Backpropagation Discrimination Geometric Analysis Interference Memory Modelling Neural Nets

A number of recent simulation studies have shown that when feedforward neural nets are trained, using backpropagation, to memorize sets of items in sequential blocks and without negative exemplars, severe retroactive interference or catastrophic forgetting results. Both formal analysis and simulation studies are employed here to show why and under what circumstances such retroactive interference arises. The conclusion is that, on the one hand, approximations to 'ideal' network geometries can entirely alleviate interference if the training data sets have been generated from a learnable function (not arbitrary pattern associations). All that is required is either a representative training set or enough sequential memory sets. However, this elimination of interference comes with cost of a breakdown in discrimination between input patterns that have been learned and those that have not: catastrophic remembering. On the other hand, localized geometries for subfunctions eliminate the discrimination problem but are easily disrupted by new training sets and thus cause catastrophic interference. The paper concludes with a formally guaranteed solution to the problems of interference and discrimination. This is the Hebbian Autoassociative Recognition Memory (HARM) model which is essentially a neural net implementation of a simple look-up table. Although it requires considerable memory resources, when used as a yardstick with which to evaluate other proposed solutions, it uses the same or less resources.     

机器人位姿精度理论研究现状及展望

本文分析了机器人位姿精度理论的研究现状, 指出其欠缺之处, 从信息论的观点出发, 探索利用信息熵的概念来研究机器人的位姿精度理论, 建立了机器人位姿信息传输模型, 并对它有望弥补解决目前机器人位姿精度理论中研究不足的可能性进行分析探讨。     

离合器盖冲压模具结构CAD系统的设计模式

本文利用结构化分析与设计思想对离合器盖冲压模具结构进行了功能分析，建立了ＩＤＥＦＯ功能模型与ＩＤＥＦ１Ｘ信息模型，并设计了一种灵活实用的表示装配关系的数据结构，建立了离合器盖冲压模具结构ＣＡＤ系统的总体模，在本文中，系统开发的方法论得了体现。     

瞄准摄像机的视觉伺服控制

本文提出了瞄准摄像机的视觉伺服控制方法。控制系统由两个模块组成：图像处理模块和瞄准控制模块。在图像处理模块中，用基于特征的方法分析动态序列图像。在瞄准控制模块中，用前馈型神经网络实现从信息空间到任务控制空间的映照。根据瞄准摄像机的模型和位姿估计算法通过精确计算得到神经网络的训练样本。用计算机仿真来验证控制方法的性能。     

基于神经网络的TC21合金本构关系模型(英文)

本构方程是描述材料变形和有限元模拟基本信息必要的数学模型,它反映流动应力与应变、应变率和温度综合作用的高度非线性关系。基于Gleeble-1500热模拟机上进行等温压缩试验获得的实验数据,系统研究TC21钛合金的流变行为,并采用BP人工神经网络建立该合金的本构关系模型。在该模型中,输入变量为应变、应变速率和变形温度,输出变量为流动应力。与传统方法相比,利用BP人工神经网络所建立的本构关系模型能够更好地表征试验数据及描述整个变形过程。     

A fuzzy logic model for machining data selection

This paper describes the development stages of a fuzzy logic model for metal cutting. The model is based on the assumption that the relationship between the hardness of a given material and the recommended cutting speed is an imprecise relationship, and can be described and evaluated by the theory of fuzzy sets. The model has been applied to data extracted from the Machining Data Handbook [Metcut Research Associates, 3rd edn, Vols 1 and 2 (1980)], and a very good correlation was obtained between the handbook data and that predicted using the fuzzy logic model. The objective of the model is to facilitate the computerization process of the vast machining information contained in machining data handbooks. Also, the proposed model suggests the possibility of developing an expert system for machining data selection based on fuzzy logic.     

Application of inverse techniques to determine heat- transfer coefficients in heat- treating operations

An existing sequential function specification algorithm for the solution of the inverse heat conduction problem (IHCP) has been applied to determine the response of both the surface heat flux and the surface temperature of flat stainless steel samples subjected to water quenching under controlled laboratory conditions that ensured one-dimensional heat flow. From this information, combined convective and radiative heat-transfer coefficients have been obtained as a function of steel surface temperature. The computer code was subsequently modified to solve the IHCP for air-cooled cylindrical carbon steel samples. In the algorithm, the problem is linearized by assuming the thermophysical properties of the steel to be fixed at values from the previous time step while estimating the current surface heat flux, which results in a more efficient code without a severe loss of accuracy. When compared with iterative (“brute force”) methods commonly used in the past, techniques like sequential function specification offer a more robust strategy for solving the IHCP. By including information on future measurements, while solving for the unknown surface heat flux at a particular time, the sequential function specification algorithm effectively prevents over-responses to measured temperatures, and large variations in calculated heat-transfer coefficients, observed when sequential matching is applied, can be reduced. Sensitivity coefficients, a measure of the response of temperature to changes in the unknown surface heat flux which are calculated with this algorithm, can be used to design experiments involving the IHCP effectively.     

