Assessment Techniques Based On Neural Network Air-conditioning Fan Status

This thesis develops a technique for equipment condition safety assessment utilizing self-organizing neural network based on the application to the equipments of SMSD (Site Material &Service Dept.) of Intel Technology (China) Co., Ltd.A new condition self-classifying system based on self-organizing neural networks is designed and employed to investigate the vibration of air-conditioning sets.Firstly, testing methodology and software for the analysis are introduced, data processing methods, such as elimination of the energy of structural vibration, are determined, and statistical reference standard is established based on analysis of a set of one year's field data for an accurate and reasonable outcome.Secondly, this thesis integrates kurtosis and characteristic information of effective value into a characteristic vector as input mode of neural networks, to carry out equipment condition classifying and assessing to ensure the signal processing is simple and condition recognizing is correct. 16 types of air-conditioning vibration conditions are assessed and self-organizing map (SOM) is employed to perform the visual clustering and feature extraction into a two-dimensional mirror plane, thus equipment vibration condition can be easily visually read out.Thirdly, this thesis points out that impacts of initialization of training sample upon the classifying capability of neural network are significant. A cross permutation of the training sample during initialization stage proves better classification. At the same time, this thesis also analyzes and confirms the influence of individual parameters upon the neural network's classification capabilities, and these parameters are: training sample, the zone size of characteristic plane, the number of self-study etc., and charts are utilized to illustrate these influences. This thesisalso highlights that localized detailing of the training sample can improve the resolution of the zones.