哪位大侠帮我翻译下

lxj311

Artificial neural networks modeling of mechanical property and microstructure evolution  in  the  Tempcore  process
Abstract
In this study, the microstructures and the mechanical properties of steel bars treated by the Tempcore process have been investigated. In the Tempcore  process, AISI 1020 steel bars of various diameters were used. In bars, unlike the self-tempering temperature and the extent of elongation, an increase in the amount of martensite was observed, which caused a consequential increase in yield and tensile strength as a function of quenching duration. The amounts of martensite ,  bainite,  pearlite and the values of elongation, self-tempering temperature, yield and tensile strength could be obtained by a new and fast method, by using artificial neural networks. A PASCAL computer program has been developed for this study.
In the numerical method, bar diameters and quenching durations were chosen as variable parameters. The numerical results obtained via the neural networks were compared with the experimental results .It appears that the agreement is reasonably good.ó2002 Elsevier Science Ltd. All rights reserved.
Keywords: Reinforcing steel; Artificial neural networks; Tempcore process; Quenching; Tempering;  Martensite
1.Introduction
In recent years, the production of high yield strength concrete-reinforcing bars has become increasingly important. High yield strength provides economical use of products.
The name Tempcore is used to define a new process for producing high yield strength concrete-reinforcing steel bars[1–3].This process consists of(i)water quenching of rolled product,(ii)cooling in air and(iii) holding in cooling bed. The martensite layer formed by quenching is reheated by the hot core and reaches a maximum self-tempering temperature after some time. Thus, the martensite layer becomes tempered, as the name Tempcore implies. Quenching duration and self-tempering temperature, which is related to the bar diameter, affect the mechanical behaviour of the bar . Either ferrite,  pearlite,  and bainite or ferrite and pearlite internal structures form below the martensite  layer.
Tempcore process has been used for several years in II_zmir Demir Celik Sanayi A.S.(_IIzmir Iron and Steel Industry Co., Turkey)for steel reinforcing bars with the chemical composition of 0.16–0.21%C,0.17–0.22%Si,
and 0.74–0.83%Mn.Experimental results revealed a linear increase in strength with increasing quenching time, while self-tempering temperature decreases linearly[4].
In the Tempcore process, expensive and prolonged experiments are required to determine the quality of  produced materials depending on varying parameters such as quenching time and bar diameter. However, using the mathematical methods, all these difficulties can be overcome easily. It was shown by Cetinel et al. that microstructures and self-tempering temperatures of bars with different diameters for different quenching durations in the Tempcore process could be determined by the finite element method more quickly than by experimental studies [4].  Therefore, in this study, a neural network method has been chosen to simulate the Tempcore  process. Experimental findings were used to perform the networks.
The calculated amount of martensite, bainite, pearlite expressed as volume percentage, mechanical property and self-tempering temperature values compare reason-ably with the experimental results. Thus, it becomes possible to predict the required quenching duration for different bar diameters and mechanical properties with-out making a change in the production line. Consequently, the mechanical properties can be predicted and controlled by optimising the quenching duration, and the production line can be predesigned as required.

2. Experimental study
Steel bars of AISI 1020 with the chemical composition of 0.17%C, 0.22%Si,0.79%Mn,0.036%P,0.041% S of different diameters(16,18,20 and 22 mm)were used as test materials. Specimens that had been taken from bars produced by the Tempcore process were cut into 12 mm long pieces, prepared for metallographic examination and etched with 3%nital solution. Mechanical properties were determined using an Instron tensile test machine. Tempering temperature was measured by a pyrometer that was placed at the cooling box. The volume percentages of   martensite ,  bainite and pearlite  phases were obtained from etched specimens.Fig.1.shows the structures of steel bars of 18 mm diameter after different quenching durations. Table 1 shows the mechanical properties, self-tempering temperatures and volume percentages of martensite, bainite and pearlite phases obtained for 18 mm diameter steel bar in an average of more than five experiments for each quenching duration studied.

