Research On General Fault Detection And Diagnosis Methods For Heating,Ventilation And Air Conditioning Systems

Heating,ventilation and air conditioning(HVAC)systems always suffer from various types of equipment faults and operational faults.The faults would lead to energy wastes and safety risks in HVAC systems.With the population of information technology,HVAC systems have been collecting a large amount of data.It is essential to develop automatic tools to monitor the system operations and diagnose the faults timely.The major challenge lies in the high complexity and large individual differences in the forms of HVAC systems.Most of the existing fault detection and diagnosis methods are predefined for specific conditions.In practices,it is challenging and time-consuming to develop specific strategies for each system using the existing methods.To tackle these challenges,this study proposes a set of general fault detection and diagnosis methods for HVAC systems,and carries out the following work from three aspects: realizing automatic understanding and analysis of the operational data,improving the generalization ability of fault diagnosis models,and determining the optimial fault diagnosis models.Firstly,with the aim of realizing automatic understanding and analysis of the operational data,a knowledge graph-based data analysis and anomaly detection method is proposed.Its basic idea is to mimic the general intelligence of human experts in understanding massive amounts of operational data from various HVAC systems,and further proposing customized anomaly detection solutions.A domain ontology is developed to allow computers understand the prior knowledge for HVAC systems data analysis and anomaly detection.Semantic rules are proposed to detect the anomalies in HVAC systems.These rules are written in abstract syntax.They can be reused in various operating situations.In the online data analysis and anomaly detection process,a customized knowledge graph is generated based on the ontology to capture the physics underlying the target system operations.The semantic rules are activated based on the knowledge graph to detect the anomalies.The results show that the proposed method can provide customized data analysis and anomaly detection solutions for different situations.It has high degrees of flexibility and generalization.Secondly,with the aim of improving the generalization ability of fault diagnosis models,a hierarchical object oriented Bayesian network-based fault detection and diagnosis method is proposed.Its basic idea is to reuse and combine the standard Bayesian network fragments to generate the system-level fault diagnosis models for HVAC systems.A Bayesian fault diagnosis class hierarchy is developed.Each class contains a Bayesian network fragment for diagnosing the local equipment faults.Inheritance is adopted to avoid repeated or inconsistent modeling of similar fault diagnosis classes.It allows sub-classes to inherit the Bayesian network fragments from their super-classes.For a specific HVAC system,the fragments are reused and combined to generate a hierarchical object oriented Bayesian network for real-time fault diagnosis.The results show that the proposed method can provide customized system-level fault diagnosis solutions for complex HVAC systems without tedious and repeated modeling works.The standard Bayesian network fragments are flexible and expansible,which can be reused in the HVAC systems of various types.Thirdly,with the aim of determining the optimial fault diagnosis models,a data-driven and knowledge-guided Bayesian network learning method is proposed.Its basic idea is to learn fault diagnosis models that have high accuracy and good model interpretation based on operational data and diagnostic knowledge.A probabilistic framework is developed to represent the expert knowledge for fault diagnosis.An improved genetic algorithm-based approach is raised for determining the optimal Bayesian network which has high consistence with both the knowledge and data.The search domain is the training data set.The diagnostic knowledge is utilize to guide the selection and evolution operations.In the online fault diagnosis process,local casual graphs are generated from the Bayesian networks for visually interpreting the fault action mechanisms.The results show that the optimal Bayesian network embeds both the prior fault-symptom relations obtained from diagnostic knowledge and the statistical fault-symptom relations learned from operational data.Thus it has higher diagnostic accuracy and good model interpretability.Finally,a general fault detection and diagnosis method is developed based on the knowledge graph and object oriented Bayesian network.The method is evaluated with a heating project.For the heating system,a customized knowledge graph is generated based on the ontology to capture the physics underlying the system operations.The semantic rules are activated based on the knowledge graph to detect the abnormal symptoms.An object oriented Bayesian network is generated based on the Bayesian fault diagnosis classes.Posterior inference is conducted to diagnose the faults based on the observed symptoms.The results show that the proposed method performs well in various operating conditions.The faults can be successfully diagnosed with incomplete diagnostic information.