摘要
特征参数的选择与提取是潜油电泵偏磨诊断至关重要的一步。对潜油电泵偏磨和碰磨过程作了力学分析,证明了利用加速度信号实现偏磨诊断的可行性。据小波变换和功率谱参数的特点以及主成分分析的优点,提出潜油电泵偏磨诊断的特征参数选择和提取方法。对加速度信号使用‘bior1.5’小波先作3层小波分解,然后逐层求取3层细节系数和第3层近似系数的功率谱系数,最后对这4维功率谱系数求取1维主成分,得到了4个代表一个样本的特征参数。该参数消除了小波分解时造成的相邻尺度的相关性,并保留了样本的本质信息和主要信息。以支持向量机作为分类器,5次交叉验证平均正确识别率高达91%,高于小波系数的功率谱系数最大值、小波系数主成分等参数。
Feature parameter selection and extraction is the crucial step to diagnose electric submersible pump （ESP） partial friction. In this paper, the mechanical analysis of the ESP partial friction and collision friction is carried out, which proves the feasibility of diagnosing ESP partial friction with acceleration signal. According to the characteris- tics of wavelet transform and power spectrum parameters and the advantage of principal component analysis （PCA） , a feature parameter selection and extraction method of ESP partial friction is proposed. Firstly, the acceleration signal is decomposed to 3 layers with ＇ biorl. 5＇ wavelet. Then, the power spectrum coefficients of 3 layer detailed co- efficients and the 3rd layer approximation coefficients are calculated for each layer. Lastly, 1 dimension principal components of the 4 dimension power spectrum coefficients are calculated, all the 4 parameters are employed as the features of the sample. These parameters eliminate the correlation in adjacent scales after wavelet decomposition, and retain the essential characteristics and main information of the sample. Taking support vector machine （SVM） as the classifier, the average recognition rate of 5 cross validations is as high as 91% . The results show that these parameters give better performance than the power spectrum maximum values of wavelet coefficients, principal components of wavelet coefficients and other parameters.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2012年第7期1527-1532,共6页
Chinese Journal of Scientific Instrument
基金
"十二.五"国家科技重大专项(2011ZX05024-002-009)资助项目
关键词
加速度
小波分解
功率谱
支持向量机
acceleration
wavelet decomposition
power spectrum
support vector machine （SVM）
