Design of parts for cellular manufacturing using neural network-based approach

A neural network approach is applied to the problem of integrating design and manufacturing engineering. The self organising map (SOM) neural network recognizes products and parts which are modeled as boundary representation (B-rep) solids using a modified face complexity code scheme adopted, and forms the necessary feature families. Based on the part features, machines, tools and fixtures are selected. These information are then fed into a four layer feed-forward neural network that provides a designer with the desired features that meet the current manufacturing constraints for design of a new product or part. The proposed methodology does not involve training of the neural networks used and is seen to be a significant potential for application in concurrent engineering where design and manufacturing are integrated.     

A self-organizing neural network approach for the design of cellular manufacturing systems

The crux problem of group technology (GT) is the identification of part families requiring similar manufacturing processes and the rearrangement of machines to minimize the number of parts that visit more than one machine cell. This paper presents an improved method for part family formation, machine cell identification, bottleneck machine detection and the natural cluster generation using a self-organizing neural network. In addition, the generalization ability of the neural network makes it possible to assign the new parts to the existing machine cells without repeating the entire computational process. A computer program is developed to illustrate the effectiveness of this heuristic method by comparing it with the optimal technique for large-scale problems.     

A joint investigation of misclassification treatments and imbalanced datasets on neural network performance

Two important factors that impact a classification model’s performance are imbalanced data and unequal misclassification cost consequences. These are especially important considerations for neural network models developed to estimate the posterior probabilities of group membership used in classification decisions. This paper explores the issues of asymmetric misclassification costs and unbalanced group sizes on neural network classification performance using an artificial data approach that is capable of generating more complex datasets than used in prior studies and which adds new insights to the problem and the results. A different performance measure, that is capable of directly measuring classification performance consistency with Bayes decision rule, is used. The results show that both asymmetric misclassification costs and imbalanced group sizes have significant effects on neural network classification performance both independently and via interaction effects. These are not always intuitive; they supplement prior findings, and raise issues for the future.     

基于自动驾驶系统的轻量型卷积神经网络优化

计算机视觉技术大量应用于自动驾驶系统,主要解决物体识别与物体分类问题,本文根据任务提出了一种轻量化的神经网络结构.为解决训练数据规模不足的问题,采用了改进型数据增强算法,使训练数据成倍增加.同时为解决使用数据生成器作为验证集,无法使用tensorboard的问题,提出了解决方案,通过卷积网络可视化方法详细研究了神经网络处理图像信息的原理并提出了优化方法.训练后的模型在验证集上准确率达到了97.5%,满足了自动驾驶系统对分类任务准确率的要求.     

An intelligent grouping algorithm for cellular manufacturing

The methodology presented in this paper will provide a means of identifying part families/machine cells using design and manufacturing characteristics simultaneously. The technique used is a self-organizing neural network called interative-activation and competition (IAC) which acts as a content-addressable memory. This neural network is used to define a similarity index of the pairwise comparisons of parts based on a variety of design and manufacturing characteristics. A bond energy algorithm partitions the matrix of part similarity indices to create part families and inferred from the part families are machine cells. A brief example will be examined as well as discussion of the results.     

Neural network-based design of cellular manufacturing systems

A neural network based on a competitive learning rule, when trained with the part machine incidence matrix of a large number of parts, classifies the parts and machines into part families and machine cells, respectively. This classification compares well with the classical clustering techniques. The steady state values of the activations and interconnecting strengths enable easier identification of the part families, machine cells, overlapping parts and bottleneck machines. Neural networks are mostly applied by treating them as a blackbox, i.e. the interaction with the environment and the information acquisition and retrieval occurs at the input and the output level of the network. This paper presents an approach where knowledge is extracted from the external and internal structure of the neural network.     

Segmentation of Assamese phonemes using hybrid soft-computational approach

Phonemes are the smallest distinguishable unit of speech signal. Segmentation of a phoneme from its word counterpart is a fundamental and crucial part in speech processing because an initial phoneme is used to activate words starting with that phoneme. This work describes an artificial neural network-based algorithm developed for segmentation and classification of consonant phoneme of the Assamese language. The algorithm uses weight vectors, obtained by training self-organising map (SOM) with different number of iterations, as a segment of different phonemes constituting the word whose linear prediction coefficients samples are used for training. The algorithm shows an abrupt rise in success rate than the conventional discrete wavelet-based speech segmentation. A two-class probabilistic neural network problem carried out with clean Assamese phoneme is used to identify phoneme segment. The classification of the phoneme segment is alone as per the consonant phoneme structure of the Assamese language which consists of six phoneme families. Experimental results establish the superiority of the SOM-based segmentation over the discrete wavelet transform-based approach.     

