【作者】 张曦；

【导师】 邵惠鹤；

【作者基本信息】 上海交通大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 随着工业过程规模的不断扩大和复杂性的日益提高,有效的性能监控、故障诊断和质量预测是保证生产安全、提高产品质量和经济效益的关键。对于复杂的工业过程来说,准确详细的数学模型往往很难得到,即使能够得到,这些理论上的等式也只能描述系统中一部分能量及物料平衡关系,这就限制了基于模型的性能监控方法的应用。另外,随着计算机技术的迅速发展,工业过程中能够测量和处理的变量越来越多,如何从海量数据中挖掘出隐藏的有用信息,从而提高过程运行的安全性和可靠性,已经成为越来越迫切需要解决的问题。统计性能监控就是在这种背景下发展起来的,并且受到了广泛关注。统计性能监控是一种基于多元统计理论的方法。它通过对测量数据的分析和解释,判断过程所处的运行状态,在线监测和识别过程中出现的异常工况,从而指导生产、减小过程故障所造成的损失和提高生产效率。另外,在现代许多企业中,由于技术或经济条件的限制,生产过程中许多与产品质量密切相关的重要参数很难通过传感器在线测量。随着市场竞争的不断加剧和生产工艺的复杂化,这一问题已成为制约生产与产品质量进一步提高的关键性因素。质量预测和估计(软测量技术)就是为了解决这一问题而产生的,质量预测通过一些可以测量的过程变量和其他一些参数,能够在线估计这些无法直接测量的参数和变量,从而为过程控制及管理决策提供支持,为生产过程的综合自动化奠定基础。本文在传统统计性能监控和质量估计方法的基础上,做了不同程度的改进并提出了一些新的监控和质量估计算法。本文的主要工作和贡献有以下几个方面:1.利用核主元分析非线性性能监控的优势,并将相似度分析引入故障诊断领域,提出了基于核主元分析和模式匹配技术相结合的性能监控和故障诊断方法。针对PCA相似度分析存在的问题,对该方法进行了改进。首先利用KPCA方法计算数据的非线性主元,然后计算不同数据集之间的非线性主元相似度;并将主元相似度、非线性主元相似度和基于距离的相似度赋予不同的权值构成综合相似度指标来进行模式匹配。TE过程仿真试验验证了该方法在非线性性能监控和故障诊断中的有效性。2.针对复杂的工业过程,综合核主元分析处理非线性数据的优点和ICA方法提取高维特征空间信息的能力,提出了基于KICA方法的非线性性能监控方法。该方法首先将数据通过非线性核函数映射到高维特征空间,然后在特征空间中进行独立主元分析和计算。通过在特征空间中构建监控统计量和控制置信限,来实现复杂化工和生物过程的监控。非线性数值仿真实例和流化催化裂化(FCCU)过程仿真研究验证了该方法的有效性。3.针对间歇过程监控的特点,将核方法引入到Fisher判别分析中,提出了基于核Fisher判别分析的间歇过程性能监控与故障诊断法。所提出的方法仅利用已获得的数据测量值对过程进行监控,避免了传统MPCA方法对未来测量值的估计,从而提高了间歇过程监控的性能。在线性能监控通过比较核Fisher特征向量之间的欧氏距离来实现,而最优核Fisher判别向量用来鉴别故障类型。与KPCA方法相比,KFDA方法不仅简化了运算,避免了对核主元个数的确定,而且可以通过求解最优核Fisher判别向量来实现故障诊断。青霉素发酵过程仿真应用表明,核Fisher方法比传统的MPCA方法能更及时地监测出过程异常情况,更准确地判断异常发生的原因。4.利用核偏最小二乘(Kernel Partial Least Squares, KPLS)非线性回归的优势,提出了基于KPLS的非线性质量估计和预测方法。首先通过非线性映射将过程数据从低维输入空间映射到高维特征空间,实现变量之间非线性相关关系的线性转化。然后在高维特征空间中利用PLS回归方法进行质量估计和预测。数值仿真实例和实际工业过程数据应用表明KPLS方法能更有效地捕捉变量之间的非线性关系,回归和质量预测效果明显优于线性PLS方法。5.针对质量估计和预测过程中由于仪表错误或过程泄漏等原因造成的过失误差及故障问题,将模式识别中的Fisher判别分析法引入过程显著误差侦破和故障监测,提出了基于Fisher判别分析和核回归的质量监控和估计方法。首先利用Fisher方法对输入数据进行在线监测,若系统运行正常,则用核主元回归方法对质量数据进行预测和估计;否则将存在故障和过失误差的数据剔除并利用贡献图法确定故障原因。实际工业过程数据仿真研究验证了该方法进行故障监测和质量估计的有效性。6.通过分析上海焦化甲醇精馏过程的特点和工艺流程,开发出了一套统计性能监控和质量估计软件,并将其应用到实际生产过程中,取得了良好效果,从而为上海焦化综合信息化平台和先进控制的成功实施奠定了坚实的基础。更多还原

