| Heating Ventilating and Air Conditioning(HVAC)system is an important piece of building equipment that maintains human thermal comfort and improves the quality of the indoor air environment.It is used in a wide range of building types and its operation can be monitored by professionals through the operational data of the HVAC system.The development of sensor technology and data acquisition technology makes it possible to obtain massive amounts of data,while the development of computer technology promotes the rapid development and application of data-driven fault detection and diagnosis methods.Three main components of HVAC systems are cooling equipment,heating equipment and ventilation systems.One of the ventilation systems is the Air Handling Unit(AHU).During the operation of this system,a variety of faults can occur,some are component faults,some are caused by improper operation,and the severity of the faults varies.If a fault in the system is not detected on time,that fault may reduce the efficiency of the air conditioning controls,interfere with equipment operation,cause damage to the equipment,and consume more building energy than faultless situation,resulting in waste.Therefore,it is crucial to develop a sound and effective Fault Detection and Diagnosis(FDD)method for HVAC.This paper will focus on the FDD methods for Air Handling Units in HVAC systems,mainly including the following.First,in terms of fault detection,this study proposes a fault detection method based on Linear Confidence Upper Bound(Lin-UCB)for action selection in Contextual Bandit(CB)algorithms.The experiments use operational data as the training environment for the Agent,and the labels of the data are used as the reward after the Agent selects an action.The Agent continuously selects a fault location or a fault-free case to obtain the maximum cumulative reward based on the measured data in the environment.The trained Agent can select the best action,i.e.,the fault location when a fault occurs or a fault-free case.The experimental results show that the method can enable the fault detection model to achieve good performance,detect the fault occurrence location online,and provide assistance in scheduling maintenance.Second,in terms of fault diagnosis,this study proposes a semi-supervised fault diagnosis method based on multi-classification Generative Adversarial Networks(GAN),replacing the binary discriminator of the original GAN with a multi-classifier.The improved multi-classification GAN is a semi-supervised learning method which means that it first learns information about data distributions or patterns present in unlabeled data,and then combines this information with limited labeled data to complete the supervised learning task.Finally,a new self-training scheme is proposed for correcting the class imbalance in the GAN with and without labeled data during training.This scheme can effectively improve the robustness of the model to handle the class imbalance situation in the training data.The experiments use real air handling unit operational data to verify the high classification accuracy of the method for fault diagnosis.Also,the multi-classification GAN can reduce the false alarm rate for the fault-free case,so that the experimental results are not biased towards the fault-free class with more training data.Third,in terms of novel fault detection,this study proposes the One Class Support Vector Machine(OC-SVM)method with data pre-processing using Principal Component Analysis(PCA)for dimensionality reduction.Since novel faults can be considered as outliers or novelty values without any labels,the studied algorithm is an unsupervised learning method.First of all,the OC-SVM is trained using data which have been preprocessed,and then,after inputting the novel fault data,it is able to separate the original data from the abnormal data using an optimized hyperplane.The experiments use the fault cases discussed in the above method as original faults and sensor bias faults as novel faults.The results show that the proposed algorithm can separate abnormal data points and validate the detection accuracy of the new fault with the addition of the fault detection algorithm.The research and intervention of the novel fault detection method makes the original fault detection and diagnosis framework more complete and more robust.This study proposes a fault detection and diagnosis method targeted to the air handling units of building HVAC.The method can realize online fault detection and high precision fault diagnosis,and can detect new types of fault conditions that occur in the later stages of system operation. |
