往复机械示功图识别方法研究

发布时间：2019-05-12 06:29

【摘要】：往复式压缩机在工程实践中被广泛应用,然而由于其结构特点,经常出现各种故障情况,传统的基于振动信号的检测方法无法排除干扰信号的影响,对于实验结果不是特别理想。本文提出了一种基于压力信号的结合不变矩算法的故障识别诊断方法,通过采集压力信号,较好地避免了杂质信号的影响,通过计算压力信号示功图的不变矩,并结合支持向量机,有效地对故障类型进行识别和检测。本文进行的工作主要有:(1)对往复式压缩机的压力信号进行采集,并进行进一步修正和归一化后得到示功图的相关信息。(2)对示功图进行二值化、形态学细化处理,降低图像的冗余信息,作为分析判断的基础和依据。(3)计算示功图的不变矩的特征量,在进行归一化处理后得到的中心距的基础上构造7个不变矩,并进行Gram-Schmidt正交化处理,这样可以使得特征向量的线性相关性的影响被有效消除。(4)采用支持向量机方法对故障特征进行分类识别,该方法具有很好的计算收敛速度,极大地缩短了特征量分类的计算时间,完全适用于分类求解问题。
[Abstract]:Reciprocating compressor is widely used in engineering practice. However, due to its structural characteristics, various faults often occur, so the traditional detection method based on vibration signal can not eliminate the influence of interference signal. The experimental results are not ideal. In this paper, a fault identification and diagnosis method based on pressure signal and invariant moment algorithm is proposed. By collecting pressure signal, the influence of impurity signal is avoided, and the invariant moment of pressure signal indicator diagram is calculated. Combined with support vector machine (SVM), the fault types are identified and detected effectively. The main work of this paper is as follows: (1) the pressure signal of reciprocating compressor is collected, and the relevant information of indicator diagram is obtained after further modification and normalization. (2) the indicator diagram is binarization and morphological refinement. The redundant information of the image is reduced as the basis and basis of analysis and judgment. (3) the characteristic quantity of the invariant moment of the indicator diagram is calculated, and seven invariant moments are constructed on the basis of the center distance obtained after normalization. The orthogonal processing of Gram-Schmidt is carried out, which can effectively eliminate the influence of linear correlation of eigenvector. (4) support vector machine (SVM) method is used to classify and identify fault features, which has a good convergence rate. The computation time of feature classification is greatly shortened, and it is suitable for classification problem.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TH45

【相似文献】