【Abstract】 In real industrial processes, effective performance monitoring and quality prediction are the key to ensure safety, enhance product quality and economy benefit. For the complex industrial process, it is difficult to achieve the accurate mathematical model. Even if it could be achieved, the equations which are predigested can only describe part relationships of energy and mass. These limit the application of methods based on rigorous mathematical model. On the other hand, with the rapid development of computer technology, a large amount of process data have been sampled and collected. How to transform these collected data into valuable information, and mine it deep-level to improve the monitoring performance becomes a challenge issue. It is one of the most active research areas in the field of process control.Statistical performance monitoring is a method based on statistical theory for online fault detection and diagnosis via analysis to the collected data. The information extracted from process data could reflect the operating status at any time, reduce the losses caused by faults and enhance product quality.Because of the limitation of technology and cost, many key variables are difficult to measure via sensors in industrial process. With the market competition become more and more furious, it has become an important factor holding back the product quality further improvement. Quality prediction and estimation (soft sensor) technology has been proposed to solve this problem. It can estimate the variables that are difficult to measure directly. The results can be used for quality control and decision support. It lays a foundation for the integrated automation.In view of the characteristics of continuous and batch industry processes, some improvements of the tradition monitoring and prediction methods have been made, and new statistical monitoring algorithms are also proposed in this thesis.The main results and contributions of this dissertation are stated as follows:1. Using the advantage of kernel component analysis (KPCA) for nonlinear monitoring and introducing the similarity analysis for fault diagnosis, a new performance monitoring and fault diagnosis method based on KPCA and pattern matching is proposed. Aiming at the existing problem in traditional PCA similarity analysis, the method is improved. Nonlinear principal component similarities are firstly calculated. Then the integrated similarity index is proposed through endowing with different weights to PCA similarity, KPCA similarity and distance similarity. Fault diagnosis is performed through pattern matching of different faults. Effectiveness of the proposed method is verified through TE process.2. In view of the complex industrial processes, a nonlinear process monitoring method based on KICA is proposed through integrating the merit of KPCA to deal with nonlinear data and ICA to extract the high-dimensional information. The data are firstly mapped into high-dimensional feature subspace. Then the ICA algorithm is performed. Performance monitoring is implemented through constructing the statistical index and control limit in the feature space. Application results to the FCCU process indicate that the proposed method can effectively capture nonlinear relationship among variables. Its performance significantly outperforms monitoring method based on ICA or KPCA.3. In view of the characteristics of batch process and using the advantage of kernel theory, a novel batch performance monitoring and fault identification strategy based on kernel fisher discriminant analysis (KFDA) is proposed. The approach only uses present data and overcomes pre-estimating the unknown part of process variable trajectory in multi-way PCA (MPCA). The key to the proposed approach is to calculate the distance of block data which are projected to the optimal kernel Fisher discriminant vector between new batch and reference batch. Through comparing distance with the predefined threshold, it can be considered whether the batch is normal or abnormal. Similar degree between the present discriminant vector and the optimal discriminant vector of fault in historical data set is used to perform fault diagnosis. Simulation results on a penicillin fermentation process demonstrate that, in comparison to the MPCA method, the proposed method is more accurate and efficient to detect and diagnose the malfunctions.4. Using the advantage of kernel partial least squares (KPLS) in nonlinear regression, a new quality estimation and prediction method based on KPLS is proposed. The basic idea of the method is to first map data in the original space into high-dimensional feature space via nonlinear kernel and then performs quality estimation and prediction. Application results to a simple example and real data in an industrial oil refinery factory show that the proposed method can effectively capture nonlinear relationship among variables and have better estimation performance than PLS and other linear approaches.5. In view of the gross error caused by sensor faults or process leakage, a novel performance monitoring and quality estimation approach based on fisher discriminant analysis (FDA) and kernel regression is proposed. FDA is first used for quality monitoring.If the process is under normal condition, then kernel regression is further used for quality prediction and estimation. Otherwise, if faults have occurred, contribution plot in the fault feature direction is used for fault diagnosis. The proposed method can effectively detect the happening fault and has better ability to predict the response variables than principle component regression (PCR) and partial least squares (PLS). Application results to the industrial fluid catalytic cracking unit (FCCU) show the effectiveness of the proposed method.6. The model of performance monitoring and quality prediction is built and a software package is developed through analysis to the characteristic and techniques of methanol process. They are successfully applied to the monitoring of methanol process in Shanghai Coking and Chemical Corporation (SCCC). What we have done lay the foundations for the performance of integrated automation and advanced control in SCCC.更多还原