In silico optimization of actuation performance in dielectric elastomercomposites via integrated finite element modeling and deep learning

Dielectric elastomers(DEs)require balanced electric actuation performance and mechanical integrity under applied voltages.Incorporating high dielectric particles as fillers provides extensive design space to optimize concentration,morphology,and distribution for improved actuation performance and material modulus.This study presents an integrated framework combining finite element modeling(FEM)and deep learning to optimize the microstructure of DE composites.FEM first calculates actuation performance and the effective modulus across varied filler combinations,with these data used to train a convolutional neural network(CNN).Integrating the CNN into a multi-objective genetic algorithm generates designs with enhanced actuation performance and material modulus compared to the conventional optimization approach based on FEM approach within the same time.This framework harnesses artificial intelligence to navigate vast design possibilities,enabling optimized microstructures for high-performance DE composites.

Neural Network Approach to Modelling the Behaviour of Ionic Polymer-Metal Composites in Dry Environments

Ionic polymer-metal composites (IPMCs) are especially interesting electroactive polymers because they show large a deformation in the presence of a very low driving voltage (around 1 - 2 V) and several applications have recently been proposed. Normally a humid environment is required for the best operation, although some IPMCs can operate in a dry environment, after proper encapsulation or if a solid electrolyte is used in the manufacturing process. However, such solutions usually lead to increasing mechanical stiffness and to a reduction of actuation capabilities. In this study we focus on the behaviour of non-encapsulated IPMCs as actuators in dry environments, in order to obtain relevant information for design tasks linked to the development of active devices based on this kind of smart material. The non-linear response obtained in the characterisation tests is especially well-suited to modelling these actuators with the help of artificial neural networks (ANNs). Once trained with the help of characterisation data, such neural networks prove to be a precise simulation tool for describing IPMC response in dry environments.

A novel method for extracting and optimizing the complex permittivity of paper-based composites based on an artificial neural network model

Measuring the complex permittivity of ultrathin,flexible materials with a high loss tangent poses a substantial challenge with precision using conventional methods,and verifying the accuracy of test results remains difficult.In this study,we introduce a methodology based on a back-propagation artificial neural network(ANN)to extract the complex permittivity of paper-based composites(PBCs).PBCs are ultrathin and flexible materials exhibiting considerable complex permittivity and dielectric loss tangent.Given the absence of mature measurement methods for PBCs and a lack of sufficient data for ANN training,a mapping relationship is initially established between the complex permittivity of honeycomb-structured microwave-absorbing materials(HMAMs,composed of PBCs)and that of PBCs using simulated data.Leveraging the ANN model,the complex permittivity of PBCs can be extracted from that of HMAMs obtained using standard measurement.Subsequently,two published methods are cited to illustrate the accuracy and advancement of the results obtained using the proposed approach.Additionally,specific error analysis is conducted,attributing discrepancies to the conductivity of PBCs,the homogenization of HMAMs,and differences between the simulation model and actual objects.Finally,the proposed method is applied to optimize the cell length parameters of HMAMs for enhanced absorption performance.The conclusion discusses further improvements and areas for extended research.

Modelling and Characterization of Basalt/Vinyl Ester/SiC Micro-and Nano-hybrid Biocomposites Properties Using Novel ANN–GA Approach

Basalt fiber reinforcement in polymer matrix composites is becoming more and more popular because of its environmental friendliness and mechanical qualities that are comparable to those of synthetic fibers.Basalt fiber strengthened vinyl ester matrix polymeric composite with filler addition of nano-and micro-sized silicon carbide(SiC)element spanning from 2 weight percent to 10 weight percent was studied for its mechanical and wear properties.The application of Artificial Neural Network(ANN)to correlate the filler addition composition for optimum mechanical properties is required due to the non-linear mechanical and tribological features of composites.The stuffing blend and composition of the composite are optimized using the hybrid model and Genetic Algorithm(GA)to maximize the mechanical and wear-resistant properties.The predicted and tested ANN–GA optimal values obtained for the composite combination had a tensile,flexural,impact resilience,hardness and wear properties of 202.93 MPa,501.67 MPa,3.460 J/s,43 HV and 0.196 g,respectively,for its optimum combination of filler and reinforcement.It can be noted that the nano-sized SiC filler particle enhances most of the properties of the composite which diversifies its applications.The predicted mechanical and wear values of the developed ANN–GA model were in closer agreement with the experimental values which validate the model.

