An improved fuzzy clustering method for cellular manufacturing

Fuzzy c-means (FCM) has been successfully adapted to solve the manufacturing cell formation problem. However, when the problem becomes larger and especially if the data is ill structured, the FCM may result in clustering errors, infeasible solutions, and uneven distribution of parts/machines. In this paper, an improved fuzzy clustering algorithm is proposed to overcome the deficiencies of FCM. We tested the effects of algorithm parameters and compared its performance with the original and two popular FCM modifications. Our study shows that the proposed approach outperformed other alternatives. Most of the solutions it obtained are close to and in some cases better than the control solutions.     

模糊聚类分析在模糊神经网络结构优化中的应用

研究了模糊聚类分析在多变量模糊神经网络的结构确定中的应用,在传统的模糊C－均值算法的基础上,给出了一个衡量聚类有效性的函数和确定模糊指数的启发式方法,并给出了应用该算法的具体的模糊神经网络模型。     

A simulation analysis of factors influencing the flexibility of cellular manufacturing

The performance of cellular manufacturing (CM) systems in a variable demand and flexible workforce environment has been examined using simulation modelling. Discrepancies between academicians and practitioners’ findings with respect to flexibility and uneven machine utilization in CM systems are discussed. The views of two parties were incorporated in simulation models to rectify the existing discrepancies. While the results of this study confirm the previous findings of academicians regarding the deterioration of the performance of CM in a variable product mix situation, it appears that those results may be significantly influenced by considering a flexible workforce. The simulation results show that the practice of using flexible crossed-trained operators can improve the flexibility of CM in dealing with an unstable demand and can reduce load imbalance inherent in machine dedication in manufacturing cells.     

利用代数重建技术根据火焰辐射图像测量煤粉火焰断面温度场

本分析了基于面阵CCD进行火焰断面温度场测量的测试原理,提出了测试系统的非线性物理模型。并对其作了适当的简化以得到线性模型,借鉴CT中的代数重建技术（ART）来进行求解。同时,考虑到测量对象的一些特征,还设计了一个模糊判断器以“筛选”出合适的解。为了验证所提出的测试模型和解法,进行了数值模拟计算和在油煤混烧实验台进行了测试。最后给出了在某电站锅炉上实际测量的结果。     

基于模糊神经网络的烟度检测系统的研究

本对全流消光式烟度检测系统中的不确定因素进行校正，归结为模糊规则，并用模糊神经网络来提取这种规则，实现了烟度检测系统的误差校正。仿真分析表明，模糊神经网络应用于烟度的检测是可行的。     

用小波神经网络检测结构损伤

用小波和神经网络ART2相结合的方法检测结构的损伤位置。给出了小波变换和人工神经网络的基本理论及其用于损伤检测的原理与特点。通过把小波变换作为神经网络的前处理来构造小波神经网络。首先通过数值试验检验了小波消噪和小波神经网络损伤检测的能力。然后在一个框架结构模型上进行了试验。实验证明这种方法使网络抗噪声能力增强,使损伤识别的效果更好。ART2网络具有自动从环境中学习的能力,能自动的给出新的识别输出。     

云制造——面向服务的网络化制造新模式

针对云制造服务平台上的海量制造服务信息所带来的信息过载问题,提出一种基于图神经网络的云制造服务推荐方法,有效克服了传统推荐方法无法利用数据高维特征的局限性。提取平台上制造服务资源的特征,根据不同的相似度计算方法将制造服务资源构建为网络图;利用邻居采样图神经网络 (graph sample and aggregate, GraphSAGE) 进行网络的表示学习,并将学习到的网络特征带入链接预测函数进行模型训练;通过对资源节点间的链接概率进行预测,完成对用户的制造服务推荐。结果表明,基于图神经网络算法的链接预测模型,其预测性能要优于所对比的共同邻居 (common neighbors, CN) 、Adamic-adar (AA) 与资源分配 (resource allocation, RA) 链接预测算法,从而取得较好的推荐效果,为解决云制造服务推荐问题提供理论依据,有助于提高用户的决策效率。

     

一种改进的模糊神经网络控制器学习算法

提出了一种改进的模糊神经网络混合学习算法,运用遗传算法优化构成隶属函数的网络结构,运用最小二乘法进行解模糊,具有更高的学习精度和更快的收敛速度,解决了在多变量系统中采用模糊神经网络时学习收敛慢且易陷入局部极小点的问题。     

蜂窝纸板切纸机的模糊神经网络智能化控制

提出并实现了一种基于模糊技术和神经网络集成的仿人智能控制模型,利用该智能控制模型对蜂窝纸板切纸机进行控制,可以实现对纸板的按任意要求的切割。仿真实验证明了该模型的可靠性和快速性。     

基于多传感器融合技术的汽车防盗系统研究

汽车防盗系统本身具有复杂性和多变性等特点，一般的汽车防盗装置大多只使用单一的传感器，单个传感器不能为防盗的决策控制提供准确的依据。选择多种传感器建立的汽车防盗系统，运用多传感器融合技术，采用模糊神经网络技术进行信息融合，对汽车被盗信息进行多方面地监测，从而获得较为可靠的信息，为准确地判断汽车的状态提供依据，具有较强的决策控制能力，达到了准确预警的目的。     

