An active diagnosis method using support vector machines is presented. The support vector machine is arecently developed pattern recognition method that has some similarities with the neural network. It has astrong pattern recognition capability with relatively easy implementation processes. By introducing thesupport vector machines, a flexible and accurate damage diagnosis procedure is formulated. Theprocedure proposed here can be extended to an automatic diagnosis with strong learning capability. Themulti-dimensional feature vectors that represent the features of damages are generated by active sensingtechnologies based on ultrasonic wave propagation. Piezoelectric transducers were used for generatingultrasonic Lamb waves in a plate and for sensing the traveling ultrasonic waves. A network ofpiezoelectric elements attached to neighbors of a bolted joint is utilized to obtain the inputs and outputscombinations in the time domain. The recorded time histories are converted to multi-dimensional featurevectors to teach support vector machines. Simplified bolted joints were fabricated using aluminum platesand bolts. The excitation frequency of the ultrasonic wave is 50KHz. Another pattern recognition method,the correlation analysis, is also applied to the same feature vectors. The better accuracy of the proposedmethod is successfully presented compared with the correlation analysis. It is also shown that theapplication of wavelet transform exhibits a drastic improvement of recognition accuracy.
展开▼
