Neural network prediction of the shunt current in resistance spot welding

An error back propagation （BP） neural network prediction model was established for the shunt current compensation in series resistance spot welding. The input variables for the neural network consist of the resistivity of the material, the thickness of workpiece and the spot spacing, and the shunt rate is outputted. A simplified calculation for the shunt rate was presented based on the feature of the constant-current resistance spot welding and the variation of the resistance in resistance spot welding process, and then the data generated by simplified calculation were used to train and adjust the neural network model. The neural network model proposed was used to predict the shunt rate in the spot welding of 20# mlid steel （in Chinese classification） （in 2. 0 mm thickness） and 10# mild steel （in 1.5 mm and 1.0 mm thickness）. The maximum relative prediction errors are, respectively, 2. 83%, 1.77% and 3.67%. Shunt current compensation experiments were peoCormed based on the neural network prediction model proposed to check the diameter difference of nuggets. Experimental results show that maximum nugget diameter deviation is less than 4% for both 10# and 20# mlid steels with spot spacing of 30 mm and 50 mm.     

Artificial neural network approach for prediction of stress–strain curve of near β titanium alloy

In the present study, artificial neural network（ANN） approach was used to predict the stress-strain curve of near beta titanium alloy as a function of volume fractions of a and b. This approach is to develop the best possible combination or neural network（NN） to predict the stress-strain curve. In order to achieve this, three different NN architectures（feed-forward back-propagation network,cascade-forward back-propagation network, and layer recurrent network）, three different transfer functions（purelin, Log-Sigmoid, and Tan-Sigmoid）, number of hidden layers（1 and 2）, number of neurons in the hidden layer（s）,and different training algorithms were employed. ANN training modules, the load in terms of strain, and volume fraction of a are the inputs and the stress as an output.ANN system was trained using the prepared training set（a,16 % a, 40 % a, and b stress-strain curves）. After training process, test data were used to check system accuracy. It is observed that feed-forward back-propagation network is the fastest, and Log-Sigmoid transfer function is giving the best results. Finally, layer recurrent NN with a single hidden layer consists of 11 neurons, and Log-Sigmoid transfer function using trainlm as training algorithm is giving good result, and average relative error is1.27 ± 1.45 %. In two hidden layers, layer recurrent NN consists of 7 neurons in each hidden layer with trainrp as the training algorithm having the transfer function of LogSigmoid which gives better results. As a result, the NN is founded successful for the prediction of stress-strain curve of near b titanium alloy.     

Artificial neural network approach for prediction of stress–strain curve of near b titanium alloy

In the present study, artificial neural network(ANN) approach was used to predict the stress–strain curve of near beta titanium alloy as a function of volume fractions of a and b. This approach is to develop the best possible combination or neural network(NN) to predict the stress–strain curve. In order to achieve this, three different NN architectures(feed-forward back-propagation network,cascade-forward back-propagation network, and layer recurrent network), three different transfer functions(purelin, Log-Sigmoid, and Tan-Sigmoid), number of hidden layers(1 and 2), number of neurons in the hidden layer(s),and different training algorithms were employed. ANN training modules, the load in terms of strain, and volume fraction of a are the inputs and the stress as an output.ANN system was trained using the prepared training set(a,16 % a, 40 % a, and b stress–strain curves). After training process, test data were used to check system accuracy. It is observed that feed-forward back-propagation network is the fastest, and Log-Sigmoid transfer function is giving the best results. Finally, layer recurrent NN with a single hidden layer consists of 11 neurons, and Log-Sigmoid transfer function using trainlm as training algorithm is giving good result, and average relative error is1.27 ± 1.45 %. In two hidden layers, layer recurrent NN consists of 7 neurons in each hidden layer with trainrp as the training algorithm having the transfer function of LogSigmoid which gives better results. As a result, the NN is founded successful for the prediction of stress–strain curve of near b titanium alloy.     

