Vibration based fault detection for Solenoid valves
Solenoid valves play a vital role in many machines and systems. If one of these devices breaks down the whole system can be affected. Because of this importance of valves, it is desirable to observe these parts to detect faults, both when they are occurring and before they can cause serious damage. Among several possible methods of observation (monitoring actuation time, electrical current, fluidic parameters and others) the observation of mechanical vibrations is a well known method of observing mechanical systems which is commonly used for observation of rotating machinery, but which includes several challenges for diagnosis of solenoid valves. This thesis investigates the possibilities and advantages of vibration analysis of fault detection for solenoid valves. New algorithms are developed to automatically segment the overall non- stationary raw data to smaller sections with a higher degree of stationarity. These new segments are interpretable in a mechanical sense and they separate different sources of vibration. Furthermore a new method to detect regions of interest in a spectrum for classification without "a priori knowledge" about the process has been developed. The experiments presented in this thesis give the evidence that these new methods of pre¬processing and feature extraction enable reliable classification results for transient signals as they occur in the vibration of a switching valve.