3. Artificial neural networks
Nowadays, engineers and scientists are trying to develop intelligent machines. Artificial neural systems are present-day examples of such machines that have great potential to further improve the quality of our life [5].
It is well known that people and animals are much better and faster at recognising images than most advanced computers. Although computers outperform both biological and artificial neural systems for tasks based on precise and fast arithmetic operations, artificial neural systems represent the promising new generation of information processing networks. Advances have been made in applying such systems for problems found intractable or difficult for traditional computation. Neural networks can supplement the enormous processing power of digital computers with the ability to make sensible decisions and to learn by ordinary experience, as humans do.
Network computation is performed by a dense mesh of computing nodes and connections. They operate collectively and simultaneously on most or all data and inputs. The basic processing elements of neural networks are called artificial neurons, or simply neurons. Often they are simply called nodes. Neurons perform as summing and non-linear mapping junctions. In some cases, they can be considered as threshold units that fire when their total input exceeds certain bias levels. Neurons usually operate in parallel and are configured in regular architectures. They are often organised in layers, and feedback connections both within the layer and toward adjacent layers are allowed. Connection strength is ex-pressed by a numerical value called weight, which can be modified [6].
Among the artificial neural networks, the elementary multilayer perceptrons (MLP) with sigmoidal transfer function have been successfully applied to solve some difficult and diverse problems[7,8]as non-linear discriminant function classifiers. The feedforward network learns from the input data by the supervision of the output data creating single linear discriminant functions by each sigmoid hidden unit and combines them.  Thus , this piecewise linear discriminant function works as a non-linear discriminator.
Training the network in a supervised manner with a highly popular algorithm known as the error back-propagation (BP)has become very popular. BP is an optimisation technique for implementing gradient descent in weight space for multilayer      feedforward   networks.
The basic idea of the technique is to efficiently compute partial derivatives of an approximating function F (w;x) realised  by the network with respect to all the elements of the adjustable weight vector w for a given value of input vector x and output vector y. The weights are adjusted to fit linear piecewise discriminant functions to feature space for the best class separability  . The difference between the network output and the supervisor output is minimised according to a predefined error function (performance criterion) such as mean square error (MSE)etc.
In this work, the neural network system has been applied with multilayer perceptron  (Fig.2) , and BP algorithm by supervised training. A computer program, which was written by the authors in PASCAL, has been used for this application.
Measuring variables such as the volume percentages of martensite, bainite, pearlite, and the values of elongation, yield and tensile strength, the self-tempering temperatures produced as a function of diameters of steel bars and quenching durations have been used for training operations.
3.1. Training of internal structure, self-tempering temperature and mechanical property data
The general aim in the training process is to teach the relations between input and output values to the program and get the results with the possible lowest errors. The input variables are the diameters of steel bars and quenching durations. The output variables are the volume percentages of martensite , bainite , pearlite and the values of elongation, self-tempering temperature yield and tensile strength. Therefore, there are two input variables and seven output variables in this application.
In neural network applications, output values are reduced to values between 0 and 1, which is called the normalization process. This was carried out by dividing the input and output values by some integer numbers.
Two different structures were studied in this work. At first, seven separate programs were used for each output (internal structures, self-tempering temperatures and mechanical properties), so each of these programs has two input nodes and a unique output node.
Secondly, another program which has two input nodes and seven output nodes was used. Experimental values were set as input and output values consisting of 31 rows in each program(some of the input and output values were kept for use in the testing process after the training was complete).The training process is always performed by ‘trial and error method ’and there is no automatic way for that when using artificial neural networks. Training iterations are made by changing the learning rates (a),momentum values(e)and node numbers of hidden layers. The best architecture is obtained in this way to  minimize  the errors. The optimum a,e and hidden layer node numbers and the final errors after 20,000 training iterations are shown in Table 2.
By running the neural network programs, the average percent errors, and the differences between the given output values and the values after training iterations can be determined. The change in the amount of martensite is shown in Fig.3 for 16 mm diameter steel bar (due to the quenching duration).In Fig.4, the change of percent elongation for a round bar of 22 mm diameter can also be seen. In these two figures, the top and the bottom lines show the maximum and minimum experimental values respectively obtained from the Tempcore process. As seen, the training results obtained are generally within the experimental limits. This means that the neural network program works fine and the values provided are reasonably good.
The errors are minimised with more iterations in neural network programs by using the appropriate a,e and hidden layer nodes. For example, as seen in Fig.5 and also in Table 2, the lowest error has been obtained with the hidden layer node number of 4 for the neural network program for self-tempering temperatures. The program for martensite can also be a good example to show the decrease of the error due to the iteration
number. As seen in Fig.6, percent error has already been reduced to reasonable values after 20,000 iterations.
Training iterations take time because of the many trials undertaken by changing the parameters of pro-grams (a,e and hidden layer node number).Training is finished when the optimum parameters are determined Advanced PCs are needed to reduce the training time. A Pentium 200 PC has been used for these programs.

