AN ARTIFICIAL IMMUNE SYSTEM FOR CLASSIFYING AERODYNAMIC INSTABILITIES OF CENTRIFUGAL COMPRESSORS

Anomaly detection prevents rotating machinery from unforeseeable faults. Using such detection methods, we consider the problem of identifying aerodynamic instabilities in centrifugal compressors by analyzing just the compressor's sound signal. Therefore, an artificial immune system (AIS)-based classification is applied. The deployed v-detectors work according to the negative selection principle and were optimized in position and shape by using a novel procedure of detector generation. This allows reducing the number of detectors while maintaining classification success. Ideas such as multiple layer AIS and fuzzy detector edges enhance the tuning of the classification response. Numerical results demonstrate the design, show the validity for the presented application and give an example for adopting soft computing methods for condition monitoring.