Study On Feature Extraction Methods For Non-stationary Signals And Apply To Hypercompressor Fault Diagnosis

Mechinery diagnosis is a branch of mechanology. Its essence is the pattern recognition of machine running condition. Its key issues are the feature extraction and classification for fault signals. The dissertation is carried out based on the project of "Research on Local Wave Method and its Engineering Application" (supported by Chinese National Nature Science Foundation, No. 50475155) and the project of research on fault diagnosis of reciprocating compressors. The working states of the secondary ethylene hypercompressor in site are studied and condition monitoring and fault diagnosis methods for reciprocating compressorss are implemented by using vibration signals. Since different components have the different failure modes and so do the characteristics of vibration signals, various modern signal processing methods, such as Local-wave method, multifractals theory, improved WPT denoising method and so on are applied to extract fault features from vibration signals with high accuracy and reliability. Based on the extracted features, the methods of intelligence diagnosis for reciprocating compressor are also studied. The validity of all these methods are tested through the secondary ethylene hypercompressors in the Low Density Polyethylene production plant. The main content is listed as follows:1. Operation and structure characteristics of reciprocating compressors are illustrated by using the C-2 secondary ethylene hypercompressor as an example. Disable modes and fault mechanism of the main components are summarized. Then vibration signals are used in reciprocating compressor condition monitoring and the location of measurement points on each cylinder is presented. The characteristics of vibration signals from all measurement points on time and frequency domain are analyzed and some statistic results are obtained. The basis of reciprocating compressor fault diagnosis by using the vibration signals is founded.2. A new approach that extracts fault features from the reciprocating compressor valves is presented after Local-wave method is deeply studied, combined with higher order statistics theory. The vibration signals from combined suction/discharge valves of the secondary ethylene hypercompressor are analyzed. According to its more AM-FMs model for vibration signals, Local-wave method is used to demodulate the vibration signal due to its multi-resolution and self-adaptation in decomposing signal adap. The instantaneous amplitude of characteristic frequency band is normalized and the higher order statistics parameters are calculated from them. The experiment results show that these parameters reflect the...