Research On Fault Detection And Diagnosis Of All-air Air Conditioning Systems Using Bayesian Network-based Diagnostic Units

All-air air conditioning systems are prone to various kinds of faults due to its complexity.The occurrence of faults can result in poor indoor environment,large energy wastes,shorten life spans of equipment,and even cause safety problems.Fault detection and diagnosis(FDD)can detect anomalies and diagnose faults as soon as possible.It provides decision supports for maintenance and life cycle commissioning of the air conditioning systems.Effective FDD is significant for safe,efficient and stable operation of all-air air conditioning systems.In the recent years,a great number of studies have been made around the FDD methods for all-air air conditioning systems.However,most of the existing methods developed FDD models for air conditioning systems of the specific type.In general,the FDD model develop for a specific system is unsuitable to other systems.To overcome this problem,this paper proposes a fault detection and diagnosis method,aiming at providing FDD solutions for all-air air conditioning systems of various types.A FDD knowledge library is developed to represent and store FDD knowledge using a group of Bayesian network-based diagnostic units.For the targeted system,the related diagnostic units are selected and interlinked to develop the integrated FDD model.Monitoring data are entered the FDD model to inference the faults locations.The main work of this paper is as follows:(1)A FDD knowledge library is developed for all-air air conditioning systems using Bayesian network-based diagnostic units.The knowledge library is composed of four sections,which represent knowledge about system components,control strategies and operation modes,sensor configurations,fault inference,respectively.The former three sections describe the prior knowledge for FDD of all-air air conditioning systems with the consideration of system diversity.The last section describe the FDD knowledge using a group of Bayesian network-based diagnostic units.The diagnostic units are defined to describe the probabilistic causal relationship among typical faults,symptoms and additional information of the commonly-used component in all-air air conditioning systems.A group of conditional rules is designed to determine whether the units are applicable.(2)A knowledge library-based FDD method is proposed for all-air air conditioning systems.It has two steps.In the first steps,the related Bayesian network-based diagnostic units are selected according the configuration information of the targeted all-air air conditioning system.The instances of these units are developed based on the conditional rules,and then interlinked together to generate a multi-layered FDD model.In the second step,diagnostic information is collected and entered into the FDD model.Backward inference is conducted to calculate the posterior probability distributions of all suspected faults.A set of rules is proposed to find the most possible fault reasons based on the backward inference results.(3)Evaluations are made on a real all-air air conditioning system to demonstrate the effectiveness of the knowledge library-based FDD method.A specific FDD model is generated for the system using the proposed method.Experiments are conducted to introduce faults into the system for obtaining faulty data.The experimental data are used to evaluate the performance of the FDD model.The results show that the generated FDD model has high diagnosis accuracy.