In this research, the fault detection and diagnosis using a model-based technique for the cracked rotor vibration system is developed and implemented. More specifically, the observer based or filter bank approach is employed in the fault detection and diagnosis process in order to detect the occurrence of a crack and diagnose the position and the depth of the crack in rotating machinery.
The fault detection and diagnosis process is consisted of two parts. The first part is the filter bank or the residual generation which generates the residual vectors corresponding to each observer. The second part is a voting algorithm which searches the observer that corresponds to the behavior of the real system.
The type of filter contained in the filter bank is the discrete time-variant Kalman filter. The filter is specifically designed to track the cracked rotor vibration system. Since the filter is time-variant, the state matrix at the current time step of the filter is updated by the state estimated value from the previous time step.
Constructing the filter bank with the presented filter allows the fault detection and diagnosis process to perform very well under the environment of the process and measurement noises which is unavoidable in real systems.
The voting algorithm evaluates every observer to find the observer behaving the closest to the real system based on the score achieved by each observer. The score is calculated by the information of the residual mean, the residual autocorrelation of each observer, the correlation coefficient between the real system measurements, and the observer outputs.
In order to evaluate the fault detection and diagnosis process performance,
the fault detection and diagnosis process is tested with the simulated real system containing various sets of system parameters. The results and discussions are presented.