表面技术
錶麵技術
표면기술
Surface Technology
2015年
9期
1-6
,共6页
杨振凯%王海军%刘明%王晶晨
楊振凱%王海軍%劉明%王晶晨
양진개%왕해군%류명%왕정신
超音速等离子喷涂%Fe基合金粉%BP神经网络%非线性拟合%输出预测%参数优化
超音速等離子噴塗%Fe基閤金粉%BP神經網絡%非線性擬閤%輸齣預測%參數優化
초음속등리자분도%Fe기합금분%BP신경망락%비선성의합%수출예측%삼수우화
supersonic plasma spray%Fe base alloy powder%BP neural network%nonlinear fitting%output prediction%parame-ters optimization
目的：基于BP神经网络具有自学习、自训练和输出预测的功能,将其应用于热喷涂过程中的参数优化问题。方法依托高效能超音速等离子喷涂系统实验平台,以Fe基合金粉末为喷涂材料,将等离子喷涂中的主气流量、电功率和喷涂距离作为模型输入,涂层沉积速率和硬度作为模型输出,不断调整隐含层节点个数,最终建立3-7-2网络结构的BP神经网络以优化工艺参数。利用优化出的工艺参数制备Fe基合金涂层,测试其性能,并计算误差。结果神经网络优化出的最优喷涂工艺参数为：主气流量96 L/min,电功率56 kW,喷涂距离95 mm。采用该工艺参数制备涂层,涂层增厚实测平均值为360μm,硬度为672 HV0.3,而模型的预测值分别为332μm和611 HV0.3,与预测值的相对误差分别为7.8%和9.1%。结论 BP神经网络对等离子喷涂参数优化问题的拟合精度比较高,误差在可以接受的范围之内。将BP神经网络运用于热喷涂工艺参数的优化具有科学性和可操作性。
目的：基于BP神經網絡具有自學習、自訓練和輸齣預測的功能,將其應用于熱噴塗過程中的參數優化問題。方法依託高效能超音速等離子噴塗繫統實驗平檯,以Fe基閤金粉末為噴塗材料,將等離子噴塗中的主氣流量、電功率和噴塗距離作為模型輸入,塗層沉積速率和硬度作為模型輸齣,不斷調整隱含層節點箇數,最終建立3-7-2網絡結構的BP神經網絡以優化工藝參數。利用優化齣的工藝參數製備Fe基閤金塗層,測試其性能,併計算誤差。結果神經網絡優化齣的最優噴塗工藝參數為：主氣流量96 L/min,電功率56 kW,噴塗距離95 mm。採用該工藝參數製備塗層,塗層增厚實測平均值為360μm,硬度為672 HV0.3,而模型的預測值分彆為332μm和611 HV0.3,與預測值的相對誤差分彆為7.8%和9.1%。結論 BP神經網絡對等離子噴塗參數優化問題的擬閤精度比較高,誤差在可以接受的範圍之內。將BP神經網絡運用于熱噴塗工藝參數的優化具有科學性和可操作性。
목적：기우BP신경망락구유자학습、자훈련화수출예측적공능,장기응용우열분도과정중적삼수우화문제。방법의탁고효능초음속등리자분도계통실험평태,이Fe기합금분말위분도재료,장등리자분도중적주기류량、전공솔화분도거리작위모형수입,도층침적속솔화경도작위모형수출,불단조정은함층절점개수,최종건립3-7-2망락결구적BP신경망락이우화공예삼수。이용우화출적공예삼수제비Fe기합금도층,측시기성능,병계산오차。결과신경망락우화출적최우분도공예삼수위：주기류량96 L/min,전공솔56 kW,분도거리95 mm。채용해공예삼수제비도층,도층증후실측평균치위360μm,경도위672 HV0.3,이모형적예측치분별위332μm화611 HV0.3,여예측치적상대오차분별위7.8%화9.1%。결론 BP신경망락대등리자분도삼수우화문제적의합정도비교고,오차재가이접수적범위지내。장BP신경망락운용우열분도공예삼수적우화구유과학성화가조작성。
Objective BP neural network has the capability of self-learning, self training and output prediction, which could be a powerful tool to research the parameter optimization problem in thermal spraying process. Methods Relying on the high-efficiency supersonic plasma spray system ( HEPJet) platform, using Fe-based alloy powder as the spraying material, the flow rate of main gas, spraying power and distance were set as the inputs of the model, while the coating deposition rate and hardness were set as model outputs. Through continuous adjustment of the number of hidden layer nodes, the BP neural network with a 3-7-2 network structure was eventually built to optimize the process parameters. The optimized parameters were then used to obtain the Fe-based alloy coating, test its performance and calculate the error. Results The optimized parameters according to the neural network opti-mized were:main gas flow 96 L/min, electric power 56 kW, spraying distance 95 mm. After the experiment, the coating hardness and deposition rate of coating were measured. Its average increment of coating thickness was 360μm, and the average increment of coating hardnessis was 672HV0. 3, while the model predicted values were 332 μm and 611HV0. 3, respectively. Comparing with the predicted values, the errors were 7. 8% and 9. 1%, respectively. Conclusion According to the results of simulation and experi-ment, the accuracy of the BP neural network for the optimization of plasma spray parameters was relatively high, and the error was ac-ceptable. It is scientific and reliable to use BP neural network to deal with the problems of thermal spraying parameters optimization.