The modified fuzzy art and a two-stage clustering approach to cell design

This study presents a new pattern recognition neural network for clustering problems, and illustrates its use for machine cell design in group technology. The proposed algorithm involves modifications of the learning procedure and resonance test of the Fuzzy ART neural network. These modifications enable the neural network to process integer values rather than binary valued inputs or the values in the interval [0, 1], and improve the clustering performance of the neural network. A two-stage clustering approach is also developed in order to obtain an informative and intelligent decision for the problem of designing a machine cell. At the first stage, we identify the part families with very similar parts (i.e., high similarity exists in their processing requirements), and the resultant part families are input to the second stage, which forms the groups of machines. Experimental studies show that the proposed approach leads to better results in comparison with those produced by the Fuzzy ART and other similar neural network classifiers.     

Cooperative Self-Organizing Maps for Consistency Checking and Signature Verification

The successful implementation of an automatic biometric system relies mainly on the consistency of the used training sets. Signatures of the same writer are similar but not identical, since they can differ both globally and locally, in location, scale, and orientation. In contrast with fingerprints, signatures that are completely authentic never exist. This paper emphasizes the application of a competitive neural network architecture for checking the consistency of the data set belonging to an individual in a biometric database. A neural network based consistency measure is proposed to quantify the intra-variability of the individuals signatures. A new democratic neural network architecture is then presented for minimization of the rejection error and maximization of the percentage of correct classification based on some well-known features and a new feature set.     

含驱动器动力学的移动机器人编队自适应控制

针对含有驱动器及编队动力学的多非完整移动机器人编队控制问题,基于领航者-跟随者[l-ψ]控制结构,通过反步法设计了一种将运动学控制器与驱动器输入电压控制器相结合的新型控制策略。采用径向基神经网络（RBFNN）对跟随者及领航者动力学非线性不确定部分进行在线估计,并通过自适应鲁棒控制器对神经网络建模误差进行补偿。该方法不但解决了移动机器人编队控制的参数与非参数不确定性问题,同时也确保了机器人编队在期望队形下对指定轨迹的跟踪;基于Lyapunov方法的设计过程,保证了控制系统的稳定与收敛;仿真结果表明了该方法的有效性。     

基于熵的自组织神经网络树

神经网络由于优越的学习和分类能力已被用于许多模式识别的问题,并取得了很好的结果。但是对于识别大样本集和复杂模式的问题,绝大多数常规的神经网络在决定网络的结构和规模以及应付庞大的计算量等方面有着种种困难。为了克服这些困难,文中提出一种基于条件类别熵的结构自适应的神经网络树;这种神经网络树由具有拓扑有序特性的子网络组成,而树的规模由条件类别熵决定。它的主要优点是对于识别大样本集和复杂模式的问题能够通过结构自适应自动地确定网络的结构和规模。实验显示这种神经网络树对于识别大样本集和复杂模式是非常有效的。     

基于双通道卷积神经网络的文本情感分类算法

针对现有深度学习方法在文本情感分类任务中特征提取能力方面的不足,提出基于扩展特征和动态池化的双通道卷积神经网络的文本情感分类算法.首先,结合情感词、词性、程度副词、否定词和标点符号等多种影响文本情感倾向的词语特征,形成一个扩展文本特征.然后,把词向量特征与扩展文本特征分别作为卷积神经网络的两个输入通道,采用动态k-max池化策略,提升模型提取特征的能力.在多个标准英文数据集上的文本情感分类实验表明,文中算法的分类性能不仅高于单通道卷积神经网络算法,而且相比一些代表性算法也具有一定的优势.     

带比例因子的卷积神经网络压缩方法

针对卷积神经网络在图像分类任务中,分类准确率高但实时性差的问题。提出了一种含比例因子的“知识提取”算法。此方法在已有的“知识提取”算法上,加入了衡量样本类间相近关系的比例因子,充实了网络压缩手段,使得神经网络可以更精确地进行“知识提取”。其原理是将比例因子误差值作为代价函数的一部分参与训练调节神经网络的神经元参数,进而使得神经网络的泛化能力更加趋近于具有更好分类表现能力的压缩参考网络。结果表明,含比例因子的神经网络压缩算法可以更细致地刻画训练集的类间相近关系,拥有比原“知识提取”算法更好的训练性能,进而训练出泛化性能更强、精度更高的神经网络。实现了在网络分类准确率下降尽量小的前提下,较大程度地减少神经网络的分类耗时,以达到网络压缩的目的。     

Advanced machine learning techniques for microarray spot quality classification

It is well known that microarray printing, hybridization, and washing oftentimes create erroneous measurements, and these errors detrimentally impact machine microarray spot quality classification. Thus, it is crucial to identify and remove these errors if automation is to replace the still common practice of visually assessing spot quality, an extremely expensive and time-consuming procedure. A major problem in microarray spot quality classification methods proposed in the literature is the correlation among the features extracted from the spots. In this paper, we propose using a random subspace ensemble of neural networks and a feature selection algorithm to improve the performance of our microarray spot quality classification method. Our best method obtains an error under the receiver operating characteristic curve (EAUR) of 0.3 outperforming the stand-alone support vector machine EAUR of 1.7. The consistency of our proposed approach makes it a viable alternative to the labour-intensive manual method of spot quality assessment.     

