上海交通大学硕士学位论文变风量空调系统传感器故障诊断研究姓名：孙勇申请学位级别：硕士专业：制冷与低温工程指导教师：晋欣桥20050101上海交通大学硕士学位论文- II -变风量空调系统传感器故障诊断研究摘 要随着中央空调系统在各种场合越来越广泛的应用中央空调系统得到了很大发展中央空调系统中的控制系统也越来越复杂不管是商用办公楼还是工业加工场所甚至是民用住宅都对空调系统提出了更高的要求要求空调系统的运行稳定舒适和节能而为了达到这些要求都需要稳定可靠的控制系统而在暖通空调系统中传感器测量信号的正确与否将直接影响系统能否正常的运行和实现需要的控制同时也影响对其它设备进行故障诊断时结果的正确性因此如何保障系统传感器的可靠性成为当前许多学者所关注的课题本文正是基于这样的背景研究如何诊断变风量空调系统的传感器故障以保证控制系统的稳定可靠地运转即当传感器的测量值偏移真实值时采用适当的方法找出具有偏差的传感器并且粗略地找出测量偏差值并使传感器的测量值回复到正常的水平在实际系统中进行故障诊断试验是非常不经济的同时还影响到系统的正常运转为了研究和验证故障诊断方法本文首先建立了变风量空调系统的数字仿真器在建立系统各部件数学模型的基础上以TRNSYS 为系统仿真平台将各部件模型连接成系统数字仿真器为了提高系统仿真器的运行速度本文在建立系统的压力流量平衡时把各部件的压力流量关系单独抽取出来建立单独的模型充分利用 TRNSYS多点叠代的特性求解系统各状态参数本文对仿真器进行了调试将系统中的各个 PID 控制器的控制参数都调节到恰当的水平使系统能够稳定运转系统的仿真结果表明系统数字仿真器能够模拟实际系统的整体运行趋势能够完成基本的控制要求为故障诊断的研究提供了必要的基础上海交通大学硕士学位论文- III -在故障诊断方面本文所提出的诊断方法是一种结合系统比较和优化的方法系统的比较和优化方法, SCOM 和多元统计方法主犯成分分析, PCA的故障诊断方法系统比较和优化方法以系统出现故障的传感器总是少数为假设条件通过检测系统各个测量值的一致性来检测传感器故障而主成分分析法利用系统正常运行条件下的测量数据来建立系统的模型从而避免了直接建立系统的解析模型系统比较和优化方法的可靠性在很大程度上取决于各个测量参数之间关系的完备性 系统越是简单对于测量参数的描述就越少 那么 SCOM方法的可靠性就越低相反系统越大传感器数目越多对于测量参数的描述就越多SCOM 方法的可靠性就越高在本文的仿真模型中SCOM 方法实现起来比较简单但是对于传感器小漂移不是很敏感主成分分析法把测量数据空间分为主成分子空间PCS和残差子空间RS在主成分子空间内包含的主要是数据的正常部分而在残差子空间内主要包含的是故障通过检测测量数据在残差子空间内的投影大小可以检测系统是否出现了故障PCA 方法实现起来比较复杂但是可以诊断出传感器的小漂移在进行采集信号的处理时本文利用小波进行噪声和粗差的滤出在进行故障恢复时本文利用变风量空调系统中的质量能量守恒和压力流量平衡来进行单个传感器故障的重构最后在本文所建立的变风量空调系统数字仿真器上对提出的故障诊断方法进行了验证验证结果表明利用本文提出的故障诊断方法能够较为准确地诊断出具有固定漂移故障的传感器并能够较好的诊断出单一故障点时传感器的固定漂移值而且能够以模块化的形式嵌入到中央空调上位机中对传感器的故障进行在线诊断研究结果表明本文提出的故障诊断方法具有较高的可行性也为其它系统传感器的故障诊断研究提供了参考依据关键词故障诊断仿真传感器固定漂移主成分分析法上海交通大学硕士学位论文- IV -SENSOR FAULT DETECTION AND DIAGNOSIS OF THE VAV AIR CONDITIONING SYSTEMSABSTRACTWith extensive application of the HVAC systems, the HVAC systems have beenimproved and become more and more complicated. Currently, the control systems inHAVC play an important role in the operation of the systems. Its stability and reliability isthe key of the performance of the systems, such as indoor comfort and energy consumption.In the HVAC systems, validity of sensor signals has important influences not only on thenormal operation of the system but also on the realization of necessary control. It also haseffects on fault detection of other equipments. Currently, many scholars focus on thestability and reliability of sensors in HVAC systems.On account of the importance of sensors, some researches have been done on how todetect and diagnose sensor fault in VAV systems in this paper in order to keep the controlsystem operating stably and reliably. In other words, in the cases of zero deviation (acommon fault for sensors), some proper methods have been put forward to find out thefaulty sensor and give out the fixed bias (i.e. fixed deviation) of the sensor approximately.So the control system can operate normally.Testing a FDD method on the installed systems is very expensive and may interruptthe normal operation of the systems. The better way to study and validate the FDDmethods is to use simulation. In this thesis, the mathematical models of all components inVAV system are set up first. Then they are incorporated with flow-pressure balance intoTRANSYS. After proper adjustment of the parameters, the simulator of VAV systems hasbeen completed. The results show that the simulator can describe the characteristics of theVAV systems and can be used as platform for control analysis. It also can be used fortesting and validating the FDD methods.The method presented in this thesis is a method integrating systematic comparison andoptimization method (SCOM) with principal component analysis method (PCA).Systematic comparison and optimization method is on the assumption that the number offaulty sensors in system is always minimum. Through the examining of consistency of allsensors reading in the system, the fault can be detected and diagnosed. The principalcomponent analysis method sets up its model through data got from normal operation ofthe system, so PCA method avoids setting up the analytical model directly.The reliability of SCOM is based on the integrity of the relationship of all data上海交通大学硕士学位论文- V -measured. If the system is more complicated, then the sensors used in the system will bemore and the relationships between all the sensors will be described more detailedly, so theSCOM method will be more reliable. The SCOM method can be realized easily in thesimulator used in this thesis, but it is not very sensitive to small bias.The PCA approach partition the measurement space into principal componentsubspace (PCS), in which normal data is included, and residual subspace (RS), in whichfault is included. Fault can be detected by detecting the projection of the measurement datain RS. The realization of PCA approach is complicated, but this approach can detect smallbias of sensors.This thesis used wavelet to filter the noise and gross error before the measured data areused for fault detection.In order to reconstruct the single sensor fault, this thesis used the law of energy andquality conservation and the balance of flow-pressure in VAV systems.Finally, th