过程能力分析在制丝工艺技术改进中的应用

为稳定、提高卷烟制丝质量,采用统计过程控制技术对制丝关键工序重要参数进行了评价,发现润叶加料、混丝掺配加香工序有待改进,对其进行改进后,再次利用统计过程控制技术进一步完善制丝质量考核体系。结果表明:①经改进后的制丝生产工艺关键指标过程能力指数PPK(Preliminary Process Capability Index)值平均提高了0.23,润叶加料均匀性标准偏差降低了56.5%,梗丝掺配均匀性由改进前的94.27%提高到97.38%;②改进前后批内、批间卷烟样品的感官质量得分波动值分别降低0.29和0.27分,说明改进后批内、批间的内在质量更趋于稳定;③完善制丝质量考核体系后,松散回潮热风温度、润叶加料入口流量、加料比例、烘丝热风温度、加香瞬时比例5个指标的平均PPK值提高6.1%。     

基于偏最小二乘法的烟叶切丝后含水率的预测及应用

基于偏最小二乘法对影响切丝后含水率的12个因素进行建模分析,结果表明,各因素对切丝后含水率的影响程度不同:松散回潮出口含水率加料出口含水率贮叶柜温度车间相对湿度(切丝时)车间温度(切丝时)加料出口温度松散回潮出口温度松散回潮热风温度贮叶时间加料前HT蒸汽流量贮叶柜相对湿度加料流量;偏最小二乘法回归模型的拟合相关性较强;通过实际生产验证,通过预测切丝含水率来预判其是否满足指标范围要求的准确度较高。此方法可以为制丝生产过程切丝后含水率的预测和控制提供理论依据和生产指导。     

烟厂制丝线温湿度控制工艺研究

针对现有制丝系统生产能力,不能实现信息化生产的要求,设计了新型制丝管控系统,介绍了制丝生产线的工作流程,重点讲述了温湿度的控制方法与原理:松散回潮由滚筒式叶片回潮机完成,PLC通过PID控制器实现水分的控制;热风润叶机采用Fuzzy-PID算法完成增湿增温控制,便于烟片丝的后续处理,提高了卷烟生产效率和质量,提高了企业竞争力。     

片烟制丝工艺流程与技术指标探讨

对片烟制丝流程设计和工艺技术指标进行了分析,对目前片烟制丝存在的问题提出了改进设想：①在片烟回潮工序增加喷嘴数量;对热风加热系统进行改进;完善蒸汽和水的计量装置;②在加料工序采用热风循环系统和料液保温系统,使加料后烟叶的水分和温度控制更加简单有效;③在烟梗回潮工序采用蒸梗机等     

回潮、加料和烘丝工序烟草生物碱的变化

为探索卷烟加工关键工序中烟草总生物碱和游离生物碱的变化规律、优化工艺参数、合理控制卷烟中烟草生物碱的含量、提高产品质量提供理论依据,采用毛细管气相色谱(GC)分离、氢火焰离子化检测(FID),分析研究了制丝工艺过程中松片回潮、润叶加料和烘丝前后3个关键工序烟草总生物碱和游离生物碱含量的变化情况。结果表明,烘丝工序对烟草总生物碱和游离生物碱的影响最为显著,润叶加料工序次之,松片回潮工序影响较小。经过这3个工序后,烟草总生物碱和游离生物碱的降低幅度分别超过15%和25%。     

制丝过程对再造烟叶物理及化学性质的影响

为了更好地指导再造烟叶在卷烟配方中的运用,为再造烟叶制丝生产提供依据,比较了制丝过程中真空回潮、松散回潮、润叶及烘丝工序后再造烟叶和叶片的物理指标、化学指标的变化情况,结果表明:①再造烟叶的吸湿性、保湿性低于片烟;在润叶工序后再造烟叶已与叶片掺配均匀;②再造烟叶的卷曲程度、燃烧速率、填充值随着制丝过程的进行而增加;定量及厚度在加工过程中变化较小;抗张强度随着制丝过程的增加有降低的趋势;③再造烟叶的损耗(1.5%～2.5%)主要在切丝工序;常规化学指标在烘丝工序有小范围的变化,在其余工序基本无变化.     