Evaluation of echo features of ultrasonic flaws and its intelligent pattern recognition

In this paper, three types of weld flaw were taken as target, evaluation and recognition of flaw echo features were studied. On the basis of experimental study and theoretical analysis, 26 features have been extracted from each echo samples.A method which is based on the statistical hypothesis testing and used for feature evaluation and optimum subset selection was explored Thus. the dimensionality reduction of feature space was brought out, and simultaneously, the amount of calculation was decreased. An intelligent pattern classifier with B-P type neural network was constructed which was characterized by high speed and accuracy for learning. Using a half of total samples as training set and others as testing set, the learning efficiency and the classification ability of network model were studied. The results of experiment showed that the learning rate of different training samples was about 100％. The results of recognition was satisfactory when the optimum feature subset was taken as the sample's feat     

Real estate appraisal system based on GIS and BP neural network

For the inefficiency and inaccuracy of appraisal method of traditional estate appraisal theory, the real estate appraisal system based on GIS and BP neural network was established. The structure of the system was designed which includes appraisal model, trade case, GIS database and query analysis module. With the help of the L-M algorithm in MATLAB software, BP neural network was improved and the trade cases were trained, then the BP neural network which has already been trained was tested. At the same time, the BP neural and GIS were put together to construct the hedonic price estimate model. The C# and ArcGIS9.3 were used to achieve the system in VS2008. City basic geographic data and real estate related information were used as the basic data in practice. The results show that the functions of querying, adding and editing the spatial data and attribute data are achieved and also the efficiency and accuracy of real estate are improved, so that the new method of real estate is provided by the system.     

Predicting model on ultimate compressive strength of Al2O3-ZrO2ceramic foam filter based on BP neural network

In present study, BP neural network model was proposed for the prediction of ultimate compressive strength of Al2O3-ZrO2 ceramic foam filter prepared by centrifugal slip casting. The inputs of the BP neural network model were the applied load on the epispastic polystyrene template (F), centrifugal acceleration (v) and sintering temperature (T), while the only output was the ultimate compressive strength (σ). According to the registered BP model, the effects of F, v, T on σ were analyzed. The predicted resul...     

Forecasting of mechanical properties of covered electrode containing La/CeO_2 based on fuzzy neutral network

In order to improve the mechanical properties of deposited metal of ilmenite type welding electrode,CeO_2/La rare earth elements were added into electrodes based on E4301 electrode, then electrodes were produced,test plates were welded, and mechanical properties were tested based on National Standards of China. For the sake of solving the problems of large amount of mechanical properties tests, long test cycle and high test cost during the conventional production process of electrode, a prediction model of the mechanical properties of deposited metal based on Takagi-Sugeno(T-S) fuzzy neural network was established. Mn, Si and C contents of medium manganese in electrode, CeO_2, and La contents of electrode and welding speed were selected as input variables of the prediction model, and the tensile strength, lower yield strength, elongation, impact energy and hardness of deposited metal were selected as output variables. Finally,predicting experiment was done under test samples, and results show that average relative prediction error of the tensile strength, lower yield strength, elongation and hardness are 0.91 %, 2.57 %, 4.94 % and 1.94 %, respectively, which reach the need of actual production. The results of prediction show that the mechanical properties of deposited metal of electrode containing rare earth can be forecasted accurately through material composition of electrode and welding parameters based on T-S fuzzy neural network model.     

Evolutionary artificial neural network approach for predicting properties of Cu- 15Ni-8Sn-0.4Si alloy

A novel data mining approach, based on artificial neural network（ANN） using differential evolution（DE） training algorithm, was proposed to model the non-linear relationship between parameters of aging processes and mechanical and electrical properties of Cu-15Ni-8Sn-0.4Si alloy. In order to improve predictive accuracy of ANN model, the leave-one-out-cross-validation （LOOCV） technique was adopted to automatically determine the optimal number of neurons of the hidden layer. The forecasting performance of the proposed global optimization algorithm was compared with that of local optimization algorithm. The present calculated results are consistent with the experimental values, which suggests that the proposed evolutionary artificial neural network algorithm is feasible and efficient. Moreover, the experimental results illustrate that the DE training algorithm combined with gradient-based training algorithm achieves better convergence performance and the lowest forecasting errors and is therefore considered to be a promising alternative method to forecast the hardness and electrical conductivity of Cu- 15Ni-8Sn-0.4Si alloy.     