基于复合神经网络的多源气动数据建模

将深度学习方法应用至气动数据建模,能够解决传统建模方法效率低、代价高的问题,具有重要的现实意义。基于复合神经网络模型对多源气动数据进行学习,利用低精度数据辅助高精度数据进行预测。与不同网络模型进行对比,验证了文中提出的复合神经网络在气动数据建模中表现优良,且泛化能力较好。

基于RBF-BP算法的Ni-SiC镀层性能预测研究

采用磁场辅助喷射电沉积技术,在不同工艺条件下制备Ni-SiC复合镀层,通过构建4×4×4×7×10×1的RBF-BP复合神经网络模型预测Ni-SiC复合镀层耐蚀性。结果表明:RBF-BP复合神经网络的预测值与真实值拟合度为0.974 97,高于单神经网络,表明复合神经网络能准确预测不同工艺参数下制备的Ni-SiC复合镀层耐蚀性。经复合神经网络预测,当电流密度为4 A/dm^(2)、喷射速度为6 m/s、SiC粒子浓度为8 g/L、磁场强度为0.8 T时复合镀层腐蚀失重最低,复合镀层的耐蚀性最好。通过镀层表征研究分析可知,该条件下镀层晶粒显著细化,镀层表面较平滑,SiC纳米粒子复合量高且分布均匀。

基于BP神经网络的雪茄原料感官质量预测模型构建

本研究旨在利用BP神经网络技术,深入分析并预测雪茄原料的常规化学成分与其感官质量之间的复杂关系。通过收集四川、湖北、云南、湖南和尼加拉瓜雪茄烟叶常规化学成分数据作为输入变量,结合雪茄原料各项感官质量指标作为输出变量,成功构建了拓扑结构为6-9-1的BP神经网络模型。该模型不仅能够准确预测雪茄原料的感官质量评吸结果,而且揭示了不同产区雪茄烟叶在化学成分和感官质量方面的独特特征。研究表明,所检测样本中,国内4个主产区雪茄烟叶总糖、还原糖、烟碱、氯含量均高于尼加拉瓜烟叶,尼加拉瓜烟叶香气质和香气量得分较高。四川烟叶刺激性得分较低,湖北产区雪茄烟叶余味得分较高,云南烟叶杂气得分较低,湖南烟叶燃烧性和灰色得分较高。本研究雪茄烟叶样本的常规化学成分和感官质量指标统计特征较好,基本服从正态分布。所构建的BP神经网络模型的预测值与实际值间差异较小,其中余味、刺激性、灰色和总分的相关系数均在0.9以上。在训练集、验证集和测试集的预测值和实际值误差中,除总分误差区间较大外,剩余多数指标误差区间在0~0.5范围内的比例达到85%以上。BP神经网络所建立的雪茄原料感官质量预测模型拟合效果较好。本研究的成功实施为基于常规化学成分快速、准确地预测雪茄原料感官质量提供了有力支持,有助于推动中式雪茄烟行业的创新发展。

Adaptive composite anti-disturbance control for heavy haul trains

In this paper,an adaptive composite anti-disturbance control of heavy haul trains(HHTs)is proposed.First,the mechanical principle and characteristics of couplers are analysed and the longitudinal multi-particles nonlinear dynamic model of HHTs is established,which can satisfy that the forces of vehicles in different positions are different.Subsequently,a radial basis function network(RBFNN)is employed to approximate the uncertainties of HHTs,and a nonlinear disturbance observer(NDO)is constructed to estimate the approximation error and external disturbances.To indicate and improve the approximation accuracy,a serial-parallel identification model of HHTs is constructed to generate a prediction error,and an adaptive composite anti-disturbance control scheme is developed,where the prediction error and tracking error are employed to update RBFNN weights and an auxiliary variable of NDO.Finally,the feasibility and effectiveness of the proposed control scheme are demonstrated through the Lyapunov theory and simulation experiments.

基于光纤传感技术的多股碳纤维复合芯导线隐蔽性缺陷检测

为解决大容量、大跨越输电用碳纤维导线由于隐蔽性缺陷无法检出而导致频繁断线的问题,提出了一种基于光纤传感技术的多股碳纤维复合芯导线隐蔽性缺陷检测方法。该方法通过搭建导线运行环境并模拟导线运行工况,采用基于分布式光纤布里渊散射的时域反射技术,检测碳纤维导线的温度和应变分布情况,并结合光时域反射技术,检测碳纤维导线中光纤的损耗情况。经过综合对比分析,获取可表征多股碳纤维缺陷隐蔽性缺陷的光纤温度、应变、损耗等信号特征量,并构建神经网络模型,将各信号特征量作为模型输入,通过模型训练确定模型内各权系数,使其能够有效地检测多股碳纤维复合芯导线隐蔽性缺陷。实验结果表明,该方法可有效获取各类光纤信号特征量,并且能够准确地检测各类导线隐蔽性缺陷,具有重要的实际应用价值。

