VP型宽频带倾斜仪故障信号的BBA-SOM智能诊断

Determination of Troubles for VP-type Broadband Inclinometer Using CEEMD Multi-scale Approximate Entropy and BBA-SOM Model

针对现有VP型倾斜仪故障诊断主要依靠人工经验和诊断流程较为复杂的问题,提出以互补集合经验模态分解(complete ensemble empirical mode decomposition,CEEMD)多尺度近似熵和二进制蝙蝠算法(binary bat algorithm,BBA)优化SOM神经网络参数的VP型倾斜仪故障诊断新方。首先,将归一化后的仪器故障信号进行CEEMD分解,对6阶本征模态函数(intrinsic mode function,IMF)求取多尺度近似熵值;然后将网络输入法按比例分为训练集和测试集,以训练集的识别率为适应度函数,应用二进制蝙蝠算法(binary bat algorithm,BBA)优化SOM神经网络的竞争层维数和网络训练次数;最后应用上述得到的BBA-SOM网络模型对倾斜仪故障特征数据进行辨识。实验表明:CEEMD多尺度近似熵判据对倾斜仪故障特征的区分效果符合预期;相对于朴素贝叶斯、AdaBoost集成学习与LDA等学习模型,BBA-SOM模型可以准确进行故障诊断;该方法对实现VP型倾斜仪故障的自动诊断有重要现实意义。

Aiming at the problem that the fault diagnosis of the existing VP-type inclinometer mainly relies on manual experience and the diagnosis process is relatively complex,a complementary ensemble empirical mode decomposition(CEEMD)multi-scale approximate entropy and binary bat algorithm(BBA)optimization were proposed.A new method for fault diagnosis of VP-type inclinometers based on SOM neural network parameters.First,the normalized instrument fault signal was decomposed by CEEMD,and the multi-scale approximate entropy value was obtained for the sixth-order eigenmode function(IMF).Then the network input is divided into training set and test set proportionally to train.The recognition rate of the set is a fitness function,and the binary bat algorithm(BBA)was used to optimize the competition layer dimension and network training times of the SOM neural network.Finally,the BBA-SOM network model obtained above was used to identify the fault characteristic data of the inclinometer.Experiments show that the CEEMD multi-scale approximate entropy criterion can distinguish the fault features of the inclinometer as expected.Compared with the learning models such as Naive Bayes,AdaBoost ensemble learning and LDA,the BBA-SOM model can accurately diagnose faults.The automatic diagnosis of VP-type inclinometer fault has important practical significance.