【作者】 兰丽丽；

【导师】 陈友明；

【作者基本信息】 湖南大学 ， 供热、供燃气、通风及空调工程， 2008， 硕士

【摘要】 随着智能建筑的发展，空调系统的规模日益扩大，供热通风空气调节（HVAC）系统设备种类及数量也迅速增加。在建筑的整个生命周期内，包括设计阶段到运行阶段，故障层出不穷，导致大部分建筑通常都无法满足设计阶段的预期要求。同时，这些故障通常在短时间内难以察觉。对空调系统进行有效的故障诊断可以减少系统的能源消耗、维持舒适的室内环境、降低设备的损耗和减少温室气体的排放。因而，故障诊断对空调系统来说是十分重要的。本文提出用主成分分析法（PCA）对空气源热泵空调系统的传感器故障和空气侧换热器表面结污故障进行诊断与检测的基本思想。基于主成分分析法进行故障检测与诊断的基本流程主要包括四个过程：PCA模型建立过程、数据采集过程、故障检测过程和故障诊断过程。主成分分析法将测量数据空间为主成分子空间（PCS）及残差子空间（RS）。在正常情况下，PCS内的数据投影主要包含的是测量数据的正常值，而RS内的数据投影主要是测量噪声。当故障发生时，由于故障的影响，RS内的投影将会显著增加，通过对RS内的投影进行分析，我们可以进行故障检测。本文使用了平方预测误差（SPE）这个统计量作为故障检测的指标，并按照下面的规则来进行故障检测：SPE(x)￡2 d，系统运行正常；SPE(x) > 2 d，系统出现故障。故障重构的本质是寻找故障测量数据所对应的正常值的一个估计值的过程。本文采用迭代法来进行传感器故障重构。迭代重构就是沿着故障方向，逐步逼近主成分子空间的过程。本文采用识别指数SVI进行传感器故障识别，判断发生故障的传感器。SVI是故障恢复前后SPE之比值，一般来说，如果SVIj接近0，则重构方向为故障发生的方向，该传感器有故障发生；否则，如果SVIj接近1，则重构方向不是故障发生的方向，该传感器无故障发生。最后，本文用现场实地测试的数据对PCA方法进行了验证。利用PCA模型对空气源热泵空调系统传感器的偏差、漂移和完全失效故障，进行了检测、识别与重构。同时，用PCA方法对空气侧换热器表面结污故障进行了检测。结果表明，PCA方法是正确、有效的。更多还原

【Abstract】 In modern society, along with the rise and great development of the intelligence buildings, the scale of HVAC systems is increasing greatly, and the equipments category and quantity are increasing numerously, failures arise inevitably. Owing to problems that arise at various stages of the building life cycle, from design planning to operation, many buildings routinely fail to perform as well as expected and satisfy performance expectations envisioned at design. The effective fault detection and diagnosis (FDD) system for HVAC systems can reduce energy consumption, maintain a comfortable indoor environment, and reduce equipment loss and the greenhouse gas emissions. Furthermore, such failures often go unnoticed for extended periods of time. Therefore, it is very necessary to carry out FDD research in HVAC systems.This paper proposes a sensor fault detection and diagnosis scheme, and a fault detection method that provides detection of condenser fouling fault of air-source heat pump water chiller/heater, using Principal Component Analysis (PCA) method on an air-source heat pump air-conditioning system. The basic processes of the PCA-based FDD include four steps: PCA model building process, data acquisition process, fault detection process and fault diagnosis process.PCA method decomposes the data space into Principal Component Subspace (PCS) and Residual Subspace (RS). In normal condition, the data are mainly located in PCS, while when there is fault occurs, the data will deviate PCS and the projection in RS will increase significantly. Thus we can detect whether there is fault occurs via detection of the data projection in RS. Squared Prediction Error (SPE) statistic is used as indicators of fault detection. This paper suggests: when SPE(x)≤δ2, the system is under normal operation condition; when SPE(x)>δ2, an abnormal condition exists. Fault reconstruction can find faulty measurement data corresponding to the estimated value of the normal process. This paper uses iterative method to carry out sensor fault reconstruction. The iterative reconstruction is the process gradually moving towards PCS along the fault direction. Then, Sensor Validity Index (SVI) is used to identify the faulty sensor. SVI is the SPE ratio before and after the fault recovery. If SVIj is close to 1, it is considered that the reconstruction direction is not the fault direction; if SVIj is close to 0, it is considered that the reconstruction direction is precisely the fault direction.Finally, this paper uses field tests data to verify the PCA methods. In the air-source heat pump air-conditioning system, a PCA model was used to carry out fault detection, identification and reconstruction for the bias, drifting and complete failure on the sensors. Meanwhile, another PCA model was used to detect the condenser fouling fault was detected by the PCA method. The results show that the PCA method is correct and effective.更多还原