基于自适应遗传BP算法的混合气体定量检测研究

针对混合气体检测问题,利用误差反向传播(BP)算法和遗传算法,提出了用自适应遗传算法优化BP神经网络的方法来实现定量检测.即利用遗传算法的全局搜索能力,对神经网络连接权值和阈值进行优化,再以优化后的初值作为BP神经网络的初始连接权值和阈值,最后用附加“动量项”的误差反向传播算法训练BP网络.设计了一个结构为7×18 ×3的3层BP网络用于一组含有5个样本的混合气体识别试验.结果表明:将自适应遗传神经网络算法应用于混合气体定量识别的训练中,收敛速度比附加“动量项”BP算法要快,而且学习精度更高,识别效率也提高了2/3.     

基于神经网络信息融合的柴油机故障诊断

本文论述了神经网络信息融合的原理与方法，首先就BP网络训练速度慢，易陷入局部极小点的问题，提出将附加动量项与自适应学习速率相结合的改进BP算法，有效地抑制了网络陷于局部极小并提高了收敛速度。最后，将振动信号与血管压力信号作为特征参数，分别采用传统BP算法，改进BP算法对供油系统的三种故障进行信息融合诊断分析。实践表明，神经网络信息融合方法非常适用于多征兆机械系统的故障诊断。     

压铸成型工艺设计与缺陷分析的神经网络模拟

对于误差逆传播 (BP)神经网络标准及改进型算法中的神经元联接权更新机制进行了分析。作为实际应用 ,基于MATLAB神经网络工具箱函数 ,研究了两个例子 ,一个是不同合金种类及不同复杂程度的型腔结构下浇注温度的选择 ;另一个是压铸工艺的缺陷分析。采用的算法分别为恒定学习率 附加动量项改进型BP算法和自适应学习率 附加动量项改进型BP算法。通过在实际模拟中比较这两种算法的训练效率 ,可得到结论 :自适应学习率 附加动量项算法是精确模拟压铸工艺映射问题的较为理想的方法。基于这一方法 ,文中最后给出了网络模拟测试的结果     

基于改进神经网络的多重化可控整流电路的故障诊断

提出了一种改进神经网络的多重化可控整流电路故障诊断方法。针对BP算法收敛速度慢、容易陷入局部极小的缺点,采用了改进的BP算法,利用附加动量学习速率可变梯度法改进神经网络学习速率,从而达到了加快网络的收敛速度与提高网络的诊断精度。文中给出了基于改进BP网络的模型训练和测试数据的诊断。测试结果表明所提出的方法正确可行,该诊断方法对多重化可控整流电路故障的识别可达到较高的精度。     

锅炉经济性运行指导专家系统

采用专家系统和模糊神经网络相结合的方法开发了锅炉经济性运行指导系统。该系统从在线监测的实时参数出发,利用适当的隶属函数将其模糊化．通过神经网络计算找出异常现象发生原因并给出专家建议,以指导操作人员进行迅速、准确地操作。本文对BP网络进行了改进,引入变步长算法来调整学习率,同时附加动量项来修改网络权值和阈值。最后利用该系统对锅炉排烟温度高的问题进行分析研究,结果表明该系统收敛速度快,精度高．能够提高电厂运行管理水平．提高了锅炉的经济性。     

PID神经网络改进研究

PID神经网络（PIDNN）是将神经网络和PID控制规律融为一体的新型前向神经元网络。对于多数系统,PIDNN可以利用已有PID控制规律的先验知识确定网络权重初值,使系统得到稳定快速的控制。当权重初值选择为随机数时,收敛速度变慢,同时可能陷入局部最小。针对这一类系统,提出附加动量项的改进算法,克服权重初值取随机数带来的问题。系统仿真结果证明改进后的PIDNN系统性能得到了明显改善。     

遗传小波网络在齿轮箱故障诊断中的应用

针对小波神经网络常用的反向传播算法普遍存在收敛速度慢,容易陷入局部极小点,网络参数的选取只能凭实验和经验来确定等缺点,提出了一种基于遗传算法优化的小波神经网络并应用于齿轮的故障诊断。仿真结果表明,该方法充分的发挥了遗传算法的全局寻优能力,小波分析的非线性逼近能力和神经网络的自学习特性,优化了系统的收敛速度和故障诊断的精度。     

一种萤火虫神经网络及在轴承故障诊断中的应用

萤火虫算法作为一种群体智能算法,具有良好的全局寻优特性,可用于解决神经网络在反向传播(Back propagation,BP)算法下易陷入局部极值点、收敛速度慢的问题。但在应用于神经网络时其参数选取缺乏有效依据或经验公式,参数取值不当时,萤火虫神经网络的训练误差无法有效收敛、种群个体之间协同寻优功能易失效。因此,提出一种双层萤火虫改进算法对其参数进行选取。用UCI数据及轴承故障试验对算法进行验证,结果表明,应用所提方法后萤火虫神经网络的训练误差收敛值显著减小,并且在训练效率、故障识别率方面均优于BP神经网络、遗传神经网络、粒子群神经网络。     

