【作者】 王洋；

【导师】 马最良；

【作者基本信息】 哈尔滨工业大学 ， 供热、供燃气、通风与空调工程， 2007， 博士

【摘要】 能源和环保问题是当今世界共同关注的热点,也是今后长期存在的问题。我国是能源生产和消费大国,面对新世纪,如何保持能源、经济和环境的可持续发展是我们面对的一个重大战略问题。而我国北方目前常用的几种供暖方式,大都是“热源消耗高品位能源、向建筑物室内提供低温热源、向环境排放废物”的单向型模式,将给我国未来的能源供应和环境保护带来巨大压力。单、双级耦合热泵系统(Single-double stage coupling heat pump system,简称SDSCHPS)是一种以热泵理论为基础,以“生态循环供暖”为理念的新型热泵供暖系统,该系统是HVAC&R领域为推动热泵技术在北方应用而进行系统创新的一个成功案例,具有较高的实际推广价值。但是,由于SDSCHPS结构的复杂性、制热运行时环境的特殊性、运行模式的多样性和多种故障的渐变性等原因,使得系统运行的可靠性、制热效率和使用寿命受到一定影响。而目前HVAC&R领域的多数故障检测与诊断系统仅有当温度、压力等超过预设值时切断设备的功能,既难以对设备当前的运行状态进行监测,又无法提供故障产生的具体原因和解决故障的具体措施,导致了暖通空调设备的运行可靠性下降、维护与维修费用的增加和大量不必要的能源浪费。为提高SDSCHPS的运行可靠性、节能性和智能化管理水平,为其在我国北方地区的推广应用提供有力保障,本文对系统中可能出现的主要故障进行深入分析和实验研究,探求系统在低温环境下的运行规律及故障与其征兆之间的关系,并初步构建融合人工神经网络和专家系统的分布式故障诊断智能系统模型,从而为SDSCHPS的故障诊断系统的开发提供较为完善的研究基础和理论框架。应用层次分解法,分别构建SDSCHPS中两大主要设备——空气/水热泵机组和水/水热泵机组的故障树,采用故障树分析法对SDSCHPS中可能出现的各种故障进行深入研究,按其出现部位,将其故障主要分为三大类,即制冷系统故障、除霜故障和控制系统故障。并采用故障出现频率和维修费用比例作为评价指标,对这些故障进行重要度分析,找出构建SDSCHPS故障诊断系统所需研究的重要故障和常见故障。由于热泵机组在“非健康”状态下运行时性能会有所降低,本文提出一种基于广义回归神经网络(GRNN)的SDSCHPS状态监测方法,充分利用GRNN良好的预测能力,通过对系统运行时制热能效比EER预测值和实际值的比较分析,实现对SDSCHPS的状态监测,完成了设备故障检测与诊断的第一个任务。并通过实验测取不同室外环境和运行模式下SDSCHPS的性能参数,并以此作为样本,通过拟合训练和预测分析,验证所构建的GRNN模型应用于SDSCHPS状态监测的可行性和有效性。分析了SDSCHPS中制冷系统的常见故障及其特征,确定对其进行故障诊断和推理所需要的特征参数和测点布置;并应用定性理论分析方法,对SDSCHPS制冷系统的常见软故障与特征参数之间的关系进行定性分析,并分别构建SDSCHPS中空气/水热泵机组和水/水热泵机组的制冷系统标准软故障模式库;然后分别构建SDSCHPS不同运行模式下制冷系统基于GRNN的故障诊断和基于规则的故障推理模型,实现了对SDSCHPS中制冷系统故障的实时诊断及推理。通过实验研究空气源热泵机组在不同程度的室外换热器堵塞故障工况下,其主要运行特性参数和性能参数的动态变化规律,验证了前文通过定性理论分析所获取的典型故障之一,即空气源热泵机组室外换热器传热效果不佳软故障发生时,故障与征兆之间对应关系的准确性,并深入分析室外换热器堵塞故障对整个空气源热泵机组运行工况和性能的危害。通过实验研究了在正常除霜、提前除霜和滞后除霜等不同除霜工况下,空气源热泵机组主要运行参数和性能参数的动态变化规律,并深入分析除霜故障对机组的运行工况和性能的危害;同时,利用概率神经网络(PNN)良好的模式识别能力,构建一个简单的除霜故障诊断模型,实现了对空气源热泵机组除霜故障的实时检测和诊断;为减少除霜故障的发生,本文还提出一种改进的除霜控制方法,即基于室外风机电流和盘管温度的除霜控制方法,利用实验结果对该方法的可行性进行分析验证,并设计出该除霜控制方法的基本流程。由于SDSCHPS控制系统中的多传感器测量值之间存在冗余关系,以信息融合理论为基础,以人工神经网络(ANN)为工具,提出一种多传感器故障诊断及容错方法,实现了对SDSCHPS中多传感器的故障诊断及信号恢复。并结合一组实验数据,分析验证该方法对SDSCHPS中常见传感器的四类故障的诊断及容错效果。针对传统的专家系统和人工神经网络应用于故障诊断时各自存在的不足,并根据SDSCHPS的结构特点和运行模式,提出一种融合人工神经网络和专家系统的分布式故障诊断智能系统,既克服了专家系统和神经网络各自的缺点,有充分利用了二者的长处,解决了单独应用专家系统和神经网络所无法解决的问题。同时,深入研究SDSCHPS分布式诊断过程中诊断任务的分析与分解、各子任务的求解、各子任务解的综合等问题,初步构建SDSCHPS的分布式故障诊断智能系统框架,从而为SDSCHPS进行故障诊断系统开发及应用提供了理论基础。本文的研究工作对推进SDSCHPS智能型机电一体化、延长设备使用寿命、提高系统的运行可靠性和效率、降低能耗、减少维护和维修费用等,有着十分重要的实际意义,而且为HVAC&R领域故障诊断智能系统的构建从方法上进行了新的探索,提供了一条切实可行的新途径。更多还原

