Research on static calibration of wind tunnel balances based on improved BP neural network
-
摘要: 针对风洞天平静态校准传统校准模型非线性误差较大的问题,采用BP神经网络建立了天平校准模型。三分量天平的BP神经网络模型为典型三层神经网络(“3–7–3”结构);BP神经网络模型校准精准度满足天平静态校准合格指标,轴向力和俯仰力矩分量校准性能优于传统模型,法向力分量校准性能则略低于传统模型。针对BP神经网络存在的不足,采用经混合策略改进的蝴蝶算法优化BP神经网络的初始权值和阈值,优化后的BP神经网络收敛精度和收敛速度得到提高。使用三分量应变天平校准数据进行了仿真实验,以天平输出信号值和天平加载载荷值作为输入和输出构建BP神经网络。传统校准模型、BP神经网络校准模型、蝴蝶算法优化BP神经网络校准模型的仿真实验结果对比表明:使用优化BP神经网络模型拟合天平校准公式,其校准性能比传统校准模型提高70%~90%,可有效消除传统校准模型非线性误差,显著提高天平静态校准精准度。Abstract: Addressing the issue of relatively large nonlinear errors in traditional calibration models for static calibration of the wind tunnel balance, researchers established a balance calibration model using the BP neural network. The BP neural network model for the three-component balance is a typical three-layer neural network, specifically manifested as a “3–7–3” structure. The precision of the BP neural network model meets the qualified criteria for static balance calibration. Its calibration performance in axial force and pitching moment components surpasses that of the traditional model, although it is slightly inferior in the normal force component. To compensate for the deficiencies of the BP neural network, an improved Butterfly Optimization Algorithm with a hybrid strategy is introduced to optimize the initial weights and thresholds. The optimized BP neural network exhibits enhanced convergence accuracy and speed. The present study utilizes the calibration data of the three-component strain gauge balance from simulation experiments, with the balance output signal values and loading load values as inputs and outputs for constructing the BP neural network. A comparison is made among between the simulation results of the traditional calibration model, the BP neural network calibration model, and the Butterfly Optimized Algorithm-BP neural network calibration model. The results indicate that the optimized BP neural network model fitting the balance calibration formula improved calibration performance by 70% – 90% compared to the traditional calibration model. It effectively eliminates the nonlinear errors of the traditional calibration model and significantly improves the precision of static balance calibration.
-
表 1 传统校准模型的校准数据
Table 1. Calibration data of traditional calibration models
指标 X Y MZ RMSE 1.900 g 0.465 g 84.661 g·mm 精准度 0.155% FS 0.034% FS 0.303% FS 表 2 神经网络模型的校准数据
Table 2. Calibration data of neural network models
分量 指标 BP BOA–BP TSBOA–BP X RMSE 0.689 g 0.453 g 0.257 g 准度 0.057% FS 0.038% FS 0.021% FS Y RMSE 0.502 g 0.326 g 0.159 g 准度 0.036% FS 0.023% FS 0.011% FS MZ RMSE 35.217 g·mm 29.405 g·mm 19.962 g·mm 准度 0.126% FS 0.11% FS 0.071% FS -
[1] 贺德馨. 风洞天平[M]. 北京: 国防工业出版社, 2001: 67-68.HE D X. Wind tunnel balance[M]. Beijing: National Defense Industry Press, 2001: 67-68. [2] 战培国. 国外风洞天平技术研究进展[J]. 飞航导弹, 2018(10): 63–68. doi: 10.16338/j.issn.1009-1319.20180123ZHAN P G. Research progress of wind tunnel balance technology abroad[J]. Aerodynamic Missile Journal, 2018(10): 63–68. doi: 10.16338/j.issn.1009-1319.20180123 [3] 战培国. 国外风洞天平校准技术研究进展[J]. 航空科学技术, 2012, 23(2): 18–20. doi: 10.3969/j.issn.1007-5453.2012.02.005ZHAN P G. Development of wind tunnel balance calibration techniques[J]. Aeronautical Science & Technology, 2012, 23(2): 18–20. doi: 10.3969/j.issn.1007-5453.2012.02.005 [4] 武家騊. 