模锻过程结合机理与数据的智能控制方法

大型模锻成形过程是一个复杂的非线性时变过程,包括锻件流变成形过程与液压系统驱动过程,以及还存在油液泄漏等众多不确定性因素,导致精准锻造过程控制异常困难。为此,在结合基于机理模型控制与数据控制优点的基础上,提出了基于物理模型结合在线顺序极限学习机的智能控制方法。该方法首先使用已知的系统信息推导出名义控制律;其次,针对模型不确定性部分,使用在线顺序极限学习机设计出该在线模型的补偿控制律;最后,建立了基于机理模型与数据模型的集成控制器,获得了最佳控制律。仿真结果表明,新方法能有效地控制复杂的锻造过程,且比现有的方法有更好的控制精度。     

基于可制造性特征的制造资源建模

可制造性评价是缩短开发周期、优化制造工艺、降低产品成本的有效途径。产品的可制造性取决于特定制造资源的加工能力。制造资源模型的开发是可制造性评价的基础。为了更好地利用制造资源信息，提出一种模糊C均值聚类算法与遗传算法相结合的混合算法，根据制造特征对制造资源进行分组，基于制造资源约束的Object-Oriented方法建立制造资源信息模型，设计可制造性评价框架。通过对32台加工设备进行划分，使用混合算法动态确定最优分组数目和该数目下的最优分组。结果表明：混合算法可靠有效，能够提高应用企业的整体绩效，增强决策的可行性，并有利于管理层做出更明智的决策。     

基于多色集合理论的产品装配序列规划研究

使用多色集合理论为产品装配建立信息模型,给出了装配基准的确定方法,并结合分层结构树图建立了产品分解图和装配图的生成算法。利用该算法可得到产品的可行装配顺序方案集合。最后结合实例给出了产品装配问题的求解思路。     

Sequential coupling simulation for electromagnetic–mechanical tube compression by finite element analysis

A sequential coupled field analysis of electromagnetic tube compression process has been performed by using the finite element method (FEM). A 2D axisymmetric electromagnetic model based on the magnetic vector potential is proposed for the calculation of magnetic field and magnetic forces; and Newmark integration method is used to calculate the dynamic plastic deformation of tube in the mechanical model. In each time step during the simulation, the transient magnetic forces acting on the tube are obtained from the electromagnetic model, which are then used as input load to the mechanical model. According to the tube deformation calculated, the geometry of the electromagnetic model is updated. Therefore, the sequential coupling simulation is realized between the electromagnetic model and the mechanical model, whose results are more accurate than the loosing coupling simulation method. The radial deformation along the outer generatrix of deformed tube wall is presented as a function of time and the plastic strain energy of electromagnetic tube compression is analyzed.     

A Hybrid Feature Recognizer for Machining Process Planning Systems

We describe a hybrid feature recognition method for machining features that integrates three distinct feature recognition methods: graph matching, cell-based maximal volume decomposition, and negative feature decomposition using convex decomposition. Each of these methods has strengths and limitations, which are evaluated separately. We integrate these methods in a sequential workflow, such that each method recognizes features according to its strengths, and successively simplifies the part model for the following methods. We identify two anomalous cases in the application of maximal volume decomposition, and their cure by introducing limiting halfspaces. Feature volumes recognized by all three methods are then combined into a unified hierarchical feature representation, which captures feature interaction information, including geometry-based machining precedence relations.     

Nonlinear comminution process modeling based on GA–FNN in the computational comminution system

A new concept named computational comminution is firstly proposed in this paper. Based on information technology, the structure of a computational comminution system (CCS) is built. The study on CCS is very different from the traditional ones for comminution, such as the study based on theoretic models, or the study based on experimental models. As one of the key technologies in CCS, a modeling framework for comminution processes is implemented particularly by employing GA–FNN that can model complex nonlinear processes such as the comminution process of cement by integrating artificial neural networks, fuzzy sets and genetic algorithms. Application results of this modeling method to the Horomill cement comminution process show that the modeling framework discussed in this paper is efficient.     

