BPNN用于含氮、硫杂环润滑油添加剂抗磨性能的定量构效关系研究

采用反向传播神经网络法(Back Propagation Neural Network,简称:BPNN)对31种含氮、硫的2-烷基黄原酸酯类润滑油添加剂的抗磨性能进行了摩擦学定量构效关系(Quantitative Structure Tribo-ability Relationship,简称:QSTR)的研究,得到了具有良好的稳定性和预测能力的BPNN-QSTR模型(R~2=0.998 4,R~2(LOO)=0.695 9,q~2=0.879 1).参考输入层的12种2D和3D结构描述符的敏感度,对影响抗磨性能的分子结构进行了相应的探讨.结果表明:分子中的N和S杂原子对其抗磨损性能有显著的影响;同时,分子长度、所含双键S原子和芳香环数量以及分子支化程度等都是影响抗磨性能的主要因素.     

基于更新径向基函数网络模型的广义Pareto分布函数拟合

广义Pareto分布函数GPD(Generalized Pareto Distribution)是一种针对随机参数尾部进行渐进插值的方法,能够对高可靠性问题进行评估。但这种方法要求样本空间较大,计算成本较高,尽管可以通过径向基函数网络RBFNN(Radial Basis Function Neural Network)辅助抽样的方法削减计算成本,但对于非线性程度较高的问题,RBFNN精度问题使得辅助抽样方法失效。针对这类问题,根据GPD的特点,提出了高效的更新RBFNN训练样本的方法,改善了RBFNN在功能函数分布尾部的精度,将RBFNN辅助抽样方法推广应用到非线性程度较高的问题,准确地得到了所有需要的尾部样本,基于该尾部样本集的GPD拟合结果与基于直接计算所有样本的GPD拟合结果完全一致。     

VIBRATION SUPPRESSION OF A FLEXIBLE PIEZOELECTRIC BEAM USING BP NEURAL NETWORK CONTROLLER

This paper aims at modeling and developing vibration control methods for a flexible piezoelectric beam. A collocated sensor/actuator placement is used. Finite element analysis (FEA) method is adopted to derive the dynamics model of the system. A back propagation neural network (BPNN) based proportional-derivative (PD) algorithm is applied to suppress the vibration. Simulation and experiments are conducted using the FEA model and BPNN-PD control law. Experimental results show good agreement with the simulation results using finite element modeling and the neural network control algorithm.     

Prediction of void fraction for gas–liquid flow in horizontal,upward and downward inclined pipes using artificial neural network

In this work, the ability of artificial neural networks (ANNs) to predict void fraction of gas–liquid two–phase flow in horizontal and inclined pipes was investigated. For this purpose, an ANN model was designed and trained using a total of 301 experimental data points reported in the literature for inclination angles between –20° and +20°. Pipe inclination angle as well as superficial Reynolds number of gas (Re_sg) and liquid (Re_sl) were chosen as input parameters of different structures of multilayer perceptron (MLP) neural networks, while the corresponding void fraction was selected as their output parameter. A hyperbolic tangent sigmoid and a linear function were employed as transfer functions of hidden and output layers, respectively, and Levenberg–Marquardt back propagation algorithm was used to train the networks. By trial–and–error method, a three–layer network with 10 neurons in the hidden layer was achieved as optimal structure of the ANN which made it possible to predict the void fraction with a high accuracy. Mean absolute percent error (MAPE) of 1.81% and coefficient of determination (R²) of 0.9976 for training data and MAPE of 1.52% and R² value of 0.9948 for testing data were obtained. Also for all data, MAPE of 1.95% and R² value of 0.9972 were calculated, and 96% data were within ±5% error band. In addition, the accuracy of the proposed ANN model was compared with the predictions from 17 void fraction correlations available in the literature for different flow patterns and horizontal and inclined flows. For all cases, the proposed ANN model gave better performance than all of the studied correlations. The results confirm the very good capability of the ANNs to predict the void fractions of gas–liquid flow in inclined pipes, regardless of flow pattern. Finally, by performing interpolation using the trained network, the void fraction values for some other conditions were predicted.     

