王金東，趙　巖，高媛媛，曹　陽，夏法鋒

(1. 東北石油大學(xué) 機(jī)械科學(xué)與工程學(xué)院，黑龍江 大慶 163318；

2. 中國(guó)石油管道大慶輸油氣分公司，黑龍江 大慶 163458)

?

基于BP神經(jīng)網(wǎng)絡(luò)的Ni-Al2O3鍍層粒子復(fù)合量預(yù)測(cè)研究*

王金東1，趙巖1，高媛媛2，曹陽2，夏法鋒1

(1. 東北石油大學(xué) 機(jī)械科學(xué)與工程學(xué)院，黑龍江 大慶 163318；

2. 中國(guó)石油管道大慶輸油氣分公司，黑龍江 大慶 163458)

摘要：采用超聲波輔助電沉積法在A3鋼表面制備了Ni-Al2O3鍍層，通過BP神經(jīng)網(wǎng)絡(luò)對(duì)不同工藝參數(shù)下制備鍍層的Al2O3粒子復(fù)合量進(jìn)行預(yù)測(cè)，最后利用透射電鏡(TEM)觀察鍍層結(jié)構(gòu)組織。結(jié)果表明，該BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)為3×8×1時(shí)，其預(yù)測(cè)值與真實(shí)值的擬合度R=0.9991，相對(duì)誤差的最大值與最小值分別為1.71%與0.74%。TEM分析表明，當(dāng)Al2O3粒子濃度9 g/L，電流密度3 A/dm2，溫度40 ℃時(shí)，Ni-Al2O3鍍層組織較為緊密，其平均粒徑約為20 nm。

關(guān)鍵詞：BP神經(jīng)網(wǎng)絡(luò)；Ni-Al2O3鍍層；粒子復(fù)合量

1引言

超聲波輔助電沉積是一種通過在電沉積過程中施加超聲波場(chǎng)，使鍍液中粒子能夠均勻的沉積于鍍層表面的方法[1-4]。目前，有關(guān)不同工藝參數(shù)對(duì)Ni-Al2O3鍍層性能影響的報(bào)道較多，但基于BP神經(jīng)網(wǎng)絡(luò)對(duì)Ni-Al2O3鍍層粒子復(fù)合量預(yù)測(cè)的研究較少[5-7]。為此，本文通過超聲波輔助電沉積法在A3鋼表面制備Ni-Al2O3鍍層，并采用BP神經(jīng)網(wǎng)絡(luò)對(duì)鍍層粒子復(fù)合量進(jìn)行預(yù)測(cè)研究，最后利用透射電鏡(TEM)觀察鍍層的組織結(jié)構(gòu)。該研究可為Ni-Al2O3鍍層在機(jī)械設(shè)備再制造技術(shù)的應(yīng)用提供一定技術(shù)參考。

2實(shí)驗(yàn)

2.1實(shí)驗(yàn)材料及工藝參數(shù)

采用尺寸20 mm×20 mm×1 mm的A3鋼片作為實(shí)驗(yàn)基材，使用純度大于99%的鎳板作為陽極。實(shí)驗(yàn)所用鍍液為瓦特型鍍鎳液，Ni-Al2O3鍍層制備過程所需試劑及工藝參數(shù)見表1。

2.2實(shí)驗(yàn)過程

采用超聲波輔助電沉積方法在A3鋼表面制備Ni-Al2O3鍍層。其中，脈沖電源是E/PS 3016-10B型脈沖電源，超聲波場(chǎng)由KQ-1500VDE型超聲波清洗器產(chǎn)生，利用XRD-7000型X射線衍射儀(XRD)對(duì)Ni-Al2O3鍍層中Al2O3粒子含量進(jìn)行測(cè)量。最后，通過Tecnai-G2-20型透射電鏡(TEM)觀察不同工藝參數(shù)下制備Ni-Al2O3鍍層組織結(jié)構(gòu)。

表1Ni-Al2O3鍍層的鍍液成分及制備工藝

Table 1 Plating conditions and process for preparing Ni-Al2O3coatings

化學(xué)試劑參數(shù)工藝條件參數(shù)NiSO4·6H2O300g/L超聲波功率180WNiCl2·6H2O40g/L電流密度1～5A/dm2H3BO335g/LAl2O3粒子濃度5～10g/L表面活性劑20mg/LpH值4.5西曲溴銨0.5mg/L溫度20～50℃

2.3粒子復(fù)合量計(jì)算

Ni-Al2O3鍍層粒子復(fù)合量的計(jì)算公式

(1)

式中，W表示鍍層中Al2O3質(zhì)量分?jǐn)?shù)(%)，M1表示Al2O3相對(duì)分子質(zhì)量，M2表示Al相對(duì)原子質(zhì)量，W0表示XRD測(cè)量鍍層中鋁元素含量(%)。

