THE DEVELOPMENT OF NEURAL NETWORK FOR THE MULTICHANNEL DIAGNOSIS OF STRUCTURE ELEMENTS DURING OPERATION

Abstract

The process of identifying damage to the structural elements in operation proposed use of a monitoring systems based on the concept of SHM that providing the structural health monitoring for its safe and reliable operation. The functional diagram of an integrated information system for monitoring and reasonably modern approach to the organization of monitoring the use of intelligent components - artificial neural networks, which provide the opportunity to realize the process of organizing information flows parallel multi-channel system. An example of a functional circuit monitoring system tanks ecological and hazardous substances and the scheme of the practical use of neural networks for solving the real problem of classification and recognition. Provides an overview of the status and structure of the classifier monitoring system to address the problem of determining the technical condition of the object of control. To detect the state of structural elements according to the analysis of diagnostic information developed probabilistic classifier based on a neural network PNN, whose architecture consists of two layers: a radially-basic neurons in the first layer and the layer of competition. Established classes for object state control and the corresponding parameters for each of them. Formed training set, which forms the basis of learning "teacher", for each class of classifier system diagnostics. Arrays are given training vectors in matrix form. Formed by different types of sets of test vectors to determine the feasibility and effectiveness of the built neural network. Determine the vector of indices of classes and a full array of connectivity of the neural network, establish membership input array information corresponding to the target class. A block diagram of a neural network diagnostic multichannel environment Matlab software, components of which are two layers, the input and output arrays and the weight matrix. Established the number of neurons of each layer constructed probabilistic network PNN multichannel diagnostics.