Stability properties of a two-dimensional system involving one Caputo derivative and applications to the investigation of a fractional-order Morris–Lecar neuronal model

Adaptive infinite impulse response filters have received much attention due to its utilization in a wide range of real-world applications. The design of the IIR filters poses a typically nonlinear, non-differentiable and multimodal problem in the estimation of the coefficient parameters. The aim of the current study is the application of a novel hybrid optimization technique based on the combination of cellular particle swarm optimization and differential evolution called CPSO–DE for the optimal parameter estimation of IIR filters. DE is used as the evolution rule of the cellular part in CPSO to improve the performance of the original CPSO. Benchmark IIR systems commonly used in the specialized literature have been selected for tuning the parameters and demonstrating the effectiveness of the CPSO–DE method. The proposed CPSO–DE method is experimentally compared with two new design methods: the tissue-like membrane system (TMS), the hybrid particle swarm optimization and gravitational search algorithm (HPSO–GSA), the original CPSO-outer and CPSO-inner, and classical implementations of PSO, GSA and DE. Computational results and comparison of CPSO–DE with the other evolutionary and hybrid methods show satisfactory results. The hybridization of CPSO and DE demonstrates powerful estimation ability. In particular, to our knowledge, this hybridization has not yet been investigated for the IIR system identification.     

Identification of Damping and Dynamic Young’s Modulus of a Structural Adhesive Using Radial Basis Function Neural Networks and Modal Data

In this paper, the radial basis function neural networks (RBFNN) was applied to the problem of identifying dynamic Young’s modulus and damping characteristic of a structural adhesive, using modal data. To identify Young’s modulus from undamped model, an appropriate RBFNN using modal data (mode shape and natural frequency) in each mode is developed. Based on a previous work, in order to identify loss factor, two approaches adopted in the identification process. In the first one, a two stage RBFNN is developed. In stage I, Young’s modulus is identified from undamped model and in stage II using the results of stage I an appropriate RBFNN is developed in each mode for identification of loss factor by implementing real parts of eigenvalues of damped model. In the second approach, a one stage RBFNN is developed using real and imaginary parts of eigenvalues of damped model to identify Young’s moduli and loss factors simultaneously. The repeatability and consistency of the method is proved by repeating the identification process for several times. The validity of results is proved by comparing the results with those identified in a previous work.     

珩磨缸套表面粗糙度预测及多目标优化研究

为了对粗珩阶段缸套内孔表面粗糙度Rk粗糙度集中的Rk、Rpk和Rvk进行预测,进而对粗珩加工参数进行优化,以珩磨压力(P)、珩磨头旋转速度(VR)和往复速度(VRe)为决定因素,Rk粗糙度集为目标响应,进行多目标优化.建立基于广义回归神经网络(Generalized regression neural network, GRNN)与响应曲面法(Response surface methodology,RSM)的粗糙度预测模型,并采用三因素三水平的全因子珩磨试验进行验证,结果表明所建立模型的预测结果与试验结果具有很好的一致性. GRNN预测模型决定系数R²的均值为0.959,RSM多元回归预测模型决定系数R²的均值为0.963,与RSM所建立的多元回归预测模型相比,GRNN预测模型在预测Rk和Rpk时,预测精度更高,预测误差更小,R²分别提高了0.025和0.020,在预测Rvk时RSM多元回归模型更优,R²提高了0.057.进一步结合响应曲面法分析了3个决定因素对粗糙度的影响显著性并进行了排序,对于...     

新型路基同心筒热冲击机理与热环境影响因子

针对新型同心筒自力发射高速热冲击载荷下热环境评估与影响因子决策问题,结合弹性变形和域动分层结合的动网格技术,求解了二维轴对称Navier-Stokes方程,分析了新型路基同心筒流场机理与热冲击特性,并确定了热环境评价指标;通过建立以优化拉丁超立方试验设计和径向基神经网络为理论基础的近似数学模型,解决了CFD自动建模困难、计算量大的难点;结合径向基神经网络训练方法,对导弹热环境的影响因子进行了智能决策研究。分析表明:倒吸进入新型同心筒内筒的低温气体有力改善了同心筒热环境;建立的近似模型精度较高,满足工程需求;对导弹热环境的影响因子从大到小依次为筒底导流板直径、筒底导流板长度、导流器高度;为导弹热环境多学科优化设计提供参考。     

Experimental studies on critical heat flux in vertical tight 37-rod bundles using freon-12

The design of tight-lattice pressurized water reactors requires the knowledge of CHF in tight rod bundles. Experimental investigations on CHF behavior in tight hexagonal 37-rod bundles were performed by using the model fluid Freon-12. About 400 CHF data points have been obtained in a range of parameters: pressure 1.0–2.7 MPa, mass flux 1.4–4.5 Mg/m² s and bundle exit steam quality −0.4 to +0.2. It is found that the effect of different parameters on CHF in the tight rod bundle is similar to that in tube geometries. The present test results agree also quantitatively well with the CHF data obtained in tubes of comparable hydraulic diameters. Some in the open literature well known CHF correlations proposed for rod bundles under-predict the test results significantly.     

