基于DBN的液压泵劣化程度评估方法研究

Research on Evaluation Method of Hydraulic Pump Degradation Degree Based on DBN

针对轴向柱塞泵中心弹簧失效故障难以有效评估的问题,提出一种基于梅尔频率倒谱系数(MFCC)和深度信念神经网络(DBN)的液压泵劣化程度评估方法。对现场采集的正常数据和3种不同程度中心弹簧失效故障的液压泵振动信号进行信号预处理,包括预加重、分帧和加窗等;对预处理后的信号进行快速傅里叶变换(FFT),得到其频率谱和功率谱,然后让其通过Mel滤波器组,得到信号的对数能量;最后对对数能量进行离散余弦变换,得到信号的倒谱系数和一阶差分系数,并以此构成特征向量。基于DBN方法搭建深度学习模型,对特征向量进行学习,将测试样本导入深度学习模型,对中心弹簧失效程度进行评估,并将倒谱系数和一阶差分系数的识别结果进行对比。结果表明:当选择倒谱系数为特征向量时,具有较高的识别精度,能够有效识别轴向柱塞泵中心弹簧的性能劣化程度。

Aiming at the problem that the central spring failure of axial piston pump is difficult to be evaluated effectively,a hydraulic pump deterioration degree evaluation method based on Mel frequency cepstrum coefficient(MFCC)and deep belief network(DBN)was proposed.The normal data and three different degrees of center spring failure data collected in the field were subjected to signal pre-processing,including pre-emphasis,framing and windowing.Fast Fourier transform(FFT)was applied to the pre-processed signal,and their frequency spectrum and power spectrum were obtained.Then they were passed through the Mel filter bank to get the logarithmic energy of the signal.Finally,discrete cosine transform was performed for logarithmic energy,the Mel frequency cepstrum coefficients and the first-order difference coefficients of the signal were obtained,forming the feature vector.A deep learning model was built based on the DBN method to learn the feature vectors,and the test samples were imported into the deep learning model to evaluate the central spring failure degree.The recognition results of the Mel frequency cepstrum coefficients and the first-order difference coefficients were compared.The results show that when the Mel frequency cepstrum coefficients are selected as the feature vectors,they have a high recognition accuracy and can effectively identify the degree of performance deterioration of the center spring of the axial piston pump.