【作者】 李畅；

【导师】 张颖伟；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2011， 硕士

【摘要】 工业过程的故障检测是流程工业监控技术的重要组成部分,它通过密切地监督生产过程的运行状态,不断地检测工业系统在生产过程中状态变化和异常信息,从而及时有效地识别出系统发生故障,并对故障的成因、幅值等因素进行预估,以提高了产品质量和经济效益。而基于数据驱动的故障检测技术正是目前流程工业中过程控制领域的研究热点之一,也是本文的研究重点。本论文的主要研究工作体现在以下几个方面：首先,论文以故障检测为研究目标,讨论了数据标准化处理的方法,重点针对偏最小二乘法、核偏最小二乘法、独立元分析法等数据处理方法进行分析。其次,论文以主元分析(PCA)方法为主线,引入了平方预测误差概念来构建基于变量贡献的图像,为工业过程的线性故障提供可靠的辨识方案。再次,论文在对多尺度核偏最小二乘法分析与研究的基础上,提出了基于多尺度核偏最小二乘法的在线测量模型,并利用该模型进行实时工业过程故障检测。最后,论文利用二次型混合指标来考察故障的幅值情况,并且提出了新的工业过程故障幅值估计算法,针对不同工业过程对象的特点,开展了具体的故障幅值计算。论文所提出的故障检测方案和算法都是针对目前已有的故障检测算法所存在的不足而提出来的,在理论意义上具有各自的优点。论文除了对具体的故障检测算法进行研究外,还对相关的算法和方案进行了算法实现和仿真实验。通过对仿真结果的分析,归纳和总结了工业过程故障检测的特点和优势,证明了本文所提出的算法的有效性和先进性。更多还原

【Abstract】 The fault detection of industrial process is an important part of the monitoring technology in the process industry, which closely monitors the operation state of the production process, constantly tests the state change and abnormal information of the system in the production process, thus to identify the system failure timely and effectively, and estimates the fault causes, magnitude and other factors expected to improve product quality and economic benefits. The data-driven based fault detection technology is one of the current hot research areas of the process control technology in the process industry, and it is also the focus of this study.The main work of the thesis includes:First, this dissertation takes the fault detection as the research targets, discusses the standardization of data processing methods, focuses on partial least squares, kernel partial least squares, independent component analysis and other data processing methods.Second, the thesis takes the principal component analysis (PCA) method as the main line, and builds the variable contribution based images by introducing the squared prediction error concept. This method could provide a reliable identification program for the linear failure of industrial processes.Third, based on the analysis and research of the multi-scale kernel partial least squares, we propose multi-scale kernel partial least squares based online measurement model, and use the model in real-time industrial process fault detection.Finally, the thesis examines the magnitude of fault by the second type of mixed indicators, and proposes a new fault magnitude estimation algorithm of industrial process. The specific calculations of the amplitude of the failure are carried out for the different characteristics of the industrial process objects.The proposed fault detection scheme and algorithms are available for the deficiencies of current fault detection algorithm, and have their own advantages in the theoretical sense. Excepting the research of the specific fault detection algorithms, the thesis also focuses on the implementation and simulation of the algorithms and programs. Through the analysis of simulation results, and summarized the industrial process fault detection features and advantages, this thesis demonstrates the effectiveness and advanced of the proposed algorithms.更多还原