Strength Prediction of Aluminum–Stainless Steel-Pulsed TIG Welding–Brazing Joints with RSM and ANN

Pulsed TIG welding–brazing process was applied to join aluminum with stainless steel dissimilar metals. Major parameters that affect the joint property significantly were identified as pulsed peak current, base current, pulse on time,and frequency by pre-experiments. A sample was established according to central composite design. Based on the sample,response surface methodology(RSM) and artificial neural networks(ANN) were employed to predict the tensile strength of the joints separately. With RSM, a significant and rational mathematical model was established to predict the joint strength.With ANN, a modified back-propagation algorithm consisting of one input layer with four neurons, one hidden layer with eight neurons, and one output layer with one neuron was trained for predicting the strength. Compared with RSM, average relative prediction error of ANN was \10% and it obtained more stable and precise results.     

Theoretical and experimental investigation on giant magnetoresistive materials with amorphous ferromagnetic layer

Pseudo-spin-valve （PSV） sandwiches using amorphous CoNbZr alloy as soft magnetic layer were fabricated by magnetron sputtering. The giant magnetoresistance （GMR） and its dependence on the thickness of magnetic layer were investigated. Anti-parallel magnetization alignments were observed in the samples with very thin CoNbZr thickness （2-4 nm） and a maximum GMR ratio of 6.5% was obtained. The Camley-Barnas semiclassical model was extended for amorphous layer based ：nagnetic sandwiches by considering that the mixed layers exist between the ferromagnetic and nonmagnetic layer. The calculated results agree with the experimental results very well, indicating that the new model gives a more realistic picture of the physical processes that take place in the magnetic sandwiches. Moreover, the calculated results for amorphous sandwiches also clarify that the occurrence of maximum GMR at very small thickness of amorphous layer is ascribed to the short mean-flee-path in amorphous materials.     

基于小波分解和BP神经网络的磁巴克豪森噪声信号分层分析研究

因小波变换具有多尺度分析的特点,在时频两域都有表征信号局部特征的能力,因此采用小波分解方法研究了不同时频成分的磁巴克豪森（MBN）信号随温度和应力变化的灵敏度问题。采用db5小波对MBN信号进行6层小波分解,提取各层分解系数的均值和均方根,并讨论分析了各特征值随所加应力以及温度变化的相对变化关系。研究表明,在试样的弹性范围内,低频系数和各层高频系数的均值和均方根都随压应力的增加而减小;各层高频系数的均值和均方根随温度的升高而降低,低频系数的均值和均方根随温度的升高而升高。最后将温度、原始MBN信号以及各分解系数的均值和均方根作为神经网络的输入,压应力作为其输出建立神经网络模型,结果表明该神经网络模型与之前没有用小波分解时的神经网络模型相比,检测应力的准确性更高。     

Prediction of Flow Stress in Cadmium Using Constitutive Equation and Artificial Neural Network Approach

A model is developed to predict the constitutive flow behavior of cadmium during compression test using artificial neural network (ANN). The inputs of the neural network are strain, strain rate, and temperature, whereas flow stress is the output. Experimental data obtained from compression tests in the temperature range ?30 to 70 °C, strain range 0.1 to 0.6, and strain rate range 10^?3 to 1 s^?1 are employed to develop the model. A three-layer feed-forward ANN is trained with Levenberg-Marquardt training algorithm. It has been shown that the developed ANN model can efficiently and accurately predict the deformation behavior of cadmium. This trained network could predict the flow stress better than a constitutive equation of the type $ \dot{\upvarepsilon } = A\sinh (\upalpha /\upsigma )^{n} \exp ( - Q/RT) $ .     

