A genetic algorithm optimized support vector machine technique for rotor crack detection and classification using vibration and displacement signatures

Rotating machinery are being used at increased speeds and loads to meat high power demands. With modern machine design trends seeking light weight machinery, the ability to detect the crack initiation and propagation at early stage is imperative for a successful diagnosis of machine condition. Undetected cracks in rotors can lead to catastrophic failure and high costs of down-time and maintenance. This paper presents the results of experimental study aiming at detecting and identifying the progression state and location of cracks in a rotor test rig running at different speeds and disk unbalance conditions. Bearing Vibration signals from accelerometers and rotor orbital displacements using inductive proximity sensors are analyzed for feature extraction and pattern recognition. Two techniques, namely the genetic algorithm (GA) and support vector machine (SVM) are implemented in a hybrid form for rotor crack fault diagnosis. The study shows that the proposed techniques greatly improved the SVM classification performance.