【Abstract】 The issue of energy sources and environment protection is always a focus in our current world, and will be a long-term problem existing in future. China is one of the countries which explode and consume great amount of energy sources every year, and in the 21 century, how to keep harmonious development is a strategic question which we have to face. At the present time, the common heating methods in north China mainly are single-direction modes, which will bring huge pressure on our future. Single double-stage coupling heat pump system (SDSCHPS) is a new kind of heating system which based on heat pump theory and zoology recycle heating concept, and also a successful case to extent air-source heat pump to the north. However, because of its complicated structure, peculiar operation environment, variety of operation modes and gradualness of many faults, the operation reliability, efficiency and life of SDSCHPS are influenced in a certain extent. Currently, most fault detection and diagnosis (FDD) systems in HVAC field only have the shut-down function when correlative temperature or pressure exceeds scheduled parameter. The existing FDD systems not only can’t monitor unit’s operation conditions, but also can’t give concrete reason and effective solution of fault. To improve its operation reliability, energy-economy and intellectualized manage ability, this dissertation theoretically analyzed and experimentally studied the mechanism of main faults which may occur in the SDSCHPS, then searched the logic relationship between the faults and symptoms, and construct its distributed FDD system which combines artificial neural network (ANN) and expert system (ES), thereby, brought integrated frame of FDD system in the SDSCHPS.By applying hierarchy-decompose method, the fault trees of the air-to-water heat pump unit and the water-to-water heat pump unit, namely, the central equipment of the SDSCHPS were set up separately, and using occurring frequency and maintenance expense as evaluation index, the important faults and common faults of the SDSCHPS were found out. According to its occurring position, the faults of the SDSCHPS were classified to there kinds, namely, refrigeration system’s faults, defrost system’s faults and control system’s faults. Owing to its performance would be lower when running in unhealthy condition, this dissertation put forward a state-monitor method of the SDSCHPS based on general regression neural network (GRNN). By utilizing the nice forecast ability of GRNN fully, the first role of the SDSCHPS’s FDD system was realized, which means that the operation state-monitor was completed by comparing and analyzed the forecasting EER and the real EER of the SDSCHPS. Then, the performance parameters of the SDSCHPS were measured under different operation conditions and modes by experiments, which acted as samples of GRNN model, and the feasibility and validity of the state-monitor method were validated by simulating and analyzing.The common faults of refrigeration system in the SDSCHPS were investigated, and the characteristic parameters were confirmed, which were needed to fault diagnosis and reason. By applying qualitative analysis method, the relationship between the common faults and its symptoms were analyzed and the standard fault mode-base of the SDSCHPS were constructed. Finally, the FDD models based on GRNN and the fault reasoning based on rule of refrigeration system under different operation modes were set up, and the real-time FDD and reasoning of the refrigeration system were realized.The relationship between one kind of common fault occurring in air-source heat pump (ASHP) and its symptom were validated experimentally. This fault will happen when the outdoor heat exchanger’s air-side of ASHP jams with ice, snow, dirt or some other stuff, which results the worse of the heat exchanging efficiency. By experiments, the dynamic variety rules of those main operation characteristic parameters were measured and analyzed, the relationship between the fault and its symptom was substantiated and different grade faults’harm on the unit’s operation conditions and performance was studied quantificationally.Similarly, the dynamic variety rules of main operation characteristic parameters and performance parameters of an ASHP unit were investigated experimentally when it run under different defrost states, such as normal defrost state, advance defrost state and lag defrost state, at the same time, the defrost faults’harm on the unit’s operation and performance was also studied. Then, applying outstanding mode-identify ability of the probability neural network (PNN), a simple defrost FDD model was set up, which helped to detect and diagnose defrost fault of ASHP in time. Finally, to diminish the defrost faults, this dissertation brought out an improved defrost control method, which was based on the change of outdoor fan current and wing surface temperature. The feasibility of this method was analyzed and validated by experiments, and its basic flow was designed elaborately.Because of the redundancy relationship between the multi-sensor measure data in the control system of the SDSCHPS, this dissertation put forward a kind of multi-sensor fault diagnosis and error-tolerance method, which was established on information amalgamation theory and ANN tool. Through simulating with a group of experiment data, the diagnosis and error-tolerance effect on four kinds of common faults, which were prone to occur in the control system of the SDSCHPS, was analyzed and validated thoroughly.Based on the limitation of traditional ES or ANN, this dissertation pointed out a distributed FDD artificial system in the SDSCHPS, which used ANN and ES simultaneously. The distributed FDD system not only could overcome the limitation of ANN or ES, but also take full advantage of these two tools, thus solve the problem which couldn’t settled by only one of them. On the other hand, in the distributed FDD system, these problems, such as the decomposition of whole FDD task, solution of every subtask, integration of each subtask’s answer, and so on, were researched thoroughly, and the theoretic frame of the distributed FDD system in the SDSCHPS was established, therefore, the theory foundation to design and apply of the distributed FDD artificial system was realized.The study work of this dissertation was very important for these problems’solution, such as advancing artificial machine-electrification incorporation, prolonging the life-span of HVAC equipment, enhancing the operation reliability and efficiency, reducing energy resources consumption, and diminishing maintenance expense, and so on. Therefore, some new methods about artificial FDD system’s construction were probed, and a new viable way was offered to settle FDD problem in HVAC field.更多还原