Reaction mechanism of roasting Zn2SiO4 using NaOH

The reaction kinetics of roasting zinc silicate using NaOH was investigated. The orthogonal test was employed to optimize the reaction conditions and the optimized reaction conditions were as follows: molar ratio of NaOH to Zn₂SiO₄ of 16:1, reaction temperature of 550 °C, and reaction time of 2.5 h. In order to ascertain the phases transformation and reaction processes of zinc oxide and silica, the XRD phase analysis was used to analyze the phases of these specimens roasted at different temperatures. The final phases of the specimen roasted at 600 °C were Na₂ZnO₂, Na₄SiO₄, Na₂ZnSiO₄ and NaOH. The reaction kinetic equation of roasting was determined by the shrinking unreacted core model. Aiming to investigate the reaction mechanism, two control models of reaction rate were applied: chemical reaction at the particle surface and diffusion through the product layer. The results indicated that the diffusion through the product layer model described the reaction process well. The apparent activation energy of the roasting was 19.77 kJ/mol.     

基于深度循环神经网络的大规模风力发电功率预测

本文针对目前风力发电功率预测存在超短期、精度差等问题,通过分析大规模风力发电功率特性和风电预测时间序列特性,提出以深度循环神经网络进行预测,结合小波系数多尺度分析的隐马尔可夫预测方法,将深度学习引入到循环神经网络中来,构建基于多尺度隐马尔可夫模型-深度循环神经网络模型的大规模风力发电功率预测模型(MHMM-DRNN)。实例验证相对误差平均值:BP神经网络模型约为31.56%,在预测过程中误差最大,ARMA模型约为23.20%,小波分析约为26.11%,而MHMM-DRNN预测模型约为16.85%,具有较好的实用性。     

Constitutive modeling of compression behavior of TC4 tube based on modified Arrhenius and artificial neural network models

Warm rotary draw bending provides a feasible method to form the large-diameter thin-walled(LDTW)TC4 bent tubes, which are widely used in the pneumatic system of aircrafts. An accurate prediction of flow behavior of TC4 tubes considering the couple effects of temperature,strain rate and strain is critical for understanding the deformation behavior of metals and optimizing the processing parameters in warm rotary draw bending of TC4 tubes. In this study, isothermal compression tests of TC4 tube alloy were performed from 573 to 873 K with an interval of 100 K and strain rates of 0.001, 0.010 and0.100 s~(-1). The prediction of flow behavior was done using two constitutive models, namely modified Arrhenius model and artificial neural network(ANN) model. The predictions of these constitutive models were compared using statistical measures like correlation coefficient(R), average absolute relative error(AARE) and its variation with the deformation parameters(temperature, strain rate and strain). Analysis of statistical measures reveals that the two models show high predicted accuracy in terms of R and AARE. Comparatively speaking, the ANN model presents higher predicted accuracy than the modified Arrhenius model. In addition, the predicted accuracy of ANN model presents high stability at the whole deformation parameter ranges, whereas the predictability of the modified Arrhenius model has some fluctuation at different deformation conditions. It presents higher predicted accuracy at temperatures of 573–773 K, strain rates of 0.010–0.100 s~(-1)and strain of 0.04–0.32, while low accuracy at temperature of 873 K, strain rates of 0.001 s~(-1)and strain of 0.36–0.48.Thus, the application of modified Arrhenius model is limited by its relatively low predicted accuracy at some deformation conditions, while the ANN model presents very high predicted accuracy at all deformation conditions,which can be used to study the compression behavior of TC4 tube at the temperature range of 573–873 K and the strain rate of 0.001–0.100 s~(-1). It can provide guideline for the design of processing parameters in warm rotary draw bending of LDTW TC4 tubes.     

