基于改进北方苍鹰算法与混合核极限学习机的齿轮箱故障诊断

Gearbox fault diagnosis based on improved northern goshawk algorithm and hybrid core extreme learning machine

中文关键词  混合核极限学习机  改进北方苍鹰优化算法  时变滤波经验模态分解  故障诊断

英文关键词  hybrid kernel extreme learning machine  improved northern goshawk algorithm  time varying filter based empirical mode decomposition  fault diagnosis

中文摘要

      针对行星齿轮箱故障诊断问题,本文提出了一种基于改进北方苍鹰优化(INGO)算法与混合核极限学习机 (HKELM)的行星齿轮箱故障诊断方法.首先,引入Savitzky-Golay(SG)滤波对齿轮箱原始信号进行去噪.利用时变滤 波经验模态分解(TVF-EMD)将去噪后的信号分解成多个本征模态函数(IMF),使用方差贡献率、相关系数和信息熵 筛选出最优的IMF.将最优IMF重构后,对重构信号进行时间同步平均(TSA)去噪以减少故障诊断模型的数据计算 量. 将Tent混沌映射、混合正弦余弦算法和Levy飞行策略用于改进北方苍鹰优化(NGO)算法,得到一种新的INGO算 法. 同时,引入余弦因子以平衡正弦余弦算法的全局和局部开发能力.最后,利用INGO算法对HKELM进行优化,用 以提高HKELM模型的故障诊断准确率.将所提方法应用于两个案例对模型进行检验,实验结果表明,本文所提方 法具有可行性和优越性.

英文摘要

      Aiming at the problem of planetary gearbox fault diagnosis, this paper proposes a fault diagnosis method of planetary gearbox based on the improved northern northern goshawk algorithm (INGO) and hybrid core extreme learning machine (HKELM). Firstly, savitzky-golay (SG) filtering is introduced to denoise the original signal of the gearbox. In addition, the time varying filtering empirical mode decomposition (TVF-EMD) is used to decompose the denoised signal into multiple intrinsic mode functions (IMF). And the variance contribution rate, correlation coefficient and information entropy are used to screen out the optimal IMFs. After the optimal IMFs are reconstructed, the reconstructed signal is denoised by time synchronization average (TSA) to reduce the data calculation amount of the fault diagnosis model. Secondly, INGO algorithm is obtained by applying Tent chaotic mapping, hybrid sine cosine algorithm and Levy flight strategy to improve the NGO algorithm. At the same time, the cosine factor is introduced to balance the global and local development capabilities of sine cosine algorithm. Finally, the INGO algorithm is used to optimize HKELM to improve the fault diagnosis accuracy of HKELM model. The proposed scheme is applied to two different public data sets to test the model, and the experimental results show that the proposed method is feasible and advantageous.