The Key Equipment Of The Food Machinery Fault Diagnosis Based On Support Vector Machines

In equipment fault diagnosis, fault confirmation and fault classification are essentially pattern recognition. Pattern recognition of equipment fault is a typical limited sample problem. In small-sample learning, it is very difficult to obtain good generalization performance by conventional machine learning method. Based on statistical learning theory, support vector machine(SVM) is the most advanced machine learning algorithm in the field of pattern recognition and shows more superiority than other methods. It can solve small-sample learning problems better by using structural risk minimization than empirical risk minimization. Moreover, by using the kernel function idea, SVM can change the problem in non-linear space to that in the linear space in order to reduce the algorithm complexity.The most predominance of SVM is proper for limited sample decision. The nature of the algorithm is acquiring connotative class information to great extent from limited samples. With its excellant generalization abilities, SVM can be used in fault diagnosis engineering practice.SVM has excellent predominance in theory, but the research on application is comparatively delayed. On the basis of systemetical study on SVM algorithm priciples, this dissertation focuses its practical research on the application of SVM in the field of equipment fault diagnosis on three aspects: the SVM fault pattern recognition, multi-class fault diagnosis and condition trend prediction for mechanical equipment based on support vector regression.The main research of this dissertation is summarized as follows:1 .Binary class classification algorithm of support vector machine and its application in mechanical fault pattern recognition under limited fault samples is studied. Appropriate binary class classification SVM is structured, with which experiments are carried out to recognize electrical motor bearing faults and the faults under noises. The result shows that SVM fault recognition method has excellent classification ability to limited fault samples and noisy fault samples. 2.Multi-class classification algorithm of support vector machine for mechanical fault diagnosis and its engineering application are studied. On the basis of binary class classification, multi-class classification algorithm is studied and applied to multiple fault diagnosis of mechanical agitated ventilation fermenter. After training on multi-class fault samples from the fermenter, fault diagnosis is carried out with the SVM multi-class fault classifier got from the training. 6 kinds of faults are recognized and diagnosed completely and accurately, achieving satisfactory diagnosing result. Compared with the experiments based upon neural networks, those based upon support vector machine have better classification ability under limited fault samples and SVM is proved to be a new effective method in fault diagnosis.3.Support vector regression algorithm and its application in equipment condition trend prediction are studied. Condition trend prediction is one of the important methods for incipient fault prognosis. Used in predicting the 30-hour peak-peak value of the vibration signal from refrigerating compressor, the predicting error is only 2.1934%, which shows high practical value of this method.