红外光谱与人工神经网络相结合识别栽培、野生黄芩和粘毛黄芩

为了识别栽培黄芩、野生黄芩和粘毛黄芩，采用非线性－线性、线性－线性、非线性－非线性三种模式的人工神经网络（ANN）分别分析各种黄芩的红外谱。我们采用42个样本作训练集，34个样本作检验集，用各种模式的ANN进行了监督性训练。当训练目标误差平方和定为0.01时，各类ANN对训练集中三类黄芩样本识别的正确率均为100%，但对检验集样本识别的结果各不相同，其识别的正确率与隐含层节点数S1有关。我们发现当S1较大时，识别正确率反而下降，可能此时网络的非线性程度过高，使其不适合于该类样本集的训练。线性－线性型ANN识别的结果随S1的变化不很大，但识别的正确率不高，基本在85％左右。非线性－线性型ANN识别的结果最佳。当S1为3时，其识别正确率超过了97％。因此该法可用以简便、快速、准确地识别这三种黄芩药材。     

Prediction and simulation of wear response of Linz–Donawitz (LD) slag filled glass–epoxy composites using neural computation

This article reports on the implementation of a soft computing technique based on artificial neural networks (ANNs) in analyzing the wear performance of a new class of hybrid composites filled with Linz–Donawitz slag (LDS). LDS is a major solid waste generated in huge quantities during steel making. It comes from slag formers such as burned lime/dolomite and from oxidizing of silica, iron etc. while refining the iron into steel in the LD furnace. In this work, hybrid composites consisting of short glass fiber (SGF) reinforced epoxy filled with different LDS content (0, 7.5, 15 and 22.5 wt%) are prepared by simple hand lay‐up technique. Solid particle erosion trials, as per ASTM G 76 test standards, are conducted on the composite samples following a well‐planned experimental schedule based on Taguchi design of experiments. Significant process parameters predominantly influencing the rate of erosion are identified. The study reveals that the LDS content is the most significant among various factors influencing the wear rate of these composites. Further, a model based on ANN for the prediction of erosion performance of these composites is implemented. The ANN prediction profiles for the characteristic wear properties exhibit very good agreement with the measured results demonstrating that a well‐trained network has been created. The simulated results explaining the effect of significant process variables on the wear rate indicate that the trained neural network possesses enough generalization capability of predicting wear rate even beyond the experimental range. Copyright © 2014 John Wiley & Sons, Ltd.     

Experimental measurement and modeling study for estimation of wax disappearance temperature

Wax deposition is a frequent problem in oil pipelines and down-stream industries. Correct prediction of wax formation conditions is required to prevent this phenomenon. In this study, wax appearance temperature (WAT) of 12 Iranian oil and condensate samples were measured using viscometry data and differential scanning Calorimetry (DSC) analysis. Also, a new empirical correlation and intelligent artificial neural network (ANN) model were developed to estimate wax disappearance temperature (WDT) of crude oils. Specific gravity, pressure, and molecular weight of oil sample were used as input variables for these models. The ANN model was trained using different hidden neurons and training algorithms. Experimental measurements studies were used for validation of the new correlation. Comparing the results indicated that the ANN model has 0.27% error while most thermodynamic models have an average error of 0.35% to 2.19%. Also, the proposed correlation can predict WDT with good accuracy and minimum input data. Results show that this correlation has a maximum error of 1.16% for 310 published experimental data and 1.19% for 9 Iranian samples.     

Nonintrusive data-based learning of a switched control heating system using POD,DMD and ANN

The aim of this work is to derive an accurate model of two-dimensional switched control heating system from data generated by a Finite Element solver. The nonintrusive approach should be able to capture both temperature fields, dynamics and the underlying switching control rule. To achieve this goal, the algorithm proposed in this paper will make use of three main ingredients: proper orthogonal decomposition (POD), dynamic mode decomposition (DMD) and artificial neural networks (ANN). Some numerical results will be presented and compared to the high-fidelity numerical solutions to demonstrate the capability of the method to reproduce the dynamics.     