Prediction of the mechanical properties of the post-forged Ti–6Al–4V alloy using fuzzy neural network

Isothermal compression of the Ti–6Al–4V alloy was conducted at a 2500 ton isothermal hydrostatic press, and the mechanical properties including ultimate tensile strength, yield strength, elongation and area reduction of the post-forged Ti–6Al–4V alloy were measured at a ZWICK/Z150 testing machine. A fuzzy neural network (FNN) was applied to acquire the relationships between the mechanical properties and the processing parameters of post-forged Ti–6Al–4V alloy. In establishing those relationships, the forging temperature, strain and strain rate were taken as the inputs, whilst the ultimate tensile strength, yield strength, elongation and area reduction were taken as the output respectively. The predicted results using the present FNN model is in a good agreement with the experimental data of the post-forged Ti–6Al–4V alloy, and the optimum processing parameters can be quickly and conveniently selected to achieve the desired mechanical properties by means of the prediction based on the fuzzy neural network model.     

A comparative study on indirect determination of degree of weathering of granites from some physical and strength parameters by two soft computing techniques

Weathering has several adverse effects on the physical, mechanical and deformation characteristics of rock. However, when determining the weathering degree of rocks, some difficulties are encountered. Ideally, the weathering degree can be determined by simple test results and reliable prediction models. Considering this situation, the purpose of the present study is to construct simple and low cost weathering degree prediction models with two soft computing techniques, artificial neural networks and fuzzy inference systems. When developing these models, model results were tested against data from specimens collected from the Harsit granitoid (NE Turkey) and data published in the literature. Model inputs are porosity, P-wave velocity and uniaxial compressive strength, and model output is weathering degree. The models developed in this study exhibited high prediction performances when checked by train and test data sets. This result shows that the models developed herein can be used for indirect determination of weathering degree. The artificial neural network model requests numerical data as the input, while the fuzzy inference system model can take numerical data and expert opinion as the input. As a conclusion, the models have a high potential when determining weathering degree of a rock for various purposes.     

Machine cell formation for production management in cellular manufacturing systems

The formation of machine-part families is an important task in the design of cellular manufacturing systems. Manufacturing cell grouping has the effect of reducing material handing cost and work in process. Among the many methods utilized in machine cells formation, the similarity coefficient method is most widely used. Production sequence and product volumes, if incorporated properly in determining the machine cells, can enhance the quality of solutions and reduce the number of intercellular movements. Measures for cell formation based on operations sequence utilizing ordinal production data are few and have many limitations, such as counting the number of the trips for each individual part instead of counting the weights of the batches. A new ordinal production data similarity coefficient based on the sequence of operations and the batch size of the parts is introduced. Furthermore, a new clustering algorithm for machine cell formation is proposed. The new similarity measure showed more sensitivity to the intercellular movements and the clustering algorithm showed better machine grouping.     

面向模块化制造单元的零件分组算法应用

零件分组是模块化制造单元形成的基础,零件特征对分组影响很大.在模糊聚类算法中引入零件特征权重,对零件分组算法进行改进.通过改变同一组零件特征的权重,在不同的权重下得到了不同的零件分组结果,证明了权重对零件分组有很大的影响.实例表明改进的模糊聚类算法能够使零件分组取得更好的效果.     

应用自适应模糊神经网络进行摩擦补偿的研究

在分析了摩探性及其对伺服系统低速特性的影响的基础上,提出了一种基于自适应模糊神经网络的摩探辨识和补偿方法,该方法既可实现摩探力矩的在线辨识和补偿,又能保证系统的稳定性。仿真结果表明,该方法是有效的,而且便于工程应用。     

大型装备制造企业期量标准智能生成系统研究与实践

大型装备制造企业产品规模大、结构复杂,通常按订单生产,产品变型设计频繁,造成产品期量标准的制定非常复杂,准确性差,大大影响了ERP的实施效果。针对该问题,提出了期量标准的智能化解决方案。利用BP神经网络及其变形网络"识别"历史数据中最相似的"零件模型",对新型零件的提前期进行"预测"。在此基础上提出了详细设计方案,开发出了相应的计算机系统,运用BP神经网络结合梯度下降法对变型零件的期量标准进行估算。     

Complete and fractional cell formation using Kohonen self-organizing map networks in a cellular manufacturing system

The primary objective of group technology (GT) is to enhance the productivity in batch manufacturing environment. The GT cell formation and fractional cell formation are done by using Kohonen self-organizing map (KSOM) networks. The effectiveness of the cell formation is measured with number of exceptional elements, bottleneck parts and grouping efficiency and the effectiveness of the fractional cell formation is measured by number of exceptional elements and the number of machines in the reminder cell. This method is applied to the known benchmarked problems found in the literature and it is found to be equal or best when compared to the other algorithms in terms of minimizing the number of the exceptional elements. The relative merits of using this method with respect to other known algorithms/heuristics in terms of computational speed and consistency are presented.