The prediction of maximum temperature for single chips’ cooling using artificial neural networks

A CFD simulation usually requires extensive computer storage and lengthy computational time. The application of artificial neural network models to thermal management of chips is still limited. In this study, the main objective is to find a neural network solution for obtaining suitable thickness levels and material for a chip subjected to a constant heat power. To achieve this aim a neural network is trained and tested using the results of the CFD program package Fluent. The back-propagation learning algorithm with three different variants, single layer and logistic sigmoid transfer function is employed in the network. By using the weights of the network, various formulations are designed for the output. The network has resulted in R ² values of 0.999, and the mean% errors smaller than 0.8 and 0.7 for the training and test data, respectively. The analysis is extended for different thickness and input power values. Comparison of some randomly selected results obtained by the neural network model and the CFD program has yielded a maximum error of 1.8%, mean absolute percentage error of 0.55% and R ² of 0.99994.     

车载MIMU/GPS/地磁系统的模糊广义RBFNN辅助滤波算法

以车载微惯性测量单元/GPS/地磁系统为研究对象,构造一类模糊广义径向基函数网络辅助滤波器,完成对基于EKF的非线性导航滤波解算,以提高导航系统参数估算精度和系统动态性能.相同条件下的仿真表明,对比标准EKF和模糊广义径向基函数网络辅助滤波方法,采用后者获得的导航参数误差均方差小,统计特性好,对姿态、航向角误差的最优估计分别控制在0.2°和0.4°以内.导航解算对微惯性测量单元误差在一定范围内的变动不敏感,保证了测量的精度.     

基于前向线性预测算法的光纤陀螺零漂的神经网络建模

在详细分析光纤陀螺零漂的基础上,提出了先用滤波算法对光纤陀螺信号进行预处理,然后采用RBF神经网络对滤波后的信号进行建模的方法。针对光纤陀螺信号特点分别采用FLP算法、小波滤波算法、解相关变步长LMS自适应滤波算法对其进行了预处理,比较三种滤波方法,小波滤波算法效果优于其它两种预处理方法,但针对基于预处理后的陀螺信号采用RBF神经网络进行建模时,小波滤波预处理后的信号在建模精度上却是最差的,而对FLP算法滤波后的信号进行RBF建模,建模精度提高了两个数量级。结果表明:基于FLP算法的RBF神经网络在光纤陀螺中的建模是有效的,可大大提高建模的精度。     

Adaptive trajectory tracking of magnetostrictive actuator based on preliminary hysteresis compensation and further adaptive filter controller

An adaptive dynamic surface control (DSC) scheme is proposed for the multi-input multi-output attitude control of near-space hypersonic vehicles (NHV). The proposed control strategy can improve the control performance of NHV despite uncertainties and external disturbances. The proposed controller combines dynamic surface control and radial basis function neural network (RBFNN) and is designed to control the longitudinal dynamics of NHV. The DSC technique is used to handle the problem of “explosion of complexity” inherent to the conventional backstepping method. RBFNN is used to approximate the unknown nonlinear function, and a robustness component is introduced in the controller to cancel the influence of compound disturbance and improve robustness and adaptation of the system. Simulation results show that the proposed strategy possesses good robustness and fast response.     

Prediction of CHF in concentric-tube open thermosiphon using artificial neural network and genetic algorithm

In this paper, an artificial neural network (ANN) for predicting critical heat flux (CHF) of concentric-tube open thermosiphon has been trained successfully based on the experimental data from the literature. The dimensionless input parameters of the ANN are density ratio, ρ _l/ρ _v; the ratio of the heated tube length to the inner diameter of the outer tube, L/D _i; the ratio of frictional area, d _i/(D _i + d _o); and the ratio of equivalent heated diameter to characteristic bubble size, D _he/[σ/g(ρ _l−ρ _v)]^0.5, the output is Kutateladze number, Ku. The predicted values of ANN are found to be in reasonable agreement with the actual values from the experiments with a mean relative error (MRE) of 8.46%. New correlations for predicting CHF were also proposed by using genetic algorithm (GA) and succeeded to correlate the existing CHF data with better accuracy than the existing empirical correlations.     

低地球轨道环境中原子氧对空间润滑油的影响

选择硅油(CPSO)、全氟聚醚油(Fomblin Z25)和硅碳氢油(SiCH)3种典型的空间用润滑油,进行地面模拟原子氧暴露试验,考察了原子氧对这3种润滑油的质量、外观、分子量以及摩擦学性能的影响.结果表明:原子氧辐照会导致润滑油质量损失、产生固体不溶物并导致润滑油分子量增大和分子量分布变宽。同时摩擦试验发现,原子氧辐照后润滑油的摩擦系数变得不稳定,有突然增大的现象.3种润滑油原子氧暴露试验结果综合分析表明Fomblin Z25较容易受原子氧影响,CPSO次之,SiCH油表现最佳.     

