基于人工神经网络的TE过程故障诊断与检测研究

发布时间：2018-08-28 15:29

【摘要】：随着现代工业技术的迅猛发展,生产过程越来越体现出连续、大批量、非线性、强耦合、时变、参数不确定等特性,故障发生的概率也不断增加,这导致现有的故障诊断方法无法适应对过程监控的要求,迫切需要发展和完善新的故障诊断方法和应对策略以保证生产过程的稳定运行。本文主要研究基于人工神经网络理论的复杂工业过程故障诊断方法,所完成的主要工作如下: 1.对目前工业过程故障诊断技术的各种典型方法的研究现状及发展趋势进行了综述,并对各种方法的优缺点和适用对象进行了分析。 2.研究了基于BP神经网络和概率神经网络的故障诊断方法。对于常规的BP算法收敛速度慢、目标函数容易陷入局部最小的问题,通过引入动量因子、学习率可变的BP算法,改进了算法性能;在概率神经网络方法中,径向基函数的分布密度SPREAD的确定影响着网络的性能,通过对不同SPREAD值的仿真结果对比分析确定适合的SPREAD值,可得到较好的诊断效果。 3.基于PCA的神经网络改进故障诊断算法研究。针对网络输入样本过多会导致诊断结果不理想的现象,本文提出先用PCA处理样本数据,降低样本数据的维数,然后将处理后的样本数据输入到网络中,减小了网络模型的规模,加快了网络的运算速度,而且降低了误报率和漏检率,提高了神经网络方法的诊断准确性。全文算法采用TE过程为仿真平台进行算法的有效性验证。
[Abstract]:With the rapid development of modern industrial technology, the production process is more and more continuous, mass, nonlinear, strong coupling, time-varying, parameter uncertainty and other characteristics, the probability of failure is also increasing. As a result, the existing fault diagnosis methods can not meet the requirements of process monitoring. It is urgent to develop and improve new fault diagnosis methods and countermeasures to ensure the stable operation of the production process. This paper mainly studies the fault diagnosis method of complex industrial process based on artificial neural network theory. The main work is as follows: 1. This paper summarizes the research status and development trend of various typical methods of industrial process fault diagnosis technology, and analyzes the advantages, disadvantages and applicable objects of these methods. 2. The fault diagnosis method based on BP neural network and probabilistic neural network is studied. For the conventional BP algorithm, the convergence speed is slow, and the objective function is easy to fall into the problem of local minimum. By introducing momentum factor and variable learning rate of BP algorithm, the performance of the algorithm is improved. The determination of radial basis function distribution density (SPREAD) affects the performance of the network. By comparing and analyzing the simulation results of different SPREAD values, the suitable SPREAD value can be determined, and a better diagnostic effect can be obtained. Research on improved Fault diagnosis algorithm based on PCA Neural Network. In view of the phenomenon that too many network input samples will lead to unsatisfactory diagnosis results, this paper proposes to use PCA to process the sample data, reduce the dimension of the sample data, and then input the processed sample data into the network to reduce the scale of the network model. It speeds up the operation speed of the network, and reduces the false alarm rate and the false detection rate, and improves the diagnostic accuracy of the neural network method. The full-text algorithm uses TE process as the simulation platform to verify the validity of the algorithm.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3;TP183

【相似文献】