Neural networks with a self-refreshing memory: Knowledge transfer in sequential learning tasks without catastrophic forgetting

We explore a dual-network architecture with self-refreshing memory (Ans and Rousset 1997) which overcomes catastrophic forgetting in sequential learning tasks. Its principle is that new knowledge is learned along with an internally generated activity reflecting the network history. What mainly distinguishes this model from others using pseudorehearsal in feedforward multilayer networks is a reverberating process used for generating pseudoitems. This process, which tends to go up to network attractors from random activation, is more suitable for capturing optimally the deep structure of previously learned knowledge than a single feedforward pass of activity. The proposed mechanism for ?transporting memory? without loss of information between two different brain structures could be viewed as a neurobiologically plausible means for consolidation in long-term memory. Knowledge transfer is explored with regard to learning speed, ability to generalize and vulnerability to network damages. We show that transfer is more efficient when two related tasks are sequentially learned than when they are learned concurrently. With a self-refreshing memory network knowledge can be saved for a long time and therefore reused in subsequent acquisitions.     

基于火焰加热的微流体器件制造系统

基于玻璃热成型技术设计一种微流体器件的制备系统,该系统基于火焰加热并通过多组电机的顺序控制可实现不同规格微玻璃管的拉制。并通过软件对加热过程中微玻璃管的温度变化建立仿真分析模型,分析加热过程中微玻璃管的温度分布变化,为微玻璃管的加热时间提供借鉴,并对不同加热时间下微玻璃管的软化区域进行探究。通过改变制备系统的加热时间、拉伸电机的转速、加热区域宽度、微玻璃管的型号可拉制出不同型号尺寸的微针、微管道等微流体器件,并利用系统样机进行了基于加热时间、拉制速度、加热宽度的实验,探究各参数对微流体器件形貌尺寸的影响。相比于传统的制备系统,该系统不仅拥有多样化的可调参数,而且可通过更换不同内径的筒夹来实现对不同尺寸微流体器件的制备,使其拉制范围更广。     

Simulation of sheet metal forming incorporating EBSD data

In this work, a ferritic stainless steel (DC04) is investigated numerically on the micro and the macro scale. Single crystal orientations extracted from 2D electron backscatter diffraction (EBSD) data are reduced by using two methodically different methods. Based on these reduced data sets, a two-scale Taylor type model is applied at the integration points of the finite elements to simulate a deep drawing process. The earing profiles of the simulated cups are compared to the experimental data. In addition, the influence of the variation of the sheet holder force on the earing profile of the drawn cup for different friction coefficients is discussed. It is found that incorporating experimental microstructural information in the two-scale simulations, realistic earing profiles are obtained. With regard to efficient computation, a coarse FE model is used, resulting in a larger average cup height than the experimental slope. Nevertheless, reducing the friction coefficient in the model, the slope of the cup height agrees very well with the experiments which is expected to be obtainable for a finer FE mesh and the experimental friction coefficient as well.     

混沌-支持向量机在加工误差预测中的应用

提出将混沌-支持向量机模型方法应用于加工误差数据预测。利用互信息法和曹氏方法进行相空间重构,并运用小数据量法计算最大Lyapunov指数,对加工误差时间序列进行混沌识别。通过最小二乘支持向量机对历史样本的学习建立预测模型,并将其预测结果与RBF神经网络预测结果进行仿真对比。结果表明,在较少的加工误差数据条件下,该模型能够有效地描述和预测加工误差的变化,具有较高的预测精度。     

自动变速器模拟仿真技术

根据自动变速器各组成部分的功能,将自动变速器分为液力变矩器、行星齿轮传动子系统、液压控制子系统、电子控制子系统4个部分,分别论述了各子系统现有的建模仿真方法.最后建立了典型自动变速器液压控制系统和行星齿轮机构在AMESim环境下的仿真模型,并进行了动态分析.     

个性化产品需求表达模型的研究

随着网络技术、信息技术的日趋成熟,以客户需求为中心的个性化产品的设计与开发已逐渐被研究者所重视.如何系统地、全面地获取客户对产品的需求,是个性化产品设计的关键.文章通过从产品的装配结构和性状特征入手,以产品需求表达结点之间的规则约束、多Agent之间的协商机制,来构建真正以客户为中心的个性化产品的需求表达模型.