改进遗传算法在减速器优化设计中的应用

针对简单遗传算法的早熟现象及不能处理带有复杂约束的优化问题,提出一种基于乘子法与伪并行遗传算法的改进遗传算法,并将其应用于斜齿轮减速器优化设计中。计算结果表明改进遗传算法,全局寻优能力强。     

基于机器视觉和神经网络的低质量文本识别研究

针对流水线产品上的文本图片含有较多噪声和缺陷,造成机器视觉产品中字符识别准确率低,鲁棒性差的问题,对文本识别中的图像预处理、字符切分和归一化、字符识别等方面进行了研究,采用机器视觉技术中基于仿射变换的预处理方法对文本图片进行了倾斜校正,保证了后续字符的精确切分;提出了一种基于改进的BP神经网络算法,显著提高了字符识别的准确率和鲁棒性,利用附加动量法和自适应学习速率方法,避免了传统BP神经网络训练过程中易陷入局部极小值的情况,并提高了神经网络模型的收敛速度。研究结果表明:该方法能使倾斜的图片得到有效校正,神经网络模型的训练效率得到有效提高,且在含有噪声和缺陷的字符图片中仍能保持较高的识别率和鲁棒性。     

基于混沌优化的多层前馈神经网络

BP算法是应用广泛的一种多层前馈神经网络模型,针对算法求解精度低、搜索速度慢、易于陷入局部极值点等问题,根据混沌理论的全局优化思想,提出采用"多次载波"技术将混沌优化和前馈神经网络相结合,利用已找到的近似最优解来启发搜索全局最优解的方法训练神经网络,以布尔函数识别、曲线逼近、模式识别3个典型应用对算法进行验证。研究结果表明,算法具有较好的泛化能力和快速全局收敛的性能,特别是针对中小规模的网络,混沌优化算法在训练时间、全局收敛率等指标方面优于BP算法。     

基于SSDAE深度神经网络的钛板电涡流检测图像分类研究

钛板电涡流成像检测易受工业现场中的噪声影响,包含噪声的检测图像往往难以提取较好的特征,从而影响分类识别精度。针对以上问题,提出了一种基于栈式稀疏降噪自编码(SSDAE)深度神经网络的钛板缺陷电涡流检测图像分类方法。将稀疏性限制引入降噪自编码器并进行逐层无监督自学习,然后将自编码器栈式组合后添加逻辑识别(LR)层,构建出SSDAE深度神经网络,网络在有监督微调后可实现钛板缺陷电涡流图像特征自动提取与分类识别。稀疏性限制的引入提高了特征学习能力,降噪自编码器的栈式组合提高了深度网络的鲁棒性。实验结果表明,相比其他常规方法,所提出方法不仅在理想环境下有更高的分类准确率,且该方法能有效抵抗噪声,在复杂工况下能更有效地对钛板缺陷进行分类识别。     

基于神经网络的薄壳多目标振动优化控制研究

针对目前主动控制方法主要集中于减振降噪方面的研究,无法满足工程中频率特性改变的需求等问题,结合神经网络的多目标并行处理能力,提出基于神经网络的多目标振动优化控制方法。首先,基于神经网络算法,构造频域主动控制架构,相较于时域方法,该架构一个控制循环只需一次傅里叶变换(Fast Fourier transform,FFT),无傅里叶逆变换(Inverse fast Fourier transform,IFFT),因此,控制时效性得到有效保证。其次,基于全局频域误差与特征频点误差,构造混合型误差评判准则,提升算法的自适应性,可靠性与抗干扰能力。再次,基于多自由度系统方程,研究了多目标控制中的可控性问题,保证控制的可行性。最后,通过大型薄壳结构的八点多目标振动优化控制,有效验证了方法的有效性与可行性。     

神经网络优化技术在磨床主轴优化中的应用

利用拉格朗日乘子法将约束优化设计问题转化为求解拉格朗日乘子法极值的问题，讨论拉格朗日乘子法极值存在的条件，基于人工神经网络非线性优化计算的优点，构造用于拉格朗日乘子法的人工神经网络学习规则和网络结构。优化计算平面磨床主轴系统的结构参数，讨论优化结果与Kuhn—Tucker准则优化结果的一致性。     

Tolerance allocation using neural networks

The purpose of tolerance allocation is to find a combination of tolerances to individual components such that the assembly tolerance constraint is met with minimum production cost. There are several methods available to allocate or apportion the assembly tolerance to individual parts. Some of the most common methods use linear programming, Lagrange multipliers, exhaustive search and statistical distributions. However, all the methods have some limitations. Moreover, most of these methods cannot account for the frequently observed mean shift phenomena that occur owing to tool wear, chatter, bad coolant, etc. This paper presents a neural networks-based approach for the tolerance allocation problem considering machines' capabilities, and mean shifts. The network is trained using the backpropagation learning method and used to predict individual part tolerances.     