2.4BP神經(jīng)網(wǎng)絡(luò)設(shè)計(jì)及表征

根據(jù)Ni-Al2O3鍍層制備工藝，本文采用3個(gè)分量作為BP神經(jīng)網(wǎng)絡(luò)的輸入層，即Al2O3粒子濃度(x1)、電流密度(x2)和溫度(x3)，采用1個(gè)分量作為輸出層，即Ni-Al2O3鍍層粒子復(fù)合量(y)，該BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)如圖1所示。

圖1　BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)圖

3結(jié)果與分析

3.1BP神經(jīng)網(wǎng)絡(luò)測(cè)試

利用Matlab7.0軟件建立BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)[8-9]，圖2為BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)在該條件下的均方根誤差、隱含層和神經(jīng)元數(shù)量的關(guān)系。由圖可見，當(dāng)BP模型的神經(jīng)元數(shù)為16個(gè)、隱含層數(shù)為8個(gè)時(shí)，BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)的均方根誤差最小，其最小值為1.32%。因此，本文采用BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)為3×8×1，經(jīng)計(jì)算，該結(jié)構(gòu)實(shí)驗(yàn)值與預(yù)測(cè)值的擬合相似度R=0.9991。

圖2BP神經(jīng)網(wǎng)絡(luò)均方根誤差、隱含層與神經(jīng)元數(shù)量之間的關(guān)系

Fig 2 Errors of the BP model obtained at different hidden layers and neuron numbers

3.2BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)

圖3為BP神經(jīng)網(wǎng)絡(luò)模擬Ni-Al2O3鍍層粒子復(fù)合量曲線。由圖可知，利用BP神經(jīng)網(wǎng)絡(luò)對(duì)Ni-Al2O3鍍層1～30#樣本數(shù)據(jù)進(jìn)行測(cè)試，其預(yù)測(cè)值與真實(shí)值變化基本一致，故BP神經(jīng)網(wǎng)絡(luò)能夠較好的模擬Ni-Al2O3鍍層中粒子復(fù)合量變化規(guī)律。因此，本文采用BP神經(jīng)網(wǎng)絡(luò)對(duì)31～40#樣本進(jìn)行預(yù)測(cè)，以此檢驗(yàn)其預(yù)測(cè)效果，其預(yù)測(cè)結(jié)果見表2。從表2中看出，BP神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)的預(yù)測(cè)值與真實(shí)值相差不大，其相對(duì)誤差的最大值與最小值分別為1.71%與0.74%。因此，BP神經(jīng)網(wǎng)絡(luò)能夠較好的模擬Ni-Al2O3鍍層粒子復(fù)合量，并為其它金屬鍍層的性能預(yù)測(cè)提供一種新方法。

圖3　Ni-Al2O3鍍層的BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)結(jié)果

Fig 3 The prediction results of BP neural network of Ni-Al2O3coatings

表2BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)結(jié)果及相對(duì)誤差

Table 2 The predicted results and relative errors by using BP neural network

樣品編號(hào)預(yù)測(cè)值/%實(shí)際值/%相對(duì)誤差/%314.2354.2871.21324.3654.4411.71334.0884.1321.06344.3994.3521.07354.2794.2450.80364.1464.1770.74374.4434.4820.87384.0654.1281.52394.2394.2911.21404.2164.2550.91

3.3工藝參數(shù)對(duì)組織結(jié)構(gòu)的影響

圖4為不同工藝參數(shù)下制備Ni-Al2O3鍍層的TEM照片。由圖4可知，當(dāng)Al2O3粒子濃度9 g/L，電流密度3 A/dm2，溫度40 ℃時(shí)，所制備的Ni-Al2O3鍍層組織較為緊密，Al2O3粒子復(fù)合量較高，其平均粒徑約為20 nm。當(dāng)Al2O3粒子濃度6 g/L，電流密度2 A/dm2，溫度30 ℃時(shí)，所制備的Ni-Al2O3鍍層組織疏松，Al2O3粒子復(fù)合量較低，其平均粒徑約為50 nm。由此可見，在適宜的Al2O3粒子濃度、電流密度及溫度等工藝參數(shù)下，可制得Al2O3粒子復(fù)合量較高的Ni-Al2O3鍍層。

圖4　不同工藝參數(shù)制備Ni-Al2O3鍍層TEM照片

Fig 4 TEM photos of Ni-Al2O3coatings prepared by different parameters

4結(jié)論

采用超聲波輔助電沉積法在A3鋼表面制備了Ni-Al2O3鍍層，并建立了3×8×1的BP神經(jīng)網(wǎng)絡(luò)模型，其輸入層為Al2O3粒子濃度、電流密度和溫度，輸出層為鍍層中Al2O3粒子復(fù)合量。該BP神經(jīng)網(wǎng)絡(luò)的相對(duì)誤差最大值與最小值分別為1.71%與0.74%。TEM分析表明，當(dāng)采用Al2O3粒子濃度9 g/L、電流密度3 A/dm2及溫度50 ℃時(shí)，所制備的Ni-Al2O3鍍層組織較為緊密，Al2O3粒子復(fù)合量較高，其平均粒徑約為20 nm。

參考文獻(xiàn)：

[1]Wu Menghua, Li Zhi, Xia Fafeng, et al. Study on the preparation of nano Ni-Al2O3composite layer by ultrasonic-electrodepositing method [J]. Journal of Functional Materials, 2004, 35(6): 776-778.

