Fault Detection & Diagnosis Of Clean Room HVAC System

With development of society and improvement of people life quality, HVAC system have been applied in many aspects of industry and life, the systems is getting huge and more complex, it lead to the faults rate increase. The faults are sure to bring significant influence to life environment and production environment. Especially for clean air-conditioning system used in pharmaceutical industry, as the special nature of its application its faults certainly lead greater harm to the people's lives and property. So it is particularly important how to effectively identify faults when the clean air conditioning failures occur. Traditional method would lead to subjective decision-making with great, and will cost a lot of manpower because of the complexity of the system.With the continuous development of artificial intelligence, Fault detection system (FDD) has been Apply in the aerospace, pharmaceutical, electronics, materials and other industries, it makes possible using artificial intelligence on the clean air conditioning system fault detection.This article describes the significance of FDD used in HVAC system, and the FDD content and methods used in HVAC systems equipment are described in article; And the recent study Status development and problems of HVAC system FDD techniques has been given, the main methods of FDD in HVAC have been classified; Five Common faults occur on clean HVAC system have been Summarized, and seven parameters which can response faults have been find; Necessity of Integrating fuzzy logic and neural network has been discussed in paper. The relationship has been built up between seven parameters and five faults use method of FNN. The method combines the merits of fuzzy logic and neural network. Avoid the problem of fuzzy rules is difficult to obtain as the lack of empirical knowledge, while making the neural network gets a clear explanation.The FDD based on FNN used in clean HVAC system has been realized using MATLAB 7.0. Network has achieved convergence value soon under the appropriate parameters when it was trained.