Optimum surface roughness in end milling Inconel 718 by coupling neural network model and genetic algorithm

In this study, optimum cutting parameters of Inconel 718 are determined to enable minimum surface roughness under the constraints of roughness and material removal rate. In doing this, advantages of statistical experimental design technique, experimental measurements, artificial neural network and genetic optimization method are exploited in an integrated manner. Cutting experiments are designed based on statistical three-level full factorial experimental design technique. A predictive model for surface roughness is created using a feed forward artificial neural network exploiting experimental data. Neural network model and analytical definition of material removal rate are employed in the construction of optimization problem. The optimization problem was solved by an effective genetic algorithm for variety of constraint limits. Additional experiments have been conducted to compare optimum values and their corresponding roughness and material removal rate values predicted from the genetic algorithm. Generally a good correlation is observed between the predicted optimum and the experimental measurements. The neural network model coupled with genetic algorithm can be effectively utilized to find the best or optimum cutting parameter values for a specific cutting condition in end milling Inconel 718.     

Optimization of injection molding process parameters using integrated artificial neural network model and expected improvement function method

In this study, an adaptive optimization method based on artificial neural network model is proposed to optimize the injection molding process. The optimization process aims at minimizing the warpage of the injection molding parts in which process parameters are design variables. Moldflow Plastic Insight software is used to analyze the warpage of the injection molding parts. The mold temperature, melt temperature, injection time, packing pressure, packing time, and cooling time are regarded as process parameters. A combination of artificial neural network and design of experiment (DOE) method is used to build an approximate function relationship between warpage and the process parameters, replacing the expensive simulation analysis in the optimization iterations. The adaptive process is implemented by expected improvement which is an infilling sampling criterion. Although the DOE size is small, this criterion can balance local and global search and tend to the global optimal solution. As examples, a cellular phone cover and a scanner are investigated. The results show that the proposed adaptive optimization method can effectively reduce the warpage of the injection molding parts.     

An improved radial basis function network for structural reliability analysis

Approximation methods such as response surface method and artificial neural network (ANN) method are widely used to alleviate the computation costs in structural reliability analysis. However most of the ANN methods proposed in the literature suffer various drawbacks such as poor choice of parameter setting, poor generalization and local minimum. In this study, a support vector machine-based radial basis function (RBF) network method is proposed, in which the improved RBF model is used to approximate the limit state function and then is connected to a reliability method to estimate failure probability. Since the learning algorithm of RBF network is replaced by the support vector algorithm, the advantage of the latter, such as good generalization ability and global optimization are propagated to the former, thus the inherent drawback of RBF network can be defeated. Numerical examples are given to demonstrate the applicability of the improved RBF network method in structural reliability analysis, as well as to illustrate the validity and effectiveness of the proposed method.     

Research on image reconstruction algorithms based on autoencoder neural network of Restricted Boltzmann Machine (RBM)

Aiming at the problem of low quality in image reconstruction of traditional image reconstruction algorithm of electromagnetic tomography(EMT), an EMT image reconstruction algorithm based on autoencoder neural network of Restricted Boltzmann Machine (RBM) is proposed. Firstly, the basic principles of EMT system and autoencoder neural network are analyzed. Autoencoder neural network is a deep learning model, which contains two parts: encoder and decoder. The encoding process of the encoder is equivalent to the object field detection process in the EMT system; the decoding process of the decoder is equivalent to the image reconstruction process. On this basis, an autoencoder neural network model is built. In this model, the RBM is used for layer by layer pre-training to obtain the initial weight and offset, and the global weight and offset are adjusted by BP algorithm. The parameter file generated in the trained autoencoder neural network is used to construct a decoder. Finally, the detected voltage value output by the EMT system is input into the decoder network to obtain the reconstructed image of the EMT. Furthermore, data with Gaussian noise and data regarding flow pattern not in training dataset are used to test the generalization ability and practicability of the network, respectively. The experimental results show that the method in this paper is a kind of EMT image reconstruction method with higher accuracy, which also provides a new means for EMT image reconstruction.     

基于几何纹理与Anscombe变换的蜂窝材料太赫兹图像降噪模型

为解决太赫兹（Terahertz,THz）图像内泊松高斯混合噪声导致芳纶纤维蜂窝材料脱粘缺陷轮廓检测精度低的问题,基于Anscombe变换与小波阈值法构建了THz图像降噪模型。高斯噪声方差为降噪模型的必要参数,但实际THz图像噪声分布未知,且噪声与纹理在高频混叠,给方差准确估计提出了挑战。为此,首先以样件纹理几何形状为先验信息,构造Benzene-ring算子去除THz图像纹理,使其小波域高频分量中仅含有噪声;然后提出改进的Logistic混沌映射提高样本集的多样性,以训练Elman神经网络准确建立高频分量与高斯噪声方差间映射关系;最后依据噪声方差估计值,基于Anscombe变换将泊松高斯混合噪声转化为高斯噪声,并利用小波阈值法与Anscombe逆变换得到了最终THz降噪图像。仿真与试验结果表明,所提出的方法降噪效果最佳并有效提高缺陷轮廓检测精度,相比于高斯滤波、小波阈值以及非局部均值法,平均梯度指标分别提升12%、33%、9%,缺陷面积绝对误差分别降低234 mm²、304 mm²、263 mm²。