Reciprocating compressors offer a broad range of capacity control and extremely high compression ratios regardless of gas molecular weight, which are of utmost requirement in the process industries such as hydrogen gas compression and the natural gas transport industry. A drawback of the reciprocating compressor system is that the maintenance costs of reciprocating compressors are approximately 3.5 times greater than those for other types of compressors. Condition-based maintenance has therefore become more prevalent in refineries and petrochemical plants because of the increased criticality of these machines. Achieving higher reliability requires continuous monitoring of the reciprocating compressors.

This invention provides a method to detect and predict valve failures using wavelet analysis, logistic regression, and neural networks. The pressure signal is a non stationary waveform, and so the features from the signal are extracted using wavelet packet decomposition. The extracted features with the temperature data are used to train a logistic regression model to classify defective and normal operation of a valve. The model, for a given set of input, will give the probability of the inputs belonging either to the normal or defective signature group. The logistic regression model is used as an indicator of system health. The wavelet features extracted from the pressure signal are used to train a neural network model to predict their future trends, which are used as indicators for performance assessment and for root cause detection of the compressor valve failures.

File Number: 08UMR004 

Web site: http://ecodevo.mst.edu

Other Information:

Case Manager: Eric Anderson (ericwa@mst.edu)