基于规范变量分析的化工过程故障诊断方法研究

发布时间：2018-04-13 21:44

本文选题：故障诊断 + 规范变量分析　；参考：《兰州理工大学》2015年硕士论文

【摘要】：随着科技的不断发展,化工过程趋于复杂化。研究化工过程的故障诊断算法对保证安全生产、避免生产设备损毁、提高产品质量发挥了关键性作用。国内外学者在故障诊断领域研究改进了众多算法以求达到快速准确的故障诊断效果。实际的化工过程一般具有复杂性高、规模庞大的特点,化工过程通常都会采集到丰富的过程变量数据,而基于数据驱动的故障诊断算法直接从化工过程中采集到的数据入手,对数据进行分析,实现对过程的故障诊断,因此深入研究基于数据驱动的故障诊断算法,不仅具有重要的理论意义,在实际生产中也具有广阔的应用前景。本文主要对规范变量分析(CVA)算法进行研究,并对该算法在故障诊断应用中存在的问题做进一步改进,主要做了以下几个方面的工作:1.针对规范变量分析(CVA)算法不能较好地处理化工过程高维数据的降维问题,提出了一种基于等距映射(ISOMAP)的化工过程故障诊断CVA算法,该算法首先采用流形学习中的ISOMAP算法,完成对初始数据的非线性降维,同时保持了数据内部的几何结构,然后基于从高维数据中提取出来的低维数据,利用规范变量分析(CVA)得出过程状态向量和监控统计量,通过TE过程仿真验证了该算法用于化工过程故障诊断的合理性。2.针对过程数据通常具有时变性,规范变量分析(CVA)在动态过程系统的故障诊断中不能得到较好的故障诊断准确率,提出了一种基于滑动窗的规范变量分析(MWCVA)算法,该算法首先建立初始的CVA模型和计算监控统计量,通过滑动窗更新过程变量数据,计算更新建模所需数据,不断实时的更新出新样本的CVA模型和监控统计量,通过TE过程仿真验证了该算法的优越性。3.针对化工过程中,往往生产多种产品,使得过程存在多个操作模态,传统的单模态故障诊断算法在用于诊断多模态化工过程时不再适用。因此,针对多模态化工过程中多高斯分布的采样数据,本文提出了一种基于高斯混合模型的核规范变量分析算法。首先采用高斯混合模型(GMM)将化工过程中的采样历史数据分类成多个高斯分量,然后利用核规范变量分析(KCVA)算法对每个高斯分量建模,计算其相应的统计量进行过程故障诊断,通过TE过程仿真验证了该算法的适用性。
[Abstract]:With the development of science and technology, the chemical process tends to be complicated.The research on the fault diagnosis algorithm of chemical process plays a key role in ensuring the safety of production, avoiding the damage of production equipment and improving the product quality.Scholars at home and abroad have improved many algorithms in the field of fault diagnosis in order to achieve rapid and accurate fault diagnosis results.The actual chemical process generally has the characteristics of high complexity and large scale. The chemical process usually collects abundant process variable data, and the data-driven fault diagnosis algorithm starts directly from the data collected in the chemical process.To analyze the data and realize the fault diagnosis of the process, it is not only of great theoretical significance to study the data-driven fault diagnosis algorithm, but also has a broad application prospect in practical production.In this paper, the canonical variable analysis (CVA) algorithm is studied, and the existing problems in fault diagnosis are further improved. The following work is mainly done: 1.In view of the fact that the canonical variable analysis (CVA) algorithm can not deal with the dimensionality reduction problem of high dimensional data in chemical process, a novel CVA algorithm for chemical process fault diagnosis based on isometric mapping is proposed. The ISOMAP algorithm of manifold learning is used in this algorithm.The nonlinear dimensionality reduction of the initial data is completed, while the geometric structure of the data is maintained. Based on the low-dimensional data extracted from the high-dimensional data, the process state vector and the monitoring statistics are obtained by using the canonical variable analysis (CVA).The rationality of this algorithm in chemical process fault diagnosis is verified by te process simulation.In view of the fact that the process data is usually time-varying, the canonical variable analysis (CVA) can not get better accuracy in the fault diagnosis of dynamic process system, a new method based on sliding window (Slide window) for canonical variable analysis (CVA) is proposed.The algorithm firstly establishes the initial CVA model and calculates the monitoring statistics. The process variable data is updated by sliding window and the data needed to update the modeling is calculated. The CVA model and monitoring statistics of the new samples are constantly updated in real time.The superiority of the algorithm is verified by te process simulation.In chemical process, many kinds of products are often produced, so there are many operating modes in the process. The traditional single-mode fault diagnosis algorithm is no longer suitable for the diagnosis of multi-modal chemical process.Therefore, in view of the sampling data of multi- distribution in multi-modal chemical process, this paper presents a new kernel canonical variable analysis algorithm based on Gao Si mixed model.Firstly, Gao Si mixed model (GMMM) was used to classify the sampled historical data in chemical process into several Gao Si components, and then the KCVA algorithm was used to model each Gao Si component and calculate the corresponding statistics for process fault diagnosis.The applicability of the algorithm is verified by te process simulation.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TQ050.7

【相似文献】