基于SAX的往复压缩机气阀故障诊断

摘要

传统的时域、频域、时频域分析方法基于振动信号对整个序列时间点分析,忽略了其中一小段时间序列数据变化规律,没有考虑其波形特征,无法有效提取特征.鉴于此,提出了基于符号聚合近似算法(SAX)的故障特征提取方法.该方法对机械振动信号SAX符号化近似表示,之后通过一定长度滑动窗口确定符号串类型,统计符号串频数,作为特征值,最后排列各个特征值,归一化后得到向量,作为该数据样本的特征向量.该方法针对周期序列中一小段数据进行对比,描述其变化规律,挖掘内在信息,形成数据模态,然后对各类模态统计分析,有效提取了原始序列的变化趋势和信息特征,克服了现有方法的不足.对油田现场压缩机气阀振动数据分析处理,用SAX方法提取特征向量后输入网格优化的SVM分类器进行了训练分类,与基于信息熵的特征提取方法进行对比,两者故障分类准确率分别为81.25％和100％.分析结果表明:SAX特征提取方法可以准确有效地提取故障特征,具有更高的故障识别率,基于SAX算法的往复压缩机气阀故障诊断方法具有可行性.所得结论可为压缩机气阀故障信号的进一步分析提供参考.%The traditional analysis methods based on vibration signal in time,frequency and time-frequency domain focus on the whole time sequence,but ignore the change characteristic and waveform tendency in a part of sequence,thus making it ineffective in feature extraction.A new fault feature extraction method based on SAX algorithm is proposed.This method transforms mechanical vibration signal into SAX symbol representation,and then determines the symbol string type through a sliding window with a certain length.The symbol string frequency is counted as the eigenvalues,and then arranged and normalized to obtain the vector as the feature vector of data sample.The method analyzes the periodic sequence on the basis of a part of sequence,describes the change characteristic and waveform tendency and digs the intrinsic information to form a data modality.Then,the statistical analysis of various modalities is conducted to effectively extract the trend of the original sequence from qualitative and quantitative perspectives and information features,thus overcoming the shortcomings of existing methods.Vibration data of on-site compressor valve in oilfield were analyzed and processed.The feature vector was extracted using SAX method.The grid-optimized SVM classifier is input for training classification and analyzed using feature extraction method based on information entropy.The accuracy of fault classification was 81.25％ and 100％,respectively.The analysis results show that the SAX feature extraction method can extract the fault features accurately and effectively,and has a higher fault recognition rate.The SAX algorithm based fault diagnosis method for gas valve on reciprocating compressor is feasible.The conclusions can provide references for further analysis of compressor signals.