基于Kohonen神经网络的分形图像编码

本文提出利用Kohonen自组织神经网络把母块分类与特征抽取结合起来有助于改善分形编码的时间。因为特征抽取减少了问题的维数并且使网络能够在一幅和实验图像分离的图像上得到训练。自组织网络为分类引入了一个领域拓扑结构，并且不需要事先指定一组适当的图像类。网络按照在训练期间观测的图像特征的分布来组织自己。结果表明，该分类方法可以将编码时间减少两个数量级并保持可观的精度和压缩性能。     

深度卷积神经网络胸片肺结节分类识别研究

针对肺结节特征复杂且不明显,难以精确诊断出胸片中是否含有肺结节的问题,提出将深度神经网络应用于肺结节分类识别之中。首先通过将胸片灰度一致化,减少由于不同设备导致胸片亮度与灰度的差异;其次采用不同的数据扩增方法使得深度卷积神经网络可以充分提取肺结节的特征;最后通过改进的神经网络架构对肺结节进行分类识别。提出的算法有效地避免了在对胸片图像进行分割时造成图像特征部分丢失的现象,同时克服了由于胸片图像的复杂造成的肺结节特征不明显的缺点。最终通过实验研究证明胸片肺结节分类识别的平均准确率达到84.2%,在医学胸片肺结节的分类识别领域上具有一定的应用价值。     

Neural Networks for Full-Scale Protein Sequence Classification: Sequence Encoding with Singular Value Decomposition

A neural network classification method has been developed as an alternative approach to the search/organization problem of protein sequence databases. The neural networks used are three-layered, feed-forward, back-propagation networks. The protein sequences are encoded into neural input vectors by a hashing method that counts occurrences ofn-gram words. A new SVD (singular value decomposition) method, which compresses the long and sparsen-gram input vectors and captures semantics ofn-gram words, has improved the generalization capability of the network. A full-scale protein classification system has been implemented on a Cray supercomputer to classify unknown sequences into 3311 PIR (Protein Identification Resource) superfamilies/families at a speed of less than 0.05 CPU second per sequence. The sensitivity is close to 90% overall, and approaches 100% for large superfamilies. The system could be used to reduce the database search time and is being used to help organize the PIR protein sequence database.     

模糊神经网络及其在时间序列分析中的应用

给出了一种新型的模糊神经网络模型.该模型不需要领域专家的知识进行指导,而是通过对样本竞争分类产生模糊规则.每类样本对应于一条模糊规则,每条模糊规则的后件部分为一个对本类样本进行过学习训练的神经网络.文章以模糊神经网络在时间序列分析中的应用为例,通过与传统的时间序列分析方法以及前向神经网络方法的对比,说明了新型模糊神经网络的有效性.     

Minimal Structure of Self-Organizing HCMAC Neural Network Classifier

The authors previously proposed a self-organizing Hierarchical Cerebellar Model Articulation Controller (HCMAC) neural network containing a hierarchical GCMAC neural network and a self-organizing input space module to solve high-dimensional pattern classification problems. This novel neural network exhibits fast learning, a low memory requirement, automatic memory parameter determination and highly accurate high-dimensional pattern classification. However, the original architecture needs to be hierarchically expanded using a full binary tree topology to solve pattern classification problems according to the dimension of the input vectors. This approach creates many redundant GCMAC nodes when the dimension of the input vectors in the pattern classification problem does not exactly match that in the self-organizing HCMAC neural network. These redundant GCMAC nodes waste memory units and degrade the learning performance of a self-organizing HCMAC neural network. Therefore, this study presents a minimal structure of self-organizing HCMAC (MHCMAC) neural network with the same dimension of input vectors as the pattern classification problem. Additionally, this study compares the learning performance of this novel learning structure with those of the BP neural network,support vector machine (SVM), and original self-organizing HCMAC neural network in terms of ten benchmark pattern classification data sets from the UCI machine learning repository. In particular, the experimental results reveal that the self-organizing MHCMAC neural network handles high-dimensional pattern classification problems better than the BP, SVM or the original self-organizing HCMAC neural network. Moreover, the proposed self-organizing MHCMAC neural network significantly reduces the memory requirement of the original self-organizing HCMAC neural network, and has a high training speed and higher pattern classification accuracy than the original self-organizing HCMAC neural network in most testing benchmark data sets. The experimental results also show that the MHCMAC neural network learns continuous function well and is suitable for Web page classification.