自适应多保真数据融合的神经网络模型

数据驱动的深度学习建模在力学、材料等不同学科中得到了较多应用。深度学习建模的精度依赖大量高保真数据。在实际应用中,高保真数据往往是少量且昂贵的,而低保真数据却是成本低廉且数量较多的。当高保真数据量过少时,深度学习建模精度较低。近期发展的多保真深度神经网络,通过融合不同保真度的数据,在高保真数据较少时,依然保持了较高的建模精度。然而,已有的多保真深度神经网络模型的精度较为依赖针对模型参数的正则化调节。当添加的正则化过强时,网络对非线性关联式的拟合能力不足;当添加的正则化强度不够时,在学习多保真数据间的线性关联关系时又会出现过拟合现象。两者都会严重影响模型的预测精度。在缺乏高保真验证数据集时,较难得到最优的正则化系数。为此,通过改进已有多保真网络模型的损失函数,引入一个与线性关联式相关的参数,提出了自适应多保真数据融合的神经网络模型。该模型能根据给定数据自适应地拟合不同保真度数据间的线性或非线性关系,对正则化依赖较小,从而提高了建模的鲁棒性。在多个标准测试案例及实际应用的翼型气动参数的预测中,该模型均能表现出较高的精度和稳定性。

基于RBF-BP复合神经网络的企业创新能力评价模型

为更加精准地测量企业创新能力,利用熵值法和相关性分析,从创新投入、创新产出、创新支撑3个维度建立企业创新能力评价指标体系,构建企业创新能力评价的RBF-BP复合神经网络模型。该模型由1个输入层、1个RBF隐含层、1个BP隐含层以及1个输出层组成,其特点是将RBF隐含层的输出作为BP隐含层的输入。十折交叉验证与随机二次抽样2种方法检验表明,与单一RBF神经网络、单一BP神经网络相比,RBF-BP复合神经网络模型的平均均方误差与平均绝对误差分别下降28.21%、15.19%和12.51%、12.55%,表明RBF-BP复合神经网络模型具有最优的数据拟合能力,更适合于企业创新能力评价。

数据—机理融合驱动的复合材料疲劳损伤渐进分析方法

随着纤维增强复合材料在航空航天领域的广泛应用,复合材料的疲劳问题日益突现。为了实现高效准确的疲劳损伤分析,提出一种数据—机理融合驱动的复合材料疲劳损伤渐进分析方法,该方法利用单隐藏层神经网络作为内聚力模型的疲劳本构进行循环载荷下的疲劳分层模拟。采用基于Paris公式的损失函数约束项实现数据—机理融合的神经网络模型训练;在采用双悬臂梁和四点弯梁验证在纯Ⅰ型、纯Ⅱ型和混合型模式下的疲劳分层扩展分析能力后,进一步采用增强双悬臂梁(R-DCB)模型对具有复杂疲劳分层前缘情况下内聚力模型的适用性进行验证。结果表明:数据—机理融合驱动的复合材料疲劳损伤渐进分析方法能够实现快速有效的复合材料分层扩展预测,为复合材料结构设计和安全保障提供了新的思路和方法。

基于BP神经网络的股票价格预测

影响股票价格变动的因素有很多,且股票数据具有高度的非线性和时变性等特征,因而采用经典线性时间序列模型可能无法完全提取非线性部分的信息.针对这一问题,建立了BP神经网络模型、PCA-BP神经网络模型、GA-BP神经网络模型和ARIMA(6,1,6)模型对上证综合指数的收盘价格进行预测.计算各预测模型下的统计指标RMSE和MAE,并对4个模型进行对比分析.结果表明,GA-BP神经网络预测模型与其它三种模型相比具有更小的误差,也就是说GA-BP神经网络预测模型对上证综合指数的收盘价格预测效果更好.

某大型固体发动机T800碳纤维壳体封头结构仿真分析和优化设计

针对15 MPa内压载荷下某碳纤维复合材料壳体结构优化设计问题,采用代理模型辅助优化算法得到最大沿纤维方向的应变最小的后封头补强模型。基于Python对ABAQUS进行二次开发构建精细化的碳纤维壳体后封头模型并进行仿真分析,预测其破坏位置并得到优化目标参数。以碳纤维缠绕层带宽、纵向缠绕层退移距离和补强位置为优化变量,最大沿纤维方向应变为优化目标,利用LHS(拉丁超立方采样)方法建立关于优化变量的参数化组合样本,结合神经网络代理模型和遗传算法对国产T800碳纤维缠绕壳体后封头最大沿纤维方向应变进行寻优求解。结果表明,在纵向缠绕层退移半径为19.98 mm、带宽为25.96 mm、补强层起始半径为579.38 mm时,最大沿纤维方向应变可以达到最小的0.0158,相比未优化设计结构下降了4.2%,为今后的大型国产碳纤维缠绕壳体封头设计提供了设计方法参考。

