| Abstract: | As the core part of reciprocating compressor,piston rod is easy to cause a serious accident when abrasion and breakage fault occur to it. Therefore,it is very important to monitor its running state. At present,a small number of reciprocating compressors have been installed on-line monitoring and diagnosis system,most of which can only monitor a single vertical subsidence of piston rod and it can't fully represent the running state of piston rod. Therefore,a method of monitoring the vertical and horizontal displacement of piston rod axis orbit is simultaneously used. In view of the characteristics that the piston rod axis orbit is disordered and difficult to extract features,purification of the axis orbit is carried out based on harmonic wavelet and then features are extracted such as vibration energy,natural frequency and the axis orbit envelope area. After that,a nonlinear local tangent space manifold learning algorithm is used to reduce the dimension of the features and obtain sensitive features. By analyzing the practical cases,the effectiveness of the method for fault monitoring and diagnosis of reciprocating compressor piston rod assembly has been verified. Finally,as BP neural network has the characteristics of solving complex nonlinear problems,the validity of the fault diagnosis method of reciprocating compressor piston rod based on harmonic wavelet and manifold learning is proved by actual case data analysis based on BP neural network. |