主成分分析-支持向量机用于肝病分类模型

将主成分分析(PCA)用于肝功能检测数据特征提取,然后用支持向量机(SVM)对乙肝、丙肝、肝硬化、正常人样本建立分类模型。采用高斯径向基函数(RBF)为核函数,调节核函数参数C及σ以建立最佳支持向量机模型。该模型对训练集的识别率为99.3%,对预测集的预测率为96.4%。结果表明:PCA-SVM法建立的肝病分类模型能较好的区分乙肝、丙肝、肝硬化及正常人,且分类效果优于传统支持向量机及人工神经网络(ANN)分类模型。     

Artificial neural network for Cu quantitative determination in soil using a portable Laser Induced Breakdown Spectroscopy system

Laser Induced Breakdown Spectroscopy (LIBS) is an advanced analytical technique for elemental determination based on direct measurement of optical emission of excited species on a laser induced plasma. In the realm of elemental analysis, LIBS has great potential to accomplish direct analysis independently of physical sample state (solid, liquid or gas). Presently, LIBS has been easily employed for qualitative analysis, nevertheless, in order to perform quantitative analysis, some effort is still required since calibration represents a difficult issue. Artificial neural network (ANN) is a machine learning paradigm inspired on biological nervous systems. Recently, ANNs have been used in many applications and its classification and prediction capabilities are especially useful for spectral analysis. In this paper an ANN was used as calibration strategy for LIBS, aiming Cu determination in soil samples. Spectra of 59 samples from a heterogenic set of reference soil samples and their respective Cu concentration were used for calibration and validation. Simple linear regression (SLR) and wrapper approach were the two strategies employed to select a set of wavelengths for ANN learning. Cross validation was applied, following ANN training, for verification of prediction accuracy. The ANN showed good efficiency for Cu predictions although the features of portable instrumentation employed. The proposed method presented a limit of detection (LOD) of 2.3 mg dm^− 3 of Cu and a mean squared error (MSE) of 0.5 for the predictions.     

Simultaneous determination of iron and aluminium by differential kinetic spectrophotometric method and chemometrics

A differential kinetic spectrophotometric method was researched and developed for the simultaneous determination of iron and aluminium in food samples. It was based on the direct reaction kinetics and spectrophotometry of these two metal ions with Chrome Azurol S (CAS) in ethylenediamine-hydrochloric acid buffer (pH 6.3). The results were interpreted with the use of chemometrics. The kinetic runs and the visible spectra of the complex formation reaction were studied between 540 and 750 nm every 30 s over a total period of 285 s. A set of synthetic metal mixture samples was used to build calibrations models. These were based on the spectral and kinetic two-way data matrices, which were processed separately by the radial basis function-artificial neural network (global RBF-ANN) method. The prediction performance of these models was poorer than that from the combined kinetic-spectral three-way array, which was similarly processed by the same method (% relative prediction error (RPE_T) = 5.6). These results demonstrate that improved predictions can be obtained from the data array, which has more information, and that appropriate chemometrics methods can enhance analytical performance of simple techniques such as spectrophotometry.Other chemometrics models were then applied: N-way partial least squares (NPLS), parallel factor analysis (PARAFAC), back propagation-artificial neural network (BP-ANN), single radial basis function-artificial neural network (RBF-ANN), and principal component neural network (PC-RBF-ANN). There was no substantial difference between the methods with the overall %RPE_T range being 5.0-5.8. These two values corresponded to the NPLS and BP-ANN models, respectively. The proposed method was applied for the determination of iron and aluminium in some commercial food samples with satisfactory results.     

Estimation of Average Comonomer Content of Ethylene/1-Olefin Copolymers Using Crystallization Analysis Fractionation (Crystaf) and Artificial Neural Network (ANN)

Summary: An artificial neural network (ANN) with a 4-3-3-1 architecture was developed to estimate average comonomer content of ethylene/1-olefin copolymers from crystallization analysis fractionation (Crystaf) results. The ANN was trained with a back propagation algorithm. It was found that average comonomer contents predicted from ANN agree well with experimental results for both training and testing data sets. The developed ANN was also used to systematically investigate the effects of chain microstructures and Crystaf operating conditions on Crystaf calibration curves.     

基于人工神经网络的大量程光纤实时距离干涉测量仪

提出了一种基于人工神经网络的全光纤化大量程实时距离干涉测量仪.采用双正弦相位调制方法,即通过同时调制半导体激光器的波长和干涉仪的光程差实现外差测量.为了扩大干涉仪的测量范围和消除输出信号中的交叉敏感,采用人工神经网络进行信号处理,把两路经过初步解调的干涉信号作为输入样本,物体距离的实际值作为输出样本,对神经网络进行训练,以使其具有良好的推广能力.实验结果表明神经网络的使用不仅扩大了距离的测量范围而且提高了测量精度.