应变天平多元校中几种数据处理方法的比较[J]. 气动实验与测量控制, 1994, 8(3): 55–58.WU J T. A comparison of methods of data processing for strain balance multi-component calibration[J]. Aerodynamic Experiment and Measurement & Control, 1994, 8(3): 55–58. [5] 李正鹏, 潘承毅, 邬阳阳. 人工神经网络在预测中的应用现状与展望[J]. 科技创新与生产力, 2023, 44(10): 27–30. doi: 10.3969/j.issn.1674-9146.2023.10.027LI Z P, PAN C Y, WU Y Y. Application status and prospect of artificial neural network in prediction[J]. Sci-Tech Innovation and Productivity, 2023, 44(10): 27–30. doi: 10.3969/j.issn.1674-9146.2023.10.027 [6] 赵传荣, 孔德仁, 王胜强, 等. 一种冲击波压力传感器的准静态校准神经网络模型[J]. 振动与冲击, 2017, 36(13): 92–95, 139. doi: 10.13465/j.cnki.jvs.2017.13.014ZHAO C R, KONG D R, WANG S Q, et al. A neural network model of quasi-static calibration for shock wave pressure sensors[J]. Journal of Vibration and Shock, 2017, 36(13): 92–95, 139. doi: 10.13465/j.cnki.jvs.2017.13.014 [7] 于振, 于万成. 一种基于BP神经网络的扭矩传感器静态校准系统[J]. 传感技术学报, 2020, 33(2): 238–244. doi: 10.3969/j.issn.1004-1699.2020.02.013YU Z, YU W C. A static calibration system of torque sensor based on BP neural network[J]. Chinese Journal of Sensors and Actuators, 2020, 33(2): 238–244. doi: 10.3969/j.issn.1004-1699.2020.02.013 [8] 张海庆. 基于LSTM循环神经网络的矿用甲烷传感器自校准研究[J]. 煤矿机械, 2022, 43(6): 168–171. doi: 10.13436/j.mkjx.202206055ZHANG H Q. Research on self-calibration of mine methane sensor based on LSTM recurrent neural network[J]. Coal Mine Machinery, 2022, 43(6): 168–171. doi: 10.13436/j.mkjx.202206055 [9] 车兵辉, 尹欣繁, 彭先敏, 等. 基于BP神经网络的天平校准数据处理方法研究[J]. 计算机测量与控制, 2020, 28(10): 165–169.CHE B H, YIN X F, PENG X M, et al. Research on data processing method of balance calibration based on BP neural network[J]. Computer Measurement & Control, 2020, 28(10): 165–169. [10] 汪运鹏, 聂少军, 王粤, 等. 卷积神经网络在风洞天平静态校准中的应用[J]. 空气动力学学报, 2023, 41(3): 25–32. doi: 10.7638/kqdlxxb-2022.0096WANG Y P, NIE S J, WANG Y, et al. Application of convolutional neural network in static calibration of wind tunnel balance[J]. Acta Aerodynamica Sinica, 2023, 41(3): 25–32. doi: 10.7638/kqdlxxb-2022.0096 [11] 李炳臻, 刘克, 顾佼佼, 等. 卷积神经网络研究综述[J]. 计算机时代, 2021(4): 8–12, 17. doi: 10.16644/j.cnki.cn33-1094/tp.2021.04.003LI B Z, LIU K, GU J J, et al. Review of the researches on convolutional neural networks[J]. Computer Era, 2021(4): 8–12, 17. doi: 10.16644/j.cnki.cn33-1094/tp.2021.04.003 [12] 余本国. BP神经网络局限性及其改进的研究[J]. 山西农业大学学报(自然科学版), 2009, 29(1): 89–93. doi: 10.3969/j.issn.1671-8151.2009.01.023YU B G. Discussion on the limitation and improvement of BP neural network[J]. Journal of Shanxi Agricultural University (Natural Science Edition), 2009, 29(1): 89–93. doi: 10.3969/j.issn.1671-8151.2009.01.023 [13] 杨洋, 陈家俊. 基于群智能算法优化BP神经网络的应用研究综述[J]. 电脑知识与技术, 2020, 16(35): 7–10, 14.YANG Y, CHEN J J. Review on application of intelligent algorithm to optimize BP neural network[J]. Computer Knowledge and Technology, 2020, 16(35): 7–10, 14. [14] 高岳林, 杨钦文, 王晓峰, 等. 新型群体智能优化算法综述[J]. 郑州大学学报(工学版), 2022, 43(3): 21–30. doi: 10.13705/j.issn.1671-6833.2022.03.007GAO Y L, YANG Q W, WANG X F, et al. Summary of new group intelligent optimization algorithms[J]. Journal of Zhengzhou University (Engineering Science), 2022, 43(3): 21–30. doi: 10.13705/j.issn.1671-6833.2022.03.007 [15] ARORA S, SINGH S. Butterfly optimization algorithm: a novel approach for global optimization[J]. Soft Computing, 2019, 23(3): 715–734. doi: 10.1007/s00500-018-3102-4