3.2. Testing
The final and most important step in this work of neural networks is to test the programs designed. The programs were tested using different input and output values that were not given for training previously. The test results were compared with the output values that were experimentally obtained and kept for testing. The testing results were compared with the experimental values as seen in Tables 3–6.In Table 3, the programs were tested for 18 mm diameter bars with 1.20 s quenching duration. The first row shows the average experimental values, the second row shows the testing results of separate programs, and the last row shows the testing results of the combined program (the one with seven outputs).In the same way, the programs were tested for 20 mm bars with 1.30 s, and for 22 mm bars with 1.50 and 1.65 s quenching durations and the results are presented in Tables 4–6 respectively.
The testing processes give exact results for many values. Most of the other results are also good enough for the internal structures and mechanical properties. The results can also be compared depending on the programs used. The combined program with seven outputs has given better results for most values, while the seven separate programs have given good results for others.

4. Conclusion
In this work, determination of the volume percent-ages of internal structures, mechanical properties and self-tempering temperature values has been achieved by using artificial neural networks. The experimental results for various bar diameters were used for training the neural network programs, and these programs were tested by the different inputs that were not used for training. The testing results were found to be reasonably good.
The calculated volume percentages of phases, self-tempering temperatures, and the values of other properties were found to be highly satisfactory in comparison with the experimental results.  Thus , it becomes possible to predict the quenching durations for required microstructure and mechanical property values for different bar diameters without making a change in the production line. Consequently  , appropriate quenching duration for a designed quality of material can be predicted and the production line can be adjusted without any experiments.
It can finally be concluded that the neural networks are able to predict internal structures, mechanical properties and self-tempering temperatures of steel bars of various diameters at different quenching durations. Calculations done by the testing process of the neural network programs took only milliseconds. Therefore , it can be reasonably assumed that the neural network programs provide a quick means of calculations conducted in this work.

References
[1]Economopoulos M, Respen Y, Lessel G, Steffes G. Application of the  Tempcore  process to the fabrication of high yield strength concrete-reinforcing bars. CRM Report 1975;45:1–17.
[2]Simon P, Economopoulos M, Nilles P. Tempcore: a new process for the production of high quality reinforcing bars. Iron Steel Eng 1984; 3:55–7.
[3]Centre de Recherches Meetallurgiques (C.R.M.). Tempcore. Lieege: Abbaye du Val-Benoit.
[4] Cetinel H, Toparl? M, OO€zsoyeller L.A finite element based prediction of the microstructural evolution of steels subjected to the Tempcore process. Mech Mater 2000; 32(6):339–47.
[5]Simon H. Neural networks. New Jersey: Macmillan; 1994.
[6]Ripley BD. Pattern recognition and neural network.
Cambridge: Cambridge University Press; 1996.
[7]Guu€rgen F, OO€nal E, Varol FG. IUGR detection by ultrasonographic examinations using neural networks. IEEE Eng
Med Biol Mag 1997; 16(3):55–8.
[8]Karhk B, OO€zkaya E, Aydin S, Pakdemirli M. Vibration of a
beam-mass system using artificial neural networks. Comput Struct 1998; 69(3):339–47.

[ 本帖最后由 lxj311 于 2008-5-2 14:40 编辑 ]

