旋转机械故障多特征匹配(分类)方法

Multi-feature matching (classification) method for rotating machinery fault

从时-频谱中高效提取出具有代表性的参数化特征,对于构建自动故障分类系统来说十分重要.但是,由于时间和频率分辨率的相互制约,加之交叉项干扰、外来噪声的影响,要实现参数化时-频特征的快速有效提取相当困难.为此,提出一种基于同步压缩小波变换时-频特征图像匹配的故障诊断新方法.首先,对多故障模式的测试样本进行适当的预处理,如去野值、去基频等;随后,利用同步压缩小波变换进行精细的时-频分析,得到具有更高时-频分辨率的时-频谱;最后,时-频谱图经过特征区间增强与图像融合预处理后生成时-频特征图像模板,利用图像匹配技术对待诊断的故障模式进行分类识别.结果表明:该方法可以快速识别复杂的故障模式,进而构建新颖高效的自动故障分类系统.

Time-frequency representation(TFR)was a powerful tool for nonstationary data analysis.Efficient parametric feature extraction from the TFR spectra was very important for constructing automatic fault classification systems.However,it was very difficult to effectively extract parametric time-frequency feature because of the limitation between time and frequency resolutions,and the interference by TFR cross-terms and the influence by additive noises.A novel approach was proposed for fault diagnosis based on matching time-frequency characteristic images from synchrosqueezing wavelet transform(SWT).Firstly,the testing samples with multiple failure patterns were preprocessed by some techniques such as outlier reduction and fundamental frequency removal;Secondly,the SWT was used for precise time-frequency analysis on them,and the TFR spectra with higher time-frequency resolution was acquired;Finally,the SWT-based TFR spectra was further preprocessed by using characteristic region enhancement and image fusion to produce the characteristic template of TFR images,and then were used for multiple fault pattern recognitions based on the technique of image matching.The experimental results showed that the proposed approach could recognize complex patterns with high efficiency,and could be used to construct a novel and fast automatic fault classification system.