3000kg/h HAUNI制丝线技改经验

对原3000kg/h HAUNI制丝生产线进行了改造.拆除了打叶机、润尖线,使全线实现了全片烟生产工艺;在润叶加料前进行三级筛分,将5～12mm碎片直接进入贮叶柜,1.5～5mm碎片与切后烟丝均匀掺兑;将打叶前的润叶基设备改造成加料机;将两个白肋烟贮柜改为双向出料,使原有的3个贮叶柜增加到5个;在切丝前增设加温加湿设备;在烟丝膨胀段增设回潮机,将KLK-G烘丝机热风由逆流式改为顺流式,烘丝后增设流动冷却床;在梗处理段一润后增设1台轴心喷汽方式的螺旋蒸梗机,在切梗前增设一道风冷装置;在松片回潮、润叶加料、叶丝回潮、梗回潮、白肋烟烘烤、加香等工序增设转子流量计和蒸气流量计,加强了流量和配比精度控制,调整了切丝水分、温度、宽度和刀门压力;调整了白肋烟烘烤机斜带、网带速度和拨辊高度,在各干燥室、冷却室、回潮室的送风口或回风口增设了整流装置.改造后制丝质量大为提高,消耗下降,稳定了产品质量.     

重点工序工艺参数与主流烟气中CO量的关系研究

对卷烟工艺参数与主流烟气中CO量的关系进行研究,结果表明:制丝过程4个重点工序中微波松散、松散回潮工序对烟支主流烟气中CO量的影响作用较小;切丝、HT+烘丝工序对烟支主流烟气中CO量的影响作用较大.切丝宽度、HT蒸汽压力、热风温度、排潮风门开度、热风风门开度和筒体转速6个工艺参数对烟支主流烟气中CO量有影响作用,其中切丝宽度和HT蒸汽压力与热风风门开度的交互作用影响显著.     

基于统计回归分析的松散回潮出口含水率精准控制系统

为解决松散回潮工序片烟出口含水率控制精度低、过程控制能力弱等问题,通过对松散回潮工序历史数据进行统计回归分析,建立了松散回潮出口含水率精准控制模型,并采用自学习算法对控制模型进行了自适应优化调整。选取南阳卷烟厂"红旗渠(天行健)"牌卷烟松散回潮的在线监测样本数据,对该控制系统的应用效果进行验证,结果表明:改进后出口含水率的控制精度显著提高,过程偏移量减少0.24%,标准偏差和极差分别减小0.078%、0.34%,过程能力指数提高0.54,有效提高了生产过程控制水平。该方法为提高制丝生产过程批次内质量稳定性提供了支持。     

提升烘丝工序批次间入口水分稳定性的研究

本文针对烘丝工序入口水分影响因素进行研究分析,以减少批次间差异性为目的。主要围绕烘丝前段的3个主要工序进行分析：贮叶工序中贮存时间和贮叶房温湿度的差异,松散回潮工序与加料工序补水能力的影响。研究其与烘丝工序批间入口水分稳定性的相关性,制定出相应的改进措施进行验证,效果提升明显。     

基于机器学习的切丝后含水率预测及控制方法

选取云烟(A)牌号制丝生产过程稳态数据样本,采用递归特征消除法分析模型的影响变量。基于车间温湿度SARIMAX预测模型,利用蒙特卡洛仿真、神经网络算法和XGBoost算法建立切丝后含水率控制模型,通过预测值与实际值对比的方法进行模型检验。结果表明,在工艺标准值±0.15%的误差范围内,切丝后含水率准确率由62.57%提升至86.49%;切丝后含水率的过程能力指数达标率由91.44%提升至97.30%。该方法实现了前后工序参数协同和精准控制,有效保证了制丝过程中切丝后含水率的稳定性。     

基于MIV-BP神经网络的成品烟丝质量预测模型构建

目的：构建卷烟制丝过程成品烟丝质量模拟预测模型。方法：使用平均影响值法（the Mean Impact Value, MIV）对制丝加工过程工艺参数进行筛选,然后通过反向传播（Back-Propagation,BP）神经系统构建起制丝关键工艺参数和成品烟丝质量的模拟模型。结果：通过模拟数据与实测数据比较,填充值的模拟预测平均相对误差为3.16%;整丝率的模拟预测平均相对误差为0.67%;碎丝率的模拟预测平均相对误差为5.33%。结论：该模型预测值与实测值之间相对误差较小,精确性高,该模型适用于卷烟制丝生产过程工艺参数仿真优化。     