Preparation of sodium manganate from low-grade pyrolusite by alkaline predesilication–fluidized roasting technique

Low concentration alkaline leaching was used for predesilication treatment of low-grade pyrolusite. The effects of initial NaOH concentration, liquid-to-solid ratio, leaching temperature, leaching time and stirring speed on silica leaching rate were investigated and the kinetics of alkaline leaching process was studied. The results show that silica leaching rate reached 91.2% under the conditions of initial NaOH concentration of 20%, liquid-to-solid ratio of 4:1, leaching temperature of 180 °C, leaching time of 4 h and stirring speed of 300 r/min. Shrinking-core model showed that the leaching process was controlled by the chemical surface reaction with activation energy E_a of 53.31 kJ/mol. The fluidized roasting conditions for preparation of sodium manganate were optimized by the orthogonal experiments using the desiliconized residue. The conversion rate of sodium manganate was obtained to be 89.7% under the conditions of silica leaching rate of 91.2%, NaOH/MnO₂ mass ratio of 3:1, roasting temperature of 500 °C and roasting time of 4 h, and it increased with the increase of silicon leaching rate.     

Separation of arsenic and antimony from dust with high content of arsenic by a selective sulfidation roasting process using sulfur

The separation of arsenic and antimony from dust with high content of arsenic was conducted via a selective sulfidation roasting process. The factors such as roasting temperature, roasting time, sulfur content and nitrogen flow rate were investigated using XRD, EPMA and SEM–EDS. In a certain range, the sulfur addition has an active effect on the arsenic volatilization because the solid solution phase ((Sb,As)₂O₃) in the dust can be destroyed after the Sb component in it being vulcanized to Sb₂S₃ and this generated As₂O₃ continues to volatile. In addition, an amorphization reaction between As₂O₃ and Sb₂O₃ is hindered through the sulfidation of Sb₂O₃, which is also beneficial to increasing arsenic volatilization rate. The results show that volatilization rates of arsenic and antimony reach 95.36% and only 9.07%, respectively, under the optimum condition of roasting temperature of 350 °C, roasting time of 90 min, sulfur content of 22% and N₂ flow rate of 70 mL/min. In addition, the antimony in the residues can be reclaimed through a reverberatory process.     

浓硫酸焙烧高钛渣的反应动力学（英文）

提出一种新方法,利用浓硫酸焙烧高钛渣提取二氧化钛,并在焙烧工艺的基础上研究焙烧反应动力学。考察焙烧温度、粒度以及酸矿比对反应速率的影响。结果表明,焙烧反应符合未反应核收缩模型。动力学实验数据、SEM和EDAX结果分析表明,用浓硫酸焙烧高钛矿渣的反应受通过固体产物层的内扩散控制。Arrhenius方程得到焙烧反应的表观活化能为18.94 kJ/mol。     

基于神经网络集成学习的冷连轧板形预测

为了实现对冷连轧带钢出口板形的预测,基于粒子群算法对小波神经网络进行了优化,将优化后的网络作为基学习器,并通过bagging算法构建集成学习预测模型,进行冷连轧带钢板形的预测。以某1 450 mm冷连轧生产线数据作为样本,比较了该模型与未经优化的小波神经网络和单个学习器的预测效果。结果表明,集成学习模型预测的带钢出口板形与实测板形的偏差更小,板形预测精度更高,模型泛化性能更好。     

A comparative study of a back propagation artificial neural network and a Zerilli–Armstrong model for pure molybdenum during hot deformation

In this study, the hot deformation behavior of molybdenum was investigated by means of thermal simulation on a Gleeble-1500 machine. The experiments were carried out under different temperatures, ranging from 1100 to 1400 °C, and with a strain rate of 1S⁻¹ to 50S⁻¹. The flow stress under the above mentioned hot deformation conditions was predicted using a back propagation (BP) artificial neural network. The architecture of the network included three input parameters: strain rate, temperature and true strain, and just one output parameter: the flow stress. One hidden layer was adopted, which include nine neurons. Compared with the prediction method of flow stress using the Zerilli–Armstrong model, the prediction method using the BP artificial neural network had higher accuracy.