The interference wave and the bright and dark soliton for two integro-differential equation by using BNNM

Interference wave is an important research target in the field of navigation, electromagnetic and earth science. In this work, the nonlinear property of neural network is used to study the interference wave and the bright and dark soliton solutions. The generalized broken soliton-like equation is derived through the generalized bilinear method. Three neural network models are presented to fit explicit solutions of generalized broken soliton-like equations and Boiti–Leon–Manna–Pempinelli-like equation with 100% accuracy. Interference wave solutions of the generalized broken soliton-like equation and the bright and dark soliton solutions of the Boiti–Leon–Manna–Pempinelli-like equation are obtained with the help of the bilinear neural network method. Interference waves and the bright and dark soliton solutions are shown via three-dimensional plots and density plots.

     

Simulation of cross-flow-induced vibration of cylinder arrays by surface vorticity method

The surface vorticity method (SVM), which is a fast and practical grid-free two-dimensional (2-D) method, and a fluid–structure interaction model incorporating the effects of cylinder motions and displacements is used to simulate the vortex-induced vibration of cylinder arrays at sub-critical Reynolds number Re=2.67×10⁴. The SVM is found to be most suitable for simulating a 2-D cylinder row with large-amplitude vibrations where the vorticity field and the fluid forces of the cylinder row change drastically, and the effect of the stream on the transverse direction vibration is very significant. The fluidelastic instability of a flexible cylinder row at small pitch ratio is also investigated, and the critical reduced velocity of the cylinder row at a reduced damping parameter S_G=1.29 is calculated, which is in good agreement with experimental and analytical results of the unsteady model. Vortex-induced vibration of a staggered cylinder array is simulated using different structural parameters. When the cylinders are relatively more flexible, the flow pattern changes dramatically and the fluid–structure interaction has a dominant impact on the flow field. Compared with grid-based methods, the grid-free SVM is a fast and practical method for the simulation of the FIV of cylinder arrays due to vortex shedding at sub-critical Reynolds numbers.     

Nonlinear reduced-order model for vertical sloshing by employing neural networks

The aim of this work is to provide a reduced-order model to describe the dissipative behavior of nonlinear vertical sloshing involving Rayleigh–Taylor instability by means of a feed forward neural network. A 1-degree-of-freedom system is taken into account as representative of fluid–structure interaction problem. Sloshing has been replaced by an equivalent mechanical model, namely a boxed-in bouncing ball with parameters suitably tuned with performed experiments. A large data set, consisting of a long simulation of the bouncing ball model with pseudo-periodic motion of the boundary condition spanning different values of oscillation amplitude and frequency, is used to train the neural network. The obtained neural network model has been included in a Simulink®  environment for closed-loop fluid–structure interaction simulations showing promising performances for perspective integration in complex structural system.

     

基于ANN的碳纤维楠竹锚杆锚固力预测研究

为减小对文物本体的破坏,本文基于新疆某土遗址加固保护中碳纤维楠竹锚杆锚固力原位测试试验,考虑锚杆直径、长度、倾斜角以及灌浆体强度、孔径、碳纤维缠绕间距等锚固力影响因素,利用人工神经网络(artificial neural network, ANN) 的误差反向传播(back propagation, BP) 算法及MATLAB 人工神经网络工具箱,建立了锚固力预测的智能模型;并以原位测试所得的数据为学习样本和检验样本,验证了该方法的适用性和可行性. 将训练好的网络模型进行扩展计算,基于L₂₅(5⁶) 正交表试验理论分析了锚固力对各影响因素的敏感性,为同类加固工程的实际应用提供参考依据.     

一种基于多尺度和改进支持向量机的光纤陀螺温度漂移建模与补偿方法

为了提升光纤陀螺温度漂移模型建模的准确性及补偿的效果,提出了一种基于改进支持向量机的多尺度建模和回归方法。首先分析了造成光纤陀螺温度漂移的关键因素,给出了建模的属性参数和温度试验。然后根据经验模态分解得到的本征模态函数排列熵的变化趋势,得出了回归精度和熵之间的变化关系,进而提出了基于信号分解的多尺度回归方法。为了提高上述多尺度回归算法的适应性,在传统支持向量机的基础上,提出了基于组合核函数的支持向量机回归算法,以适应不同特性的回归数据集。为了进一步提高回归精度,基于降低回归数据复杂度的分段回归思想,在上述多尺度回归的基础上提出了双-多尺度回归,并验证了方法的有效性。最后,将提出的算法以实际的光纤陀螺温度漂移数据进行验证,结果表明,相比于传统的支持向量机和反向传播神经网络具有更好的回归精度,温度漂移模型也更加精确,以均方误差指标为例,回归精度提升了两个数量级。