基于多因素分析的烘丝机入口含水率预测模型的建立与应用

为了保障制丝过程中烘丝机入口含水率的稳定性,采用Pearson相关分析的方法,确定烘丝机入口含水率的主要影响因素,并用神经网络算法和多元回归分析方法建立含水率预测模型。通过模型求解,实现给定烘丝机入口含水率计算松散回潮机回潮加水比例参考值的目的。采用模型预测值与实测值对比的方法进行检验。结果表明:烘丝机入口含水率设定值为19.2%时,采用本方法得到的烘丝机入口含水率均值为19.21%,优于改进前的19.09%,且误差标准偏差由0.43%降到0.26%,批次间烘丝机入口含水率的波动得到改善。      

基于LSTM网络的粮食干燥机水分预测与优化

粮食干燥机的出机粮食水分预测有助于实现干燥机的智能化控制,从而可以减少干燥过程中的粮食损耗,对于粮食产后干燥环节有着重大意义。通过机器学习的方式进行预测,可以规避传统数学模型所存在的一系列缺陷。本文根据连续式谷物干燥机所提取的数据特征,提出了一种基于优化长短期记忆神经网络（LSTM）的稻谷出机水分预测模型。试验结果表明,出机水分与Min、To2、To3、Td1、Td2、Td3具有十分明显的相关性,通过设定不同的网络参数,确立了批尺寸50,学习率0.001,迭代次数50,时间步长50,神经元数100*100时效果最佳,此外还发现增加训练数据量,可以有效提高LSTM网络预测性能。将本研究建立的LSTM模型与BP、ELMAN、NARX等算法以及普通LSTM网络（无dropout,单隐藏层）进行比较。结果发现,相较于其他网络模型,本文所采用的LSTM模型可以更好的预测稻谷出机水分,其平均绝对误差（MAE）、均方根误差（RMSE）和决定系数（R2）分别为0.12%、0.20%和0.94。本研究所采用的优化LSTM模型具有较高预测精度,稳定性以及泛化性,可以为粮食干燥机的水分预测控制提供参考。     

紫外光谱结合化学计量学检测初榨橄榄油掺伪研究

以紫外光谱为技术手段,结合偏最小二乘法和BP人工神经网络2种化学计量学方法建立了初榨橄榄油/混合橄榄油二元掺伪体系的定量预测模型.试验结果表明,2种统计模型定量预测性能良好,偏最小二乘模型的训练集交叉验证均方根误差RMSEcv和预测集均方根误差RMSEP均达到0.011,预测值与真实值相关性达到0.996 2;BP人工神经网络迭代次数为61步,训练集拟合残差为9.684×10-5,网络预测值和真实值相关系数为0.998 3,对于5％以上掺伪比例的油样BP神经网络能够精确地预测.     

Modeling of thermal properties of multilayered fabrics by ANN consisting of polypropylene needle-punched nonwovens

This paper presents the prediction of thermal and evaporative resistances of multilayered fabrics meant for cold weather conditions using artificial neural network (ANN) model. Thermal and evaporative resistances of fabrics were evaluated using sweating guarded hot plate method. The significance and interdependency of thickness on other fabric and process parameters and its effect on prediction performance of ANN model is analyzed in detail. For this purpose, two different network architectures were used to predict the thermal properties of multilayered fabrics. In both the networks, three-layer structure consisting of input, hidden and output layers was used. First, network was constructed with four input parameters, namely linear density of fiber, mass per unit area, punch density, and thickness of nonwoven fabric which predicts thermal and evaporative resistances. Second network was made with three input parameters, namely linear density, mass per unit area, and punch density. The network parameters were optimized to give minimum mean square error (MSE), mean absolute error percentage, and good correlation coefficient. The trend analysis was conducted and influence of various input parameters on the thermal properties of multilayered fabrics was studied. The significance of each input parameter in the prediction of thermal properties was studied by carrying out sensitivity analysis. The mean square error of the test dataset before and after the exclusion of the corresponding input parameter is taken for analysis. The input parameters were ranked based on the MSE ratio of test dataset. The predicted thermal properties of multilayered fabrics are correlated well with the experimental values. It was observed that the ANN model with minimum input parameters, namely linear density of fiber, mass per unit area, and punch density can predict the thermal properties of multilayered fabrics with good accuracy.     