吳蒙華, 李智, 夏法鋒, 等. 納米Ni-Al2O3復(fù)合層的超聲電沉積制備[J]. 功能材料, 2004, 35(6): 776-778.

[2]Xia Fafeng, Liu Chao, Wang Fan, et al. Preparation and characterization of Ni-TiN coatings deposited by ultrasonic electrodeposition [J]. Journal of Alloys and Compounds, 2010, 490(1-2): 431-435.

[3]Ma Chunyang, Wu Menghua, Qu Zhijia.Technology of ultrasonic-electroless plating Ni-P-SiC nanocomposite coating [J]. Heat Treatment of Metals, 2011, 36(4): 89-92.

馬春陽, 吳蒙華, 曲智家. 超聲波-化學(xué)鍍Ni-P-SiC納米復(fù)合鍍層的工藝研究[J]. 金屬熱處理, 2011, 36(4): 89-92.

[4]Xia Fafeng, Huang Ming, Ma Chunyang, et al. Effect of electrodeposition methods on corrosion resistance of Ni-SiC nanocomposite coatings [J]. Journal of Functional Materials, 2013, 44(16): 2429-2431.

夏法鋒, 黃明, 馬春陽, 等. 電沉積方式對(duì)Ni-SiC納米鍍層耐腐蝕性能的影響[J]. 功能材料, 2013, 44(16): 2429-2431.

[5]Li Xingyuan, Zhu Yongyong, Xiao Guorong. Application of articial neural networks to predict sliding wear resistance of Ni-TiN nanocomposite coatings deposited by pulse electrodeposition [J]. Ceramics International, 2014, 40(8): 11767-11772.

[6]Xia Fafeng, Jiao Jinlong, Ma Chunyang, et al. Forecast the microhardnesses of the Ni-TiN nanocoatings by AR model [J]. Journal of Functional Materials, 2012, 43(2): 140-143.

夏法鋒, 焦金龍, 馬春陽, 等. 基于AR模型的Ni-TiN納米鍍層顯微硬度預(yù)測(cè)研究[J]. 功能材料, 2012, 43(2): 140-143.

[7]Abdel A. Hard and corrosion resistant nanocomposite coating for Al alloy [J]. Materials Science and Engineering A, 2008, 474: 181-187.

[8]Han Zhiguo, Wang Jiming, Chen Zhigao. Management performance evaluation in petrochemical engineering construction project by using artificial neural network [J].Acta Petrolei Sinica (Petroleum Processing Section), 2010, 26(3): 317-323.

[9]Zhou Huangbin, Zhou Yonghua, Zhu Lijuan. Implementation and comparison of improving BP neural network based on MATLAB [J]. Computing Technology and Automation, 2008, 27(1): 28-31.

周黃斌, 周永華, 朱麗娟. 基于MATLAB的改進(jìn)BP神經(jīng)網(wǎng)絡(luò)的實(shí)現(xiàn)與比較[J]. 計(jì)算技術(shù)與自動(dòng)化, 2008, 27(1): 28-31.

Prediction on the Al2O3contents in Ni-Al2O3coatings by using BP neural network

WANG Jindong1, ZHAO Yan1, GAO Yuanyuan2, CAO Yang2, XIA Fafeng1

(1. School of Mechanical Science and Engineering, Northeast Petroleum University, Daqing 163318, China;2. Daqing Oil and Gas Branch, China Petroleum Pipeline, Daqing 163458, China)

Abstract：Ni-Al2O3 coatings were prepared by ultrasonic-electrodeposition method on the surface of A3 steel, and the particle contents of Ni-Al2O3 coatings were predicted by BP neural network. The microsturctures of Ni-Al2O3 coatings were observed by using TEM. The results indicate that the schematic of the BP model is 3×8×1, and the fitting similarity is 0.9991. The maximal and minimal relatives of this model are 1.71% and 0.74%, respectively. TEM presents that the microstructure of Ni-Al2O3 coatings, which deposited at Al2O3 particle concentration of 9 g/L, current density of 3 A/dm2 and temperature of 40 ℃, has a fine structure and the average particle size is approximately 20 nm.

Key words：BP neural network；Ni-Al2O3 coating；particle content

DOI:10.3969/j.issn.1001-9731.2016.01.048

文獻(xiàn)標(biāo)識(shí)碼：A

中圖分類號(hào)：TG1.43

作者簡(jiǎn)介：王金東(1962-)，男，山東濰坊人，博士，教授，博士生導(dǎo)師，從事石油設(shè)備再制造技術(shù)研究。

基金項(xiàng)目：國(guó)家自然科學(xué)基金資助項(xiàng)目(51474072)；中國(guó)博士后科學(xué)基金資助項(xiàng)目(2015M581425)

文章編號(hào)：1001-9731(2016)01-01226-03

收到初稿日期：2015-04-15 收到修改稿日期：2015-07-20 通訊作者：夏法鋒，E-mail: xiaff@126.com