复合地层地质参数与水下大直径盾构掘进参数相关性预测研究

为探索复合地层地质参数与水下大直径盾构掘进参数的内在联系,以汕头海湾隧道工程为背景,采用BP神经网络方法构建相关性预测模型,实现从地质参数到掘进参数的量化预测。研究结果表明:(1)通过建立基于掘进环数的地质参数及盾构稳定掘进时的掘进参数数据库,可有效表征地质及设备特性;(2)通过数据预处理、优选网络结构及训练函数,构建基于BP神经网络的相关性预测模型,总推进力、刀盘转速、掘进速度和刀盘扭矩平均误差小于6%,最大误差小于18%,预测效果较好,基本达到工程要求。基于BP神经网络相关性预测模型,输入地质参数即可定量预测掘进参数,结构简单、运算方便、精度较高。

基于组合神经网络的软件命名实体识别仿真

当前软件命名实体识别方法忽略了对命名实体标签的预测,存在精度等级(P@N)、F1值和KS值均偏低问题。提出基于组合神经网络的软件命名实体识别方法。将识别命名实体问题转化成“SBEIO”标签预测问题,在组合神经网络模型的基础上提取字、词特征,并将两者结合得到词向量特征,以此预测出最优标签序列。构建支持向量机分类器,将标签序列进行分类,根据超平面分割出分类结果,利用决策函数确定出最优分类命名实体,实现软件命名实体识别。实验结果表明,所提方法具有较高的精度等级(P@N)、F1值以及KS值。

An improved neural network model for prediction of mechanical properties of magnesium alloys

An improved neural network model was developed for prediction of mechanical properties in the de-sign and development of new types of magnesium alloys by refining the types of input variables and using a more reasonable algorithm. The results showed that the improved model apparently decreased the prediction errors, and raised the accuracy of the prediction results. Better preprocessing parame-ters were found to be [0.15, 0.90] for the tensile strength, [0.1, 0.9] for the yield strength, and [0.15, 0.90] for the elongation. When the above parameters were used, the relativity for predicition of strength was bigger than 0.95. By using improved ANN analysis, more reasonable process parameters and compo- sition could be obtained in some magnesium alloys without addition of strontoum.

基于人工神经元网络的PA6/POE/POE-g-MAH共混物力学性能预测

将人工神经元网络与全析因实验设计相结合,预测了PA6/POE/POE-g-MAH三元共混物的拉伸和缺口冲击强度,研究了注塑温度、注射速度、注射时间和冷却时间对共混物以上力学性能的影响。结果表明,当三元共混物质量比为74.5/20/5/0.5时,材料的缺口冲击强度均大于95 kJ/m^(2),POE分散相粒径约为200~400 nm。通过析因实验设计发现,注塑温度、注射速度以及两者的交互效应对共混物拉伸强度影响显著,注塑温度、冷却时间以及两者的交互效应是影响冲击强度的显著因素。采用4-14-1的ANN模型可以较好地预测三元共混物拉伸强度和缺口冲击强度,预测结果与实验结果回归曲线的斜率均大于97%。与多元线性回归模型相比,ANN模型预测性能明显更佳,其Pearson相关系数大于0.97。

沥青材料微观组成与宏观性质关联性研究

为了研究沥青微观组成与宏观性质的关联性,选取10种沥青对其微观组成和宏观性质进行分析,通过红外谱图分析不同沥青官能团结构的区别,利用灰色关联法对影响60℃动力黏度、针入度指数、残留针入度比的微观性质进行关联度排序,并建立BP神经网络模型进行性质预测。结果表明:红外谱图分析不能实现对不同种类沥青的精确分辨;芳香分质量分数对60℃动力黏度和针入度指数的影响程度最大,而胶体不稳定指数是影响残留针入度比的最重要因素;BP神经网络模型的预测值与真实值的拟合效果较好,最小相对误差可低至-0.61%,最大相对误差不高于10%。

基于机器学习的土壤锰污染程度预测模型构建

为解决传统土壤锰污染程度预测模型预测精度不足的问题,研究在融合小波神经网络与协同鸟群算法的基础上构建了深度复合神经网络的土壤锰污染程度预测模型。对研究提出的融合算法进行性能测试,结果显示研究提出的融合算法在误差小于10%及10%~20%的占比为75%~76%,该算法误差性能优于其他算法。对基于深度复合神经网络预测模型进行性能对比实验,结果显示其在四川和重庆2个数据集上的运算时间分别为26.6 s和24.5 s,较其他模型运算时间短,且其收敛速度更快。综合以上结果可以发现,研究提出融合算法及深度复合神经网络土壤锰污染程度预测模型在运算速度、运算精确度上优于对比算法与模型,具有实际应用价值。