人工神经网络和响应面法优化薏苡仁酒发酵条件

采用Box-Behnken试验设计对薏苡仁酒的发酵条件进行优化，并对Box-Behnken（BB）试验结果分别进行响应面法（RSM）和人工神经网络（ANN）分析。结果表明，RSM、ANN优化发酵条件分别为薏苡仁∶糯米为1∶2（g∶g）、酵母A1接种量为4.7%、温度为31.7 ℃、初始pH为3.0；薏苡仁∶糯米为1∶1.9（g∶g）、酵母A1接种量为4.2%、温度为28.1 ℃、初始pH为3.0，ANN、RSM分别在其最优条件下的实际值和预测值都基本一致。ANN、RSM拟合模型的相关系数（R）、决定系数（R2）、均方误差（MSE）、均方根误差（RMSE）、平均绝对误差（MAE）、平均绝对百分比误差（MAPE）分别为0.994 5、0.988 9、0.011 7、0.108 4、0.072 2、0.486 3%和0.983 6、0.967 5、0.028 9、0.170 1、0.143 7、0.985 7%。ANN具有更高拟合能力和准确性，拟合效果更好，更适合应用于薏苡仁酒发酵条件优化。     

Employing artificial neural networks for minimizing surface roughness and power consumption in abrasive machining of wood

One of the biggest challenges in machining processes of wood is to detect the optimum values of process parameters for reducing the final production cost. In the present study, the effects of various process parameters on surface roughness and power consumption in abrasive machining process of wood using experimental data collected from the literature were modeled by artificial neural networks (ANNs). The results have indicated that accurate prediction of the experimental data by neural network models was achieved with the mean absolute percentage error (MAPE) less than 2.51 % for power consumption and 2.65 % for surface roughness in the testing phase. Besides, the values of determination coefficient (R²) were found as 0.994 and 0.985 in the prediction of surface roughness and power consumption by the ANN modeling, respectively. Based on the results, it can be said that by means of the proposed models the surface roughness and power consumption can easily be predicted with very high degrees of accuracy in abrasive machining process of wood. Consequently, the present study can effectively be applied to the wood industry to reduce the time, energy consumption and high experimental costs because it eliminates the need for a large number of experiments.     

Drying kinetics prediction and quality effect of ultrasonic synergy vacuum far-infrared drying of Codonopsis pilosula

By using ultrasonic synergy vacuum far-infrared drying (US-VFID), the effects of different conditions on the drying kinetics, functional properties, and microstructure of Codonopsis pilosula slices were studied. The sparrow search algorithm (SSA) was used to optimize the back-propagation (BP) neural network to predict the moisture ratio during drying. With the increase of ultrasonic frequency, power and radiation temperature, the drying time of C. pilosula was shortened. The drying time of US-VFID was 25% shorter than VFID, when radiation temperature was 50°C, ultrasonic power was 48 W, and frequency was 28 kHz. The SSA-BP neural network, the average absolute error prediction was 0.0067. Compared with hot air drying (HAD), the total phenolic content and antioxidant activity of C. pilosula by US-VFID were increased by 29.47% and 8.67%, respectively, and a reduction in color contrast of 16.19%. The dilation and generation of microcapillary of C. pilosula were more obvious. The study revealed US-VFID could be used for the selection and process control of agro-processing methods for C. pilosula products.     

Genetic Algorithm�CArtificial Neural Network Modeling of Moisture and Oil Content of Pretreated Fried Mushroom

In this research, the effect of different pretreatments (osmotic dehydration and gum coating) on moisture and oil content of fried mushroom was investigated, and artificial neural network and genetic algorithm were applied for modeling of these parameters during frying. Osmotic dehydration was performed in solution of NaCl with concentrations of 5% and 10%, and methyl cellulose was used for gum coating. Either pretreated or control samples were fried at 150, 170, and 190 °C for 0.5, 1, 2, 3, and 4 min. The results showed that osmotic dehydration and gum coating significantly decreased (0–84%, depending upon the processing conditions) oil content of fried mushrooms. However, moisture content of fried samples diminished as result of osmotic pretreatment and increased by gum coating. An artificial neural network was developed to estimate moisture and oil content of fried mushroom, and genetic algorithm was used to optimize network configuration and learning parameters. The developed genetic algorithm–artificial neural network (GA–ANN) which included 17 hidden neurons could predict moisture and oil content with correlation coefficient of 0.93 and 96%, respectively. These results indicating that GA–ANN model provide an accurate prediction method for moisture and oil content of fried mushroom.