Centrifugal Compressor Instrumentation for Developing an Anomaly Detection Method Using Vibration- and Acoustic Measurement Data

In various industrial applications, acoustic measurement data is used for condition-based monitoring and predictive maintenance. These concepts enable the early detection of faulty machine behavior as well as the efficient planning of maintenance. Time-resolved measurements offer a promising approach for fault identification since every industrial process has a unique spectral fingerprint that allows for the differentiation between operating points as well as normal and faulty machine behavior. For this purpose, the acoustic reference state of the machine is determined and compared with measurement data obtained during operation to quantify the changes in the acoustic signal. In the field of turbomachinery, however, this method is not yet widely used due to the complexity of the process and the high overall noise level. In this paper, a single-stage centrifugal compressor is equipped with high-frequency accelerometers and free-field microphones in addition to conventional process data instruments for anomaly detection. Time-resolved spectral evaluation methods and their advantages for gaining information about sequence of effects and distinguishing time-dependent phenomena are presented. Moreover, an approach for specifying a confidence interval to determine the degree of amplitude fluctuation in frequency domain between measurements is shown. In order to further verify the repeatability of the measurement result, the impact of material warming and compressor reassembly on the measurement data is examined statistically. Furthermore, a method for synchronizing time-resolved measurements in frequency domain is elaborated. Sensor positioning for optimal measurement results is discussed and the reference state with which future measurements will be compared with is determined using the above-mentioned techniques.