Study Of Circular Self-Configuring Algorithm And Application In Fault Diagnosis Of The Blower

Based on the neural network theory, self-configuring algorithm on neural network generalization is presented in this paper. Self-configuring algorithm is a correlated branch cutting method in multi-layer feed forward network based on BP algorithm. In order to delete and combine the hidden neurons, and simplify the network structure, it introduces the conception of the correlation coefficient and dispersity, analyses the correlation between the output and hidden neurons. A method applied linear regress is always to get the combine rule; generally speaking, the problem is complex nonlinear relation. So a self-configuring algorithm based on polynomial regress is introduced, which is instead of linear regress, then we integrate the MOBP to accelerate the convergence speed, viz fast self-configuring algorithm. In the experiment, after pruning multi-networks with the same or different hidden neurons if the parameters are invariability in self-configuring algorithm, we will get the simplified network containing different hidden neurons, it means that there is un-identical in the convergence of network structure. The cause is mainly the randomness of the initial weight value and offset value. To solve this problem, a circular self-configuring algorithm based on the self-configuring algorithm, randomness and divide-and conquer is presented.Analyzing the fault mechanism of suction blower and setting an example of centrifugal blower in a famous automobile manufacture plant, a suction blower fault diagnosis mode based on the circular self-configuring algorithm is established. It can take the normalized vector of different frequency peak energy amounts on 9 bands in vibration signal frequency of every test-point as the input vector of network and the fault pattern as the output vector, and use Matlab to simulate and study. The result shows: Circular self-configuring algorithm can effectively solve the un-identical convergence of network structure, which isn't the character of the self-configuring algorithm.。 Circular self-configuring algorithm can converge at simplest (or round) network structure, which doesn't achieve by self-configuring algorithm. After using circular self-configuring algorithm to prune network, the complex degree of knowledge expressed by network will decrease largely, so network has more effective knowledge expression of fault pattern.