admin

我是用软件翻译的，可能不正确，请参考！

人工神经网络模型的力学性能和显微结构的演化，在tempcore进程
摘要
在这项研究中，微观结构和力学性能的钢筋治疗由tempcore过程中已进行调查。在tempcore过程中，艾西1020钢筋各种直径使用。在酒吧，不同的自我回火温度和程度，伸长率，交易数额增加马氏体的观察，造成了相应的增加，产量和拉伸强度作为一个功能的淬火时间。数额马氏体，贝氏体，珠光体和价值观的伸长率，自回火温度，产量和拉伸强度，可以得到一个新的和快速的方法，用人工神经网络。 1帕斯卡的计算机程序已经开发这项研究。
在数值计算方法，酒吧，直径和淬火时间被选为可变参数。数值计算的结果获得通过神经网络的比较实验结果。看来，这一协议是合理的good.ó2002 Elsevier公司科学有限公司保留所有权利。
关键词：钢筋;人工神经网络; tempcore进程;淬火;回火;马氏体
1.introduction
在近年来，生产高屈服强度混凝土钢筋已变得日益重要。高屈服强度的规定，节约利用的产品。
名称tempcore是用来定义一个新的进程，为生产高屈服强度混凝土钢筋酒吧[ 1-3 ] ，这个过程构成（一）水淬轧制产品， （二）冷却空气中和（三）控股在冷床。马氏体层所形成的淬火加热是由热芯和达到最大限度的自我回火温度后的一段时间。因此，马氏体层成为锻炼，正如其名tempcore意味。淬火的时间长短和自我回火温度，这是关系到钢筋直径，影响力学性能的大律师。无论是铁素体，珠光体，贝氏体或铁素体和珠光体的内部结构下面的表单马氏体层。
tempcore过程中已使用数年，在ii_zmir demir celik sanayi （ _iizmir钢铁工业有限公司，土耳其）为钢筋与化学成分的0.16-0.21 ％ c ,0.17 - 0 0.22 ％ ，硅，
和0.74-0.83 ％ mn.experimental结果显示，线性增加，强度随淬火时间，而自回火温度跌幅线性[ 4 ] 。
在tempcore过程中，昂贵的和长期的实验，以确定质量的生产材料，视乎不同的参数，如淬火时间和钢筋直径。不过，使用数学方法，所有这些困难是可以克服的容易。结果表明，由切蒂内尔等人。说的微观结构和自我回火温度的酒吧不同直径不同的淬火时间，在tempcore过程中可确定的有限元方法更快速，比由实验研究[ 4 ] 。因此，在这项研究中，神经网络方法已被选为模拟tempcore进程。实验结果被用来执行网络。
计算的数额马氏体，贝氏体，珠光体表示，由于货量的百分比，力学性能和自回火温度值比较理性地与实验结果。因此，它成为可能预测所需的淬火时间为不同的酒吧，直径和力学性能与淘汰作出了变化，在生产线。因此，力学性能，可以预见和控制的优化淬火的期限，及生产线，可预视需要。

2 。实验研究
钢筋的艾西1020年与化学成分的0.17 ％ ， c ， 0.22 ％ ，硅，锰0.79 ％ ， 0.036 ％ ，磷， 0.041 ％ ， S的不同直径（ 16,18,20和22毫米）被用来作为测试材料。标本已采取的从酒吧所产生的tempcore过程中被削减到12毫米长的乐曲，准备金相检验和蚀刻与3 ％ nital的解决办法。力学性能，确定使用斯特朗拉伸试验机。回火温度测量由高温计，这是放置在冷却中。量的百分比马氏体，贝氏体和珠光体阶段，获得了从蚀刻specimens.fig.1.shows结构的钢筋18毫米直径后，不同的淬火时间。表1显示的力学性能，自回火温度和体积的百分比马氏体，贝氏体和珠光体阶段取得了18毫米直径钢筋在平均超过5实验为每个淬火的时间研究。

3 。人工神经网络
现在，工程师和科学家们正试图发展智能机器。人工神经系统是目前天的例子，这样的机器有很大的潜力，以进一步改善我们的生活质素[ 5 ] 。
这是人所共知的人民和动物得多，更好和更快地在辨别图像比最先进的电脑。虽然电脑超出生物和人工神经系统的任务，基于精确和快速算术运算，人工神经系统的代表有前途的新一代信息处理网络。取得了进展，在应用这种系统中发现的问题，棘手或难以为传统的计算。神经网络可以补充的巨大处理能力的数码电脑的能力，作出明智的决定，并了解由普通的经验，为人类做的。
网络计算是由一个密集的网格计算节点和连接。他们经营的集体，并同时对大部分或全部的数据和投入。基本处理单元的神经网络被称为人工神经元，或干脆神经元。他们往往只是所谓的节点。神经元执行的总结和非线性映射的路口。在某些情况下，他们可以被视为门槛的单位，消防时，他们的总投入超过了某些偏见的水平。神经元通常是平行运作和配置在定期架构。他们往往在举办的层次，和反馈连接都层和对邻近层是不允许的。连接强度是当然的压力数值所谓的重量，可以进行修改， [ 6 ] 。
其中人工神经网络，初等多层感知器（的MLP ）与sigmoidal传递函数已成功地应用于解决一些锑

