Time-Frequency Analysis of Transient Vibrations to Improve Turbo-Compressor Start-Up

Diagnosis of turbo-compressors during start-up is a particularly challenging task. One of the reason is the reduced set of instruments that monitor this procedure. It is cumbersome to adjust lubrication and steam valves while controlling the speed and dynamic stability. In order to get the turbo-compressor out of a high vibration zone, it is important to be able to predict instabilities associated to the start-up process. Thus, it is necessary to have a measurement system with the ability of fault detection, especially at early stages of fault appearance. In this way, the start-up time can be significantly reduced. Although recent developed diagnosis methods use information from different sources and measurements, data structures are not designed to carry predictive information related to the turbo-compressor health. Therefore, it is important to extract early warning signals related to instability conditions. Vibration signals during machine start-up are non-stationary in nature, and conventional techniques, such as Fourier transforms and time series analysis, have difficulties to extract the full features of the vibrations signature. In this paper, the features of start-up vibrations in rotational systems like those found in turbo compressors are investigated by time-frequency analysis, and appropriate analysis of the transient vibration during compressor start-up is presented.