基于ANN的涡轮盘等温成形过程模拟电路模型

为提高锻件质量和成品率,有必要建立一种适合于实时控制的锻件成形过程模型.利用有限元模拟技术对涡轮盘的等温成形过程进行了虚拟正交试验,通过对成形过程的载荷--行程曲线的分析,建立了粉末高温合金涡轮盘件等温成形过程的人工神经网络(ANN)模型,并将其映射成模拟电路模型.以此模拟电路模型为参考模型,应用于模型参考自适应控制(MRAC)系统,对涡轮盘件等温成形过程进行控制.结果表明,所建立的ANN模型及其模拟电路模型对粉末高温合金涡轮盘件等温成形过程的拟合精度很高,且控制参数始终与模型输出相吻合,为实现盘件成形过程的实时控制奠定了基础.     

Modelling and prediction of cut shape for plasma arc cutting based on artificial neural network

Abstract

An artificial neural network approach for the modelling of plasma arc cutting processes is introduced. Neural network models have been proposed for predicting the cut shape and estimating the special cutting variables. The implementation of artificial neural networks in the modelling of cutting processes is discussed in detail. The performance of the neural networks in modelling is presented and evaluated using actual cutting data. Moreover, prediction applications of the above neural network models are described for various cutting conditions. It is shown that estimated results based on the proposed models agree well with experimental data; the neural network models yield good prediction results over the entire range of cutting process parameters spanned by the training data. The testing and prediction results show the effectiveness and satisfactory prediction accuracy of the artificial neural network modelling. The developed models are applicable to carbon steel.     

The inverse of material properties of functionally graded pipes using the dispersion of guided waves and an artificial neural network

Using guided circumferential wave dispersion characteristics, an inverse method based on artificial neural network (ANN) is presented to determine the material properties of functionally graded material (FGM) pipes. The group velocities of lowest modes at six lower frequencies are used as the inputs of the ANN model. The distribution function of the volume fraction of the FGM pipe is fitted to a polynomial, then the outputs of the ANN are the coefficients of the fitting polynomial. The Legendre polynomial method is employed as the forward solver to calculate the dispersion curves for the FGM pipe. Levenberg–Marquardt algorithm is used as numerical optimization to speed up the training process of the ANN model.     

基于NARX神经网络的磁流变阻尼器模型研究

为了能够较精确地拟合磁流变阻尼器在低速区的非线性及滞回特性,提出一种基于NARX神经网络的磁流变阻尼器动力学模型。以力学性能实验数据为前提,利用LMBP算法训练NARX神经网络的串-并行结构,然后将串-并行结构转化为并行结构,使得NARX神经网络模型能够在不同电流值下得到与实验结果相吻合的输出值。最后将实验结果和模型结果进行对比,通过比较两者所得的结果发现NARX神经网络模型的计算结果与实验结果的最大相对误差仅为3.77%,而且该模型能够表征磁流变阻尼器在低速区的非线性及滞回特性,证明NARX神经网络模型在处理磁流变阻尼器的非线性及滞回特性力学行为具有较高的拟合精度。     

复合材料棒材半固态挤压工艺参数的神经网络预测方法

针对半固态挤压复合材料棒材成形时工艺参数难于选取、试验工作量大的问题 ,采用人工神经网络技术与试验相结合的方法 ,通过对样本处理、神经网络模型参数及收敛性等进行分析 ,建立了工艺参数ANN预测模型 ,可以对复合材料半固态挤压成形的关键工艺参数进行预测 ,预测值与试验值吻合较好 ,最大误差不超过 0 72 % ,为复合材料半固态挤压成形的应用开辟了有效途径     

Artificial neural networks and acoustic emission applied to stability analysis in gas metal arc welding

Abstract

The present paper describes the application of neural networks to obtain a model for estimating the stability of gas metal arc welding (GMAW) process. A neural network has been developed to obtain and model the relationships between the acoustic emission (AE) signal parameters and the stability of GMAW process. Statistical and temporal parameters of AE signals have been used as input of the neural networks; a multilayer feedforward neural network has been used, trained with back propagation method, and using Levenberg Marquardt's algorithm for different network architectures. Different welding conditions have been studied to analyse the incidence of the parameters of the process in acoustic signals. The AE signals have been processed by using the wavelet transform, and have been characterised statistically. Experimental results are provided to illustrate the proposed approach. Finally a statistical analysis for the validation of the experimental results obtained is presented. As a main result of the study, the effectiveness of the application of the artificial neural networks for modelling stability analysis in welding processes has been demonstrated. The regression analysis demonstrates the validity of neural networks to predict the stability of welding process using the statistical characterisation of the signal parameters of AE that have been calculated.     

