METHOD OF BATCH TRAINING OF NEURAL NETWORK WITH DELAY IN INPUT LAYER FOR INTEGRATED DIAGNOSTICS OF THE STATE OF FAN INSTALLATION OF THE MAIN AIRING

Abstract

Currently, the increasing of operational safety is one of the major problems that exist in mining industry. The problem of mining equipment accident rate is caused by avalanche accumulating part of its physical resource exhaustion. Fan installation of the main airing, which ensures normal vital activities
of mine personnel, is of the most importance among mining equipment. Therefore, an important task is to develop a software component, designed to diagnose the state of it and to be used in computer systems. The problem of building effective methods, providing high speed of diagnostics model
training, as well as a high probability, adequacy and speed of recognition of signals, which contain the vibrational information, lies at the heart of this objective. At present, as a tool for vibration diagnostics, such calculation methods as: kurtosis, crest factor, RMS value, envelope spectrum are most
commonly used. However, when using these markers separately for diagnosis of fan installations of the main airing condition, the probability of error is no less than 0.05. On the other hand, the processing speed of vibrational information is poor. Therefore, the development of methods for intelligent integrated diagnostics of fan installations of the main airing is relevant. As the use of artificial neural networks in the diagnosis gives a tangible advantage, which is that: the interaction between the factors is studied on finished models; it does not require any assumptions regarding the distribution of factors; a priori information about the factors can be omitted; the initial data can be highly correlated, incomplete or noisy; it is possible to conduct the analysis of systems with a high degree of nonlinearity; fast model development; high adaptability; the analysis of systems with a large number of factors; it does not require a complete enumeration of all possible models; the analysis of systems with
nonuniform factors, neural network method of diagnosis is used in the article. The aim of the study is to develop a method for analysis of the process of changing the condition of fan installations of the main airing. The model structure of an artificial neural network, which is a multilayer perceptron with
time delay in the input layer and provides a comprehensive analysis of diagnostic factors, is determined in the article. As a criterion for evaluation the efficiency of neural network model diagnostics the minimum of mean-square error is chosen. As a result of numerical study, it has been found that ten
hidden neurons does not change the RMS error significantly, the proposed network provides the diagnostic results with minimum deviation. The use of the proposed batch training mode has made it possible to accelerate the forward stroke for approximately. The created algorithms can be used to solve problems associated with the diagnosis of electromechanical objects.