Wavelet Neural Network Based Sensor FDD Method In HVAC Systems

In order to save energy and improve indoor air quality, the optimal control strategies of heating, ventilation and air conditioning (HVAC) systems have become more and more complex. However, the key to realize those optimal strategies is the accuracy and reliability of sensors. Unfortunately, some sensors inevitably occur fixed or drift biases after a relative long-term operation in the systems. These faults may mislead the controller. As a result, the aim of advanced optimal strategies cannot be achieved at all. Thus, seeking suitable method to discover sensor faults when they are biased on HVAC systems will be solved firstly in this paper. On the other hand, HVAC systems are so large and complex that they concern a lot of sensors. Once some sensors have faults, how to isolate them one by one is another important point in this paper.Based on surveying and concluding the methods of fault detection and diagnosis (FDD) , Wavelet Neural Network(WNN) method is chosen to be the base algorithm and used for data-driven for sensors FDD in HVAC system. WNN uses the data of signal's frequency realm feature obtained by Wavelet Analysis as the training samples to train Neural Network, and takes the results of Neural Network as the final judgment. Based on WNN algorithm, there're two FDD methods which are respectively called Single Information Diagnosis and Combined Information Diagnosis are presented in the paper.Single Information Diagnosis uses the data of one single sensor to diagnose the faults of its own. Taking the advantage of Wavelet Analysis for the acquisition of signal's frequency realm feature and Neural Network for feature studying, It is validated by simulation tests that Single Information Diagnosis can effectively realize the diagnosis of sensors'fixed bias faults and drift bias faults in HVAC system. However, since there are large scales of sensors in HVAC system and complicated signal linkages among them, Single Information Diagnosis, regardless of the relevance between sensors, may have mistakes in diagnosis to some degree. To diagnose fault for multi-sensors in HVAC system , Combined Information Diagnosis, which combining the relevance between sensors based on energy and mass balances, is set up. The results of validation test show that Combined Information Diagnosis incorporated the modules based on energy and flow-pressure balances can diagnose the faults effectively.