随着工业生产效率、节能环保等工业需求的日益增长，规模庞大且结构复杂的工业过程日益增多。这些过程一旦发生故障会造成巨大的经济损失并对生产安全形成威胁，因此利用过程监控技术来提高工业过程的可靠性和安全性具有重要意义。本文研究的主要问题是如何提高工业过程故障监控的准确性和快速性。围绕这一问题，本文重点研究了基于数据驱动技术和因果分析的故障监控方法。全文的主要内容和贡献如下：为了能够有效增加故障识别的准确性，本文提出了一种基于规范变量分析(CVA)的贡献图方法，进一步根据监控统计量分别在规范变量空间和残差空间中的变化情况定义了两种故障贡献值。据此，将故障变量划分为状态空间属性故障变量和残差空间属性故障变量，可以有助于加深对工业故障特性的理解。理论分析和案例研究表明结合使用这两种故障贡献值能够更加有效地进行故障识别。为了进一步诊断引起故障的根原因和提高故障诊断的准确率和快速性，本文还提出了一种综合CVA和费舍尔判别分析(FDA)的故障诊断方法——CVA-FDA。该方法先利用融入数据重叠度因素的改进CVA算法对数据进行预处理，然后再用FDA算法进行故障诊断，并进一步研究了在CVA降维变换中如何选取最优的时滞阶次l和CVA降维阶次a_CVA，使得经过预处理后的数据在后续的FDA诊断故障中更充分地利用数据间的动态信息并降低计算量。为了适用于不能事先获知故障数据集情况的应用需求，同时考虑过程关联结构故障，本文提出了一种综合数据驱动技术CVA和因果分析的故障监控方法。另外，本文进一步分析了因果信息在特征表示以及故障监控尤其是故障溯源方面的重要作用。理论分析和案例研究表明所提出的基于因果依赖(CD)特征表示的CVA监控方法能够有效提高故障监控的性能，尤其对于多故障类型的情形而言。本文还提出了一种闭环辨识方法以及基于此的因果分析方法——Interleaved Data Pair Upper Diagonal (IDPUD)算法，只需要经过一步矩阵UD分解就能够同时辨识所有变量通道的模型（包括模型阶次和模型系数）。该方法既可以提高计算效率，又可以同时确定变量的因果关系以及具有因果关系变量之间的通道模型。这一算法为解决综合数据驱动技术与因果分析监控方法需要事先已知过程因果信息的问题提供了一个良好的解决途径。

With the increasing demands of production efficiency, energy-saving as well as environmental protection, many industrial processes with large-scale and complicated structures emerge recently. The industries will face tremendous economic loss and security threats when faults occur. Therefore, it is meaningful to improve the reliability and safety of industrial processes by the utilization of process monitoring techniques. The problem how to improve the accuracy and efficiency of fault monitoring is studied in this dissertation, where the monitoring approaches based on data driven techniques and causal analysis are investigated. The main contributions of the dissertation are as follows:To improve the accuracy of fault identification, a canonical variate analysis (CVA)-based contribution method is proposed for the identification of variables most closely associated with a fault, where two types of contributions are developed based on the variations in the canonical state space and in the residual space. The two contributions are used to categorize faulty variables into state-space faulty variables and residual-space faulty variables, which enhances the understanding of the character of each fault. Theoretical analysis and case studies demonstrate that the combined utilization of the two contributions can improve the performance of identifying faults.To improve the accuracy and efficiency of fault diagnosis, a combined CVA and FDA approach (denoted as CVA-FDA) for fault diagnosis is proposed, which employs CVA for pretreating the data in the first step and subsequently utilizes FDA for classifying faults. In addition, an optimization technique based on the overlapping degrees of training and validation data is utilized to select an optimized time lag l and CVA dimensionality reduction order aCVA in the CVA step, which enables the pretreated data to provide the second step of FDA for fault diagnosis with (i) more dynamical information in the data handled and (ii) decreased computational burden.To satisfy the condition that fault data can not be obtained in advance, as well as to improve the current research shortage on the monitoring of process structural faults which occur frequently in industrial processes, a data-driven approach (CVA) in conjunction with the causal analysis for the monitoring of faults associated with changes in process structures is proposed. In addition, the importance of causal information to the feature representation as well as to the fault monitoring particularly for the determination of the root causes is studied. Theoretical analysis and case studies demonstrate that the proposed CVA approach based on the feature representation of CD can improve the monitoring performance, especially for the monitoring of multiple faults.In order to solve the problem that the hybrid monitoring methods of data driven techniques and causal analysis often assume the causal connectivity of process variables to be known a priori, a causalty analysis method based on closed-loop identification, named Interleaved Data Pair Upper Diagonal (IDPUD) algorithm, is proposed. After a single step of UD factorization, the proposed IDPUD algorithm can simultaneously estimation of all path models (including model orders and parameters), which favors the causalty analysis with two advantages: (i) lower computational cost; and (ii) simultaneously determine the causal relationship among process variables as well as the model between causally related variables.