The Research On Remote Fault Diagnosis Technology For Marine Asynchronous Motor

Asynchronous motors are main support devices for both marine propulsion and power systems. Marine safety may be threatened if any motor in the propulsion and power systems is not function normally. To increase the system safety, we have to implement a timely repairing and maintaining system for motors. It is very important that the motor fault is timely detected based on the remote fault diagnosis system for marine motors.This thesis firstly studied the characteristic of systematic frame and data processing for the remote fault diagnosis system based on Internet. In view of the open-sea environment for ocean ships, a hybrid frame of marine remote fault diagnosis system was introduced of performing both on-line and off-line. On-line fault diagnosis is implemented based on web which adopts browser/server frame communicating by INMARSAT. Off-line fault diagnosis is implemented based on Email which adopts client/server frame. The proposed fault diagnosis technique being compared with other fault diagnosis methods, the eigenvalue for motor fault diagnosis based on motor current signature analysis is achieved a better result, which measure circuit is non-aggressive.We considered the data which supply to remote fault diagnosis system is short, unsynchronized and polluted with strong noise because of special marine operation environment. In addition, power spectral density of eigenfrequency during incipient fault period is low. For above problem, this thesis analysed the characteristic of three general spectral analyses methods using differently simulated data.A detect method of motor incipient fault based on multi-taper method was introduced. A balance problem between frequency resolution and variance was studied, and the optimal balance value was chosen to be applied for remote fault diagnosis. By selecting high energy tapers, we eliminated root leakage of eigenfrequency, and brought the shape of eigenfrequency to be distinguishable. Experimental studies based on the data for artificial fault were conducted and results show that multi-taper method has a better steady and antinoise performance comparing with three other methods. The eigenfrequency processing and fault classifying method based on Wavelet Packet Transform was introduced, aiming at classification for the outbreak fault. Preprocessing the motor current based on Hilbert transform was validated that has a better performance, aiming for avoiding leakage of supply frequency, and various decompose coefficient along with the vary of motor load. Frequency analysis range for every node of wavelet packet transform was studied, and be corresponded to range of current eigenfrequency. There is 25Hz interval between the two adjacent current eigenfrequencies, and actual value of current eigenfrequency presents a 2～15 Hz negative error because the slip is less than 1. Then, utilizing six-layer wavelet packets decompose for preprocessed signal, because that there was only one eigenfrequency in every bandwidth of wavelet packet decomposition, which solved the uncertainty problem of motor operation parameters. The overlap of frequency analysis range between the adjacent nodes was studied, and could be decreased by increasing the number of crests of wavelets. Simulation results show that the method is validated. The root mean square (RMS) method for reconstructed node coefficient that presents motor fault is more feasible than the RMS for node coefficient. Experiments were conducted using the data for artificial faults and results show that fault classifying method based on Wavelet Packet Transform is valid.A classify method combining rough set theory with self organizing feature map, was introduced aiming at accumulated fault information. The clustered problem for continuous attribution was studied; slow convergent speed could be achieved by using proportion function. During the training phase, neighborhood region was changed based that Gaussian function was chosen as neighborhood function. The disturbance owing to wavelet overlap was effectively decreased, by using training samples composed with two mean-squared roots for the adjacent nodes. The discernibility matrix was constructed based on rough set theory. The diagnosis rules were extracted from the discernibility matrix. Application studies based on the data for artificial fault were conducted and results show that the above method is feasible.