Ti-17合金本构关系的人工神经网络模型

开发了一个基于神经网络的Ｔｉ１７ 合金的本构关系模型。首先利用ＴｈｅｒｍｅｃｍａｓｔｏｒＺ 型热模拟机等温压缩Ｔｉ１７ 合金, 研究在不同变形温度、变形程度和应变速率等工艺参数条件下流动应力的变化情况。然后用实验所得的热变形工艺参数与性能间的数据训练人工神经网络。训练结束后的神经网络变成为一个知识基的本构关系模型。利用该模型预测的流动应力的值与实验结果间的误差较小。     

基于BP神经网络的TC11钛合金工艺-性能模型预测

材料工艺与性能的关系具有复杂、非线性交互等特点。本文根据TC11钛合金力学性能与其影响因素之间的映射关系,以大量的试验数据为基础,建立了BP神经网络模型。模型的输入包括锻造温度、锻后冷却方式等热加工工艺参数;输出为常用的力学性能指标,即抗拉强度、屈服强度、延伸率和断面收缩率。运用该模型对TC11钛合金力学性能进行了预测,并通过试验数据对模型的预测精度进行了可靠性验证。同时,运用已建立的神经网络模型对TC11钛合金工艺参数与力学性能的关系进行了分析。结果表明,所建立的力学性能预测模型具有良好的外推能力,并且可以很好地反映出该合金的工艺-性能之间的复杂关系。     

A neural network model for the numerical prediction of the diameter of electro-spun polyethylene oxide nanofibers

This paper investigates the viability of neural network as a tool for predicting the diameter of fiber formed by an electrospinning process. Published experimental data for polyethylene oxide (PEO) aqueous solution is used to train and test the neural network model. Concentration, conductivity, flow rate, and electric field strength are used as the input variables to the neural network model. Network model selection, training and testing were conducted using the k-fold cross validation technique which is demonstrated to be the most suitable scheme for the size of dataset used in this study. A statistical study was conducted to establish 95% confidence intervals on the bias and on the limits of agreement between the experimental data and the predicted data. The computer simulation results show a very good agreement between the data, demonstrating the viability of neural network as a promising tool for predicting fiber diameter. While the proposed neural network approach is not intended to model the complete complex physics of the electrospinning process, it is demonstrated to provide an accurate nonlinear mapping between the four salient input variables and the diameter of the formed fiber. This study provides some potential insights into exploring neural network model-based feedback control techniques to regulate nanofiber diameter in an electrospinning process.     

神经网络应用于离子渗氮性能预报与工艺优化

以离子渗氮为例,建立了渗层性能预报神经网络模型;通过误差逆传播（BP）网络实现了了逆映射,建立了工艺参数设计的神经网络模块。试验结果验证了性能预报神经网络模型与工艺优化神经网络模型的可靠性,为解决离子渗氮性能预报与工艺优化设计问题提供了一条先进、合理的途径。     

Comparison of multiple regression and back propagation neural network approaches in modelling top bead height of multipass gas metal arc welds

Abstract

With the advanced developments and automation of the welding process, the use of process optimisation techniques has increased. The objective of the present paper is to describe process optimisation techniques for the gas metal arc (GMA) welding process, based on experimental results generated by the process. Back propagation (BP) neural network and multiple regression methods are employed to study relationships between process parameters and top bead height for robotic multipass welding process, and to select a suitable model that provides the weld final configuration and properties as output and employs the process parameters as input. The process parameters, namely pass number, arc current, welding voltage and welding speed are optimised to produce the required top bead height. These techniques have achieved good agreement with the experimental data and yielded satisfactory results. Also, the BP neural network that was developed was compared to the empirical equations for predicting top bead height through additional experiments, and it was evident that the BP neural network was considerably more accurate than multiple regression techniques.     

Neural network solution of the inverse vibration problem

The feasibility of using general regression neural networks (GRNN) to solve the inverse vibration problem of cracked structures was investigated. The case study used in the investigation was a steel cantilever beam with a single edge crack. The first six natural frequencies were used as network inputs, and crack size and crack location were the outputs. The effect of the number of frequency inputs to the network on prediction accuracy was quantitatively evaluated. The results show that GRNN is a powerful instrument for predicting crack size and location over a wide range, and that the prediction accuracy increases with larger number of vibration modes.     

An approach of artificial neural network to robotic welding process modelling

Ａｎａｐｐｒｏａｃｈｏｆａｒｔｉｆｉｃｉａｌｎｅｕｒａｌｎｅｔｗｏｒｋｔｏｒｏｂｏｔｉｃｗｅｌｄｉｎｇｐｒｏｃｅｓｓｍｏｄｅｌｌｉｎｇ￥ＬｉＹａｎ（ＨａｒｂｉｎＲｅｓｅａｒｃｈＩｎｓｔｉｔｕｔｅｏｆＷｅｌｄｉｎｇＪｏｈｎＮｏｒｒｉｓｈａｎｄＴ．Ｅ．Ｂ...     

