whose pH is 5.26. The eluting flow rate is 1.3ml/min. Under these conditions, the calibration curves were obtained over the concentration range of 2-4. 5mg/L for Cl", 2-80 mg/L for NO.," and 4-100 mg/L for SO/". The method has high precision and can be applied to the determination of the wastewater of leather manufacture and ground water with satisfactoryresults.4. The analytical conditions of determination of Cl~ , N(V and S042" by single-column LPIC of UV-photometric detection were selected through experiments. The exchange capacity of the separation column (<i> 5 X 85mm) is 0. 08mmol/g. The eluent is 0. 3mmol/l potassium acid phthalate whose pH is 5. 0. The eluting flow rate is 1. 3ml/min. Under these conditions, the calibration curves were obtained over the concentration range of 10-100 mg/L for Cl", 10-100 mg/L for N(V and 50-500 mg/L for SO/’. The method has high precision and can be applied to the determination of the wastewater of leather manufacture and ground water with satisfactory results.5. The theory analysis and experiments showed that the direction of ion peak is positive because of the increase of conductance in single-column ion chromatography with the method of direct conductance detection. The direction of the injection peak and the ions peak is the same as the total concentration of the determination ion is usual larger than the concentration of the eluent. The interference of the injection peak is large. The direction of the ions peak is negative because of the decrease of absorbency in single-column ion chromatography with indirect UV-photometric detection. The direction of the injection peak and the ions peak is reverse as the total concentration of the determination ion is usual larger than the concentration of the eluent. The interference determination by injection peak is small.6. The sensitivity obtained with different detection methods relates to the solute types and the chromatography conditions. The sensitivity obtained with conductance detection relates to the conductance difference between the eluent and the detected ion. The sensitivity obtained with indirect UV-photometric detection relates to the molar absorbency difference between the eluent and the detected ion. There is no fixed conclusion of the sensitivity of which method is higher. Experiments should be done for making the conclusion for different eluting system and detected ion.7. It is known from the literature analysis and experiments that the baseline stability of indirect UV-photometric detection is superior to that of the conductance detection.IK Study on modeling of ion chromatography retention usingANNOptimization of mobile phase composition is an important aspect of method development in ion chromatography. Retention mapping method describes the chromatographic behaviour of solutes in the design space by its response surface, which shows the relationship between the response, i.e., the rentention time of solute, and several input variables, i.e., the components of the mobile phase.There are many factors that affect the retention behavior of ion. Retention model belongs to nonlinear multi-variable complex system. Though many scholars have studied the model, the mechanism of the complex system cannot be predominated completely. Neural network is a software model; it is able to accommodate all types of linear and nonlinear relationships by learning the relations from the data themselves, and its computing speed is very fast. The whole study process needs no knowledge of mechanism, so it has huge advantage in dealing with multi-variable nonlinear system and unknown model system. Neural network takes all solute data as a matrix to establish a model, so it can make good use of the information of the whole matrix.Though the study on neural network is systematical and sufficient and train algorithms are also mature, neural network is not a general-duty formula that can be used easily. Rational network architecture and feasible train algorithm should be selected through deep study on idiographic situation.Modeling of retention of single column ion chromatography using artificial neural network is very important. Modeling of retention establishes a firm foundation for optimizing experiment conditions, thusthe retention time of each solute for each mobile phase composition can be predicted and the optimization of experiment conditions can be guaranteed.Systematical study on modeling retention of ion chromatography with ANN has been done in this paper. Takeing the usually used BP network theoretics as a foundation and combined it with new fruit, the suitable network architecture, transfer function and train algorithm was selected by systematically study on the application of ANN in ion chromatography. Some evolvements have been made. The main contents are as follows:1.the retention model of the ion chromatography was established by using the literature data. A three-layer backpropagation neural network type with one hidden layer was used. The input layer consists of two neurons representing the eluenting flow rate and the concentration respectively, the hidden layer consists of five neurons, the output layer consists of seven neurons representing the retention time of F~, Cl", N02~, S042~, Br", N03" and PO43" respectively. The input values were scaled between -1 and 1; the output values were transformed logarithmically. The log-sigmoid transfer function was used for computation the activities of hidden layer nodes, the linear transfer function was employed for the computation activities of output, and the Bayes regularization train algorithm was applied. The average relative error of the training set is 1.84%. The average relative error of the testing set is 1.97%. The average relative error of the testing set is a little higher than that of the training set. The generalization capability of the network is excellent.2. We can know from the theoretics analysis that the statistical properties of each subset of the data divided with genetic algorithm are more similar than that of the data divided by random. It was proved by the case study that it is possible for dividing the ion chromatography data using genetic algorithm.3. Uniform design table was used to design the original weight of the neural network. It was proved by the case study that it is an easyand good method using uniform design table to design the original weight for neural network whose performance depends on the original weight. 4. Retention model of LPIC was established using ANN. A three-layer backpropagation neural network type with one hidden layer was used. The input layer consists of two neurons representing the concentration of phthalic acid and tris respectively, the hidden layer consists of four neurons, and the output layer consists of two neurons representing the retention time of N03" and SO/"respectively. The input values were scaled between -1 and 1; the output values were transformed logarithmically. The hyperbolic tangent transfer function was used for computation the activities of the hidden layer nodes, the linear transfer function was employed for computation the activities of the output, and the Bayes regularization train algorithm was applied. The average relative error of the training set is 0.82%. The average relative error of the testing set is 2. 54%. The average relative error of the testing set is higher than that of the training set. The generalization capability of the network is excellent.更多还原