Data derived time-frequency segmentation of non-stationary vibration signals

Mechanical devices like solenoid valves play a vital role in many machines and systems. If one of these devices breaks down the whole system can be affected. Thus it is desirable to monitor the condition of solenoid valves and detect failures before they can cause serious damage. Typical failures for solenoid valves are internal leakage, caused by waste in the device, or deterioration of seat and seal and stuck to open or closed, caused by increased friction. The vibration of the observed device due to the impact of the movable part in the valve is monitored and analysed to perform this monitoring task. The vibration impulse has a transient and non-stationary character and has a similar shape to an acoustic emission burst signal. The frequency range of the vibration impulse is from DC to approximately 20kHz. In this paper a new technique is proposed to perform segmentation of the raw signal in the time-and frequency-domain to isolate more stationary segments in the overall signal for the purpose of classification of faulty and unfaulty devices.