3.2 。测试
最后和最重要的一步在这方面的工作，神经网络是测试设计的程式。节目进行了测试使用不同的输入和输出的价值观被认为是不给予培训以前。测试结果相比，输出值进行了实验，并得到了不断进行测试。测试结果进行了比较与实验值可以看出，在表3 - 6.in表3 ，程序进行了测试为18毫米直径的酒吧与1.20 s淬火的时间。第一行显示的平均实验值，第二行显示的测试结果分开程序，最后一列显示的测试结果合并计划（ 1与7产出） 。在同样的方式，程序测试20毫米的酒吧与1.30 s ，并为二十二毫米酒吧与1.50和1.65 s淬火持续时间和结果载列于表4月6日分别。
测试过程中给予确切的结果，对于许多的价值观。其他大部分的结果也不够好，为的内部结构和力学性能。结果也可以比较依赖于程序使用。合并计划，与7产出已作出更好的结果，大部分的价值观，而七个独立的程序已作出了良好的效果为他人。

4 。结论
在这项工作中，测定音量％ -岁的内部结构，力学性能和自我回火温度值已经取得了用人工神经网络。实验结果为各种酒吧，直径被用于训练神经网络的程序，这些程序进行了测试，由不同的投入被认为是不用于培训。测试结果发现相当不错。
计算量的百分比阶段，自回火温度，和价值观的其他属性被认为非常满意，在比较与实验结果。因此，它成为可能预测淬火时间为所需的显微组织和力学性能的价值观不同的酒吧，直径没有改变，在生产线。因此，适当的淬火时间，设计质量的材料可以预见，及生产线可作调整，没有任何实验。
它终于可以得出这样的结论：神经网络是能够预测的内部结构，力学性能和自我回火温度钢筋直径各在不同的淬火时间。计算所做的测试过程中的神经网络程序，仅用了毫秒。因此，可以合理地假定该神经网络的程序提供了快速的方式进行计算，在这方面的工作。

参考
[ 1 ]伊科诺莫普洛斯米， respen y ， lessel克， steffes g.的适用范围tempcore过程中，以制造高屈服强度混凝土钢筋。 CRM的报告， 1975年; 45:1-17 。
[ 2 ]西蒙磷，伊科诺莫普洛斯米， nilles页tempcore ：新工艺生产高品质的钢筋。钢铁摘要： 1984年; 3:55-7 。
[ 3 ]中心recherches meetallurgiques （ c.r.m. ） 。 tempcore 。 lieege ： abbaye杜val -贝努瓦。
[ 4 ]切蒂内尔h后， toparl ？男，面向对象€ zsoyeller香格里拉的有限元基于预测的显微结构的演变钢受到该tempcore进程。机械马特2000年32 （ 6 ） :339 - 47 。
[ 5 ]西蒙每小时神经网络。新泽西网队：麦克米伦; 1994年。
[ 6 ] ripley屋宇署。模式识别和神经网络。
剑桥：剑桥大学出版社， 1996年。
[ 7 ] guu € rgen男，面向对象€宇空实验室英，法varol纤维蛋白原。胎儿宫内发育迟缓的超声检测考试用神经网络。的IEEE工程
医学生物学评论1997年第16 （ 3 ） :55 - 8 。
[ 8 ] karhk b ，面向对象€ zkaya英，艾登s ，帕克代米尔利米振动1
束质量系统利用人工神经网络。 1998年计算机结构; 69 （ 3 ） :339 - 47 。

lxj311

就是软件翻译的不是很到位  我在修改吧   谢谢了

hldxiesj

就是软件翻译的不是很到位，不过还能看

lxj311

还好   这篇翻译已经解决了   谢谢大家的帮助哦

xyzhlan

什么翻译软件能介绍下吗

baker

同样的也想知道

qinsdau

用google全文翻译就行，不过全文翻译质量肯定是不行，给校对留下的工作量应该做好心理准备，呵呵