基于人工神经网络的金属土壤腐蚀预测方法

将神经网络用于金属土壤腐蚀研究，利用神经网络的学习特征和高度的非线性特征，以土壤理化性能，腐蚀时间，Ａ３钢在土腐蚀试验１，２，８个月的腐蚀数据作为网络训练样本，对土壤中埋片２４个月的Ａ３钢腐蚀速率进行预测，并对结果进行了分析。     

Analysis of hard-facing appearance of specific powdered superalloys for PTA-coating processes

Artificial neural network (ANN) modeling and multiple linear regression (MLR) analysis have been used to develop a powder hard-facing process using high-energy plasma-transferred (HEPT) heating. HEPT heating can produce coatings with minimal surface roughness. An optimal procedure was developed involving the least number of process parameters but producing the most desirable performance characteristic. The quality characteristic of interest is the surface roughness after HEPT processing, utilizing the “the-smaller-the-better” criterion. Process performance was evaluated with respect to the signal-to-noise ratios, which were obtainable through experiments. The experimental results conclude that ANN models demonstrate a greater accuracy of predicting the surface appearance than the MLR models in terms of prediction error and the coefficient of determination. The results also reveal the most significant process control parameters. The predicted value of powder hard-facing roughness, through the implementation of optimal settings, produces a satisfactory result. The confirmation experiment showed that the ANN method achieved the expected optimal design goals for the HEPT powder hard-facing, thereby justifying the reliability and feasibility of the approach.     

Adaptive compensation of quasi-static errors for an intrinsic machine

An adaptive compensation strategy for quasi-static error correction in intrinsic machines is proposed and tested. The proposed methodology consists of systematic modelling of the machine forward kinematics, including quasi-static errors, as well as direct modelling of the inverse kinematics using nonlinear regression analysis. The result is a model which is a hybrid of physical modelling and regression analysis modelling. In addition, the methodology includes a compensation strategy of the machine contouring errors using the state observer technique for on-line adaptive compensation. A CMM is chosen as a test bed for validation of the proposed methodology. Systematic modelling is carried out in two stages for the forward and inverse kinematics. Regression based models are verified using two different tests. The statistical analysis of variance technique (ANOVA) is used to select the best model in addition to model testing using an independent set equal to approximately 10% of the fitting data. The obtained models are then employed in two compensation strategies; one for the measurement error correction, and another one for the contouring error correction by motion command modification in the forward control path. For contouring tests, the CMM behavior at different thermal states is estimated using experimentally obtained Effective Coefficient of Thermal Expansion (ECTE). Simulations of the machine in contouring selected trajectories are carried out over a range of thermal states. Results obtained show an improvement in the CMM performance to a level close to the machine resolution. The CMM performance is tested using the standard ASME B.89.1.12M-1990 evaluation test, as well as a novel modified version of the test accounting for a thermally varying environment. Machine errors are significantly reduced using the proposed methodology.     

减小点焊质量神经网络监测模型误差的措施

点焊过程监测信息与质量参数之间关系也含有较大的非线性,用线性模型去描述这样的关系将导致模型误差的增加。为了更好地描述点焊过程监测信息与质量参数之间的复杂关系,文中将神经网络理论用于点焊过程模型化。在建立点焊质量神经网络监测模型的过程中,发现,训练过程中的“假饱和”现象是减小网络模型误差的主要障碍。为此,分析了各种减小网络模型误差的可能途径,提出了相应的改善措施,并通过试验证明,所提出的观点是正确的、措施是行之有效的。     

冷拔内直筋管凸筋拉缩的研究

根据凸筋拉缩变形的力学条件,研究了工艺因素对凸筋拉缩的影响规律。采用多元非线性曲线拟合的方法,建立了凸筋拉缩的数学模型,可用于凸筋拉缩量的预测及管坯规格的选择。     

Void closure prediction in cold rolling using finite element analysis and neural network

Cold rolling is used to eliminate void defects in cast materials thus improving the material performance during service. A comprehensive procedure is developed using finite element analysis and neural network to predict the degree of void closure. A three-dimensional nonlinear dynamic finite element model was used to study the mechanism of void deformation. Experiments were conducted to investigate void closure during the cold flat rolling process. Experimental results are compared to the three-dimensional finite element predictions to validate the model. The void reduction predictions from finite element analysis are in good agreement with experimental findings. Plastic strain, principal stress distribution around the void and void reduction ratio are presented for various case studies. As finite element simulation is time-consuming, a back-propagation neural network model is also developed to predict void closure behavior. Based on the correlation analysis, the reduction in sheet thickness, the dimension of the void and the size of the rollers were selected as the inputs for the neural network. The neural network model was trained based on results obtained from finite element analysis for various simulation cases. The trained neural network model provides an accurate and efficient procedure to predict void closure behavior in cold rolling.