The right monitoring and preventative maintenance strategies can mitigate compressor failure and shutdown. In this study, a statistical analysis was performed on data recorded from compressors before offshore installation. The main statistical tool, principal component analysis (PCA) was used to identify patterns in recorded data and to identify different operating modes from the recorded data. PCA techniques are available in statistical packages like StatSoft*. A related technique, partial least squares analysis (PLS) was used to predict outputs from measured input variables. PLS can be used to create ‘soft sensors’ or ‘virtual gauges’ which are also used to monitor abnormal behaviors such as unexpected changes in bearing temperature. Matrikon’s models were built using data like suction and discharge pressures and temperatures and other variables such as bearing temperatures and gas composition.
After cleansing, data was partitioned into five equipment modes representing shutdown, idling, start up, full load and normal operation. This allowed other recorded data sets to be automatically classified according to the operating mode. It then became relatively easy to identify abnormal operating modes such as that encountered when there was a problem with a bearing.
Offshore data is often compressed before transmission to the shore but this should be avoided. Models must be able to detect small changes in the data which are obscured by data compression. Most current techniques rely on thermodynamic equipment models. Data driven statistical models are more robust and are easier to retrain. In reality, a combination of statistical and thermodynamic models would likely make for a truly effective approach to condition monitoring.
* See www.statsoft.com/textbook/stpls.html.
© Oil IT Journal - all rights reserved.