| 摘要 | 随着社会经济的迅猛发展和人们生活水平的不断提高，空调系统应用的越来越广泛，已经变的必不可少了。当空调系统发生故障时，如果不能被及时诊断，就会造成能源和设备上的损失，影响正常的生产和生活。由于传统的故障诊断方法己经不能满足实际的要求，所以国内外学者纷纷进行空调系统的故障诊断技术研究，在众多的研究中，最引人注目的是人工神经网络方法。
本课题以城建学院HVAC实验室的冷水机组空调系统为研究对象，进行了相关的实验研究。在空调系统的故障诊断研究中，把人工神经网络作为故障诊断系统的推理机。在建立故障诊断系统的神经网络模型时，分别对神经网络的隐含层数、隐含层单元个数、训练函数、学习速率等进行了详细的讨论。在训练神经网络时，为了使网络能更好的收敛，对网络的输入参数进行了预处理，并比较了模糊化、离散化和归一化三种方法的优劣，最后确定归一化的方法，作为神经网络输入参数的预处理方法。
为了取得空调系统在运行时故障与系统参数之间的对应关系，设置加热器、冷冻水泵、冷却水泵等六种故障进行模拟实验。通过空调系统的故障实验获取了大量的运行参数，对这些参数进行了详细的分析，并建立空调系统的故障征兆集。利用Visual Basic语言的可视化优势，编制了空调系统的故障诊断软件，其中在用VB调用MAT LAB时，作了大量的工作，最终利用ACTIVEX自动化协议实现了两者的相互调用。
利用故障征兆集对神经网络进行训练，并使用未训练的数据对软件进行验证，故障诊断软件能够很好的诊断出故障原因。
研究结果表明，把人工神经网络应用于空调系统的故障诊断是一种切实可行的方法。 |
| Abstract | With the rapid development of social economy and constant improvement of people's living station, the air conditioning systems are used more and more widely. They have already become essential. If the faults of the air conditioning systems can't be diagnosed in time, it will cause losses on the energy and damage the equipment, this will break down the normal of production and living. Because the traditional fault diagnoses methods couldn’t meet the actual demand, the domestic and international scholars are researching the applying methods of fault diagnose technology on air conditioning, the most noticeable one of all these methods is the Artificial Neural Network.
The research object of this paper is a chilled water air conditioning system in the HVAC laboratory of USST. While studying the faults diagnoses for the air conditioning system, the artificial neural network is taken as the reasoning machine of the faults diagnoses system. When setting up the neural network model of the faults diagnoses system, the hidden-layers, the neurons of the implied layer, the training function and the learning velocity of the neural network, etc have been discussed respectively in detail. While training the neural network, in order to enable the neural network kind more convergence, the input parameters of the neural network have been pretreated, the fuzzification, discretization and normalization methods have been compared.
In order to make the corresponding relation between trouble and systematic parameter when air conditioner system operating, we set up the electric heater, chilled water pump, cooling pump, etc, six kinds of troubles of air conditioning system to imitate the experiment.
A large number of operation parameters have been obtained and analysed through the trouble experiment of the air conditioning system. Then the trouble sign collections of the air conditioning system have been established. With the visual advantage, we use the Visual Basic to program the software of the faults diagnoses of the air conditioning system. Large amount of test work has been done in reading the calling of MATLAB to VB, at last the ACTIVEX automatic agreement makes it function.
Utilize the trouble sign collections of the air conditioning system to train the neural network, the data that was not trained to validate the software, the software can run very good in faults diagnosis.
Results of study indicate that it is a feasible method to apply the Artificial Neural Network in the diagnosis of air conditioning system. |
