(268d) Fault Detection and Identification in Petrochemical Systems Using Unknown Input Observers and Input Reconstruction
AIChE Annual Meeting
2009
2009 Annual Meeting
Computing and Systems Technology Division
Process Monitoring and Fault Detection
Tuesday, November 10, 2009 - 1:30pm to 1:50pm
Over the past few decades, monitoring the operation of complex chemical plants has become a challenging task. Stringent product specifications, high throughput, long treatment chains and the typically high number of reflux streams reduce the time available to respond to sudden and undesired operating conditions. Additionally, these properties hamper the assessment of such conditions and, more specifically, the determination of their root causes. Consequently, this can, in addition to the usually high workload of operators as well as the high level of automation, interfere with the acquisition of reliable expert knowledge to troubleshoot such anomalous events. Thus, for experienced plant operators to successfully detect abnormal operation conditions and diagnose their root causes requires support through intelligent monitoring software. The desired properties for such automated fault detection and identification (FDI) schemes are therefore (i) to have a small detection delay, (ii) to prevent false alarms and (iii) to offer a detailed analysis of the fault propagation to correctly diagnose an abnormal event.
This work proposes the concept of combining the model-based process control (MBPC) and multivariate statistical process control (MSPC) frameworks. Yoon and MacGregor (2000) discussed the potential that such a combined approach may offer. The research literature, however, has not yet offered a substantial treatment of such a combined approach. The MSPC methodology can handle larger numbers of highly correlated variables, which typically occur in large-scale industrial plants. Consequently, embedding causal model-based approaches into the multivariate statistical framework has the potential to address some of the difficulties that conventional MSPC may run into including the handling auto- and cross-correlation between and among recorded process variables (Xie et al. 2006). In contrast, MBPC approaches rely on a state space representation that explicitly describes the inherent dynamics encapsulated within these recorded variables.
The difficulty to obtain the required state space model representation of such complex systems, however, increases substantially with the complexity of the process. Besides the immense costs of obtaining a rigorous model for large plants, it is not always desirable for control or process monitoring purposes to describe the dynamic behaviour in every detail and for the entire range of operation (Van Overschee and De Moor 1996). Given that most industrial processes in the petrochemical industry operate around a small selection of operating regions, the incorporation of subspace model identification (SMI) alleviates this problem. Based on well-known statistical tools such as canonical variate analysis, accurate process models can be identified from periods of normal operation condition (NOC), which are suitable for the application in a FDI scheme.
The use of MBPC as a process-monitoring tool is accompanied by the application of state observers (Luenberger 1971). Several schemes have been introduced to fulfil the task of fault detection and identification using state observers, including generalized observers, or a bank of observers (Isermann 2006). To address the robustness of the state estimates to unmeasured disturbances and plant or parameter uncertainties, the concept of unknown input observers (UIO) has been introduced by Bhattacharyya (1978) and further developed in the research literature (Kudva et al. 1980; Watanabe and Himmelblau 1982; Kurek 1983; Yang and Wilde 1988; Guan and Saif 1991; Darouach et al. 1994; Hou and Müller 1994; Chen et al. 1996; Valcher 1999; Simani et al. 2003; Xiong and Saif 2003; Hui and Zak 2005; Barbot et al. 2007; Fang et al. 2008) to decouple state estimation from unknown or uncertain inputs. For sensor fault detection, Saif and Guan (1993) introduced an augmented state space model which includes a dynamic term to consider such an incident as an unknown model input. More precisely, for detecting sensor faults, the robust state estimation of the augmented system provides a direct estimate of the fault signature. Refinements of this concept involving the reconstruction of the process inputs using output estimates to detect process faults are proposed in Xiong and Saif (2003).
The presentation introduces the combined approach involving MBPC and MSPC, shows how identified state space models and a multivariate statistical analysis of model residuals can be applied to the MBPC approach, how the utilization of observers and the fault reconstruction work can enhance the fault detection and diagnosis capability of MSPC. The utility of this combined approach is demonstrated through a total of three application studies that consists of industrial data sets. These processes include an industrial furnace of a catalytic reforming unit, an industrial distillation unit and a reactive distillation unit. The application studies highlight that the proposed framework of SMI, observer-based residual generation and statistical residual analysis can be used to detect different sensor, input and process faults.
LITERATURE
Barbot, J. P., M. Fliess, et al. (2007). An algebraic framework for the design of nonlinear observers with unknown inputs. Decision and Control, 2007 46th IEEE Conference on.
Bhattacharyya, S. (1978). "Observer design for linear systems with unknown inputs." Automatic Control, IEEE Transactions on 23(3): 483-484.
Chen, J., R. J. Patton, et al. (1996). "Design of unknown input observers and robust fault detection filters." International Journal of Control 63(1): 85 - 105.
Darouach, M., M. Zasadzinski, et al. (1994). "Full-order observers for linear systems with unknown inputs." Automatic Control, IEEE Transactions on 39(3): 606-609.
Guan, Y. and M. Saif (1991). "A novel approach to the design of unknown input observers." Automatic Control, IEEE Transactions on 36(5): 632-635.
Hou, M. and P. C. Müller (1994). "Fault detection and isolation observers." International Journal of Control 60(5): 827-846.
Huazhen, F., S. Yang, et al. (2008). A new algorithm for simultaneous input and state estimation. American Control Conference, 2008.
Hui, S. and S. H. Zak (2005). "Observer design for systems with unknown inputs." International Journal of Applied Mathematics and Computer Science 15(4): 431-446.
Isermann, R. (2006). Fault-Diagnosis Systems - An Introduction from Fault Detection to Fault Tolerance, Springer Verlag GmbH.
Kudva, P., N. Viswanadham, et al. (1980). "Observers for linear systems with unknown inputs." Automatic Control, IEEE Transactions on 25(1): 113-115.
Kurek, J. (1983). "The state vector reconstruction for linear systems with unknown inputs." Automatic Control, IEEE Transactions on 28(12): 1120-1122.
Luenberger, D. (1971). "An introduction to observers." Automatic Control, IEEE Transactions on 16(6): 596-602.
Saif, M. and Y. Guan (1993). "A new approach to robust fault detection and identification." Aerospace and Electronic Systems, IEEE Transactions on 29(3): 685-695.
Simani, S., R. Patton, et al. (2003). Model-Based Fault Diagnosis in Dynamic Systems Using Identification Techniques, Springer-Verlag New York, Inc.
Valcher, M. E. (1999). "State observers for discrete-time linear systems with unknown inputs." Automatic Control, IEEE Transactions on 44(2): 397-401.
Van Overschee, P. and B. De Moor (1996). Subspace identification for linear systems: Theory, Implementation, Applications, Kluwer Academic Publishers.
Watanabe, K. and D. M. Himmelblau (1982). "Instrument fault detection in systems with uncertainties." International Journal of Systems Science 13(2): 137 - 158.
Xie, L., U. Kruger, et al. (2006). "Statistical Monitoring of Dynamic Multivariate Processes Part 1: Modeling Autocorrelation and Cross-correlation." Industrial and Engineering Chemistry Research 45(5): 1659-1676.
Xiong, Y. and M. Saif (2003). "Unknown disturbance inputs estimation based on a state functional observer design." Automatica 39(8): 1389-1398.
Yang, F. and R. W. Wilde (1988). "Observers for linear systems with unknown inputs." Automatic Control, IEEE Transactions on 33(7): 677-681.
Yoon, S. and J. F. MacGregor (2000). "Statistical and causal model-based approaches to fault detection and isolation." AIChE Journal 46(9): 1813-1824.