Design of Sensor Network Based on the Signed Directed Graph of the Process for Efficient Fault Diagnosis

Abstract

An optimally located network of sensors is a prerequisite for successful application of fault diagnosis techniques. Most of the previous work in the area of fault diagnosis deals with methodologies for identifying possible faults, given sensor data. Available literature suggests that very little work has been done on methods for optimally locating the sensors for efficient fault diagnosis. Some algorithms based on the concepts of observability and resolution were discussed in our previous work (ref 1: Raghuraj et al. AIChE J. 1999, 45 (2), 310). These algorithms are based on a digraph (DG) representation of the process. In this article, the sensor location work is extended to use the signed directed graph (SDG) representation of the process. Various issues involved in utilizing the SDG of the process for the problem of sensor location are discussed. Algorithms for sensor network design based on the SDG of the process are detailed and applied to two chemical engineering case studies. It is shown that better design can be obtained by using the SDG instead of the DG.

Integration of process design and control: A review

“…They used the graphbased method of maximum weight matching for determining the best control structure. Similarly, Bhushan and Rengaswamy (2000) applied signed directed graphs for designing sensor networks. The disadvantage of signed directed graphs is that they cannot describe time propagation properties.…”

Realizability Constrained Selection of Residual Generators for Fault Diagnosis With an Automotive Engine Application

“…See for example [32], [13], [33], [34], [35] which all propose different methods for placing sensors to achieve a certain fault isolability requirement, similar to the objective in Section IV-B. Other works, e.g., [36] propose techniques to find optimal sensor locations.…”

Robust sensor network design for fault diagnosis

“…It is the set of variables affected by that particular fault. Placing a sensor on any one of the variables in the faultset will ensure that the corresponding fault is observed (Raghuraj et al, 1999;Bhushan & Rengaswamy, 2000b). For fault resolution (or diagnosis), the same idea is extended and faultsets are generated for fictitious fault nodes that represent the ability to resolve between faults pairwise.…”

“…For fault resolution (or diagnosis), the same idea is extended and faultsets are generated for fictitious fault nodes that represent the ability to resolve between faults pairwise. Placing a sensor on any one of the variables in these faultsets would distinguish between the two faults that are represented by the fictitious fault node (Raghuraj et al, 1999;Bhushan & Rengaswamy, 2000b). The solution to both the observability and resolution problems is the well-known set cover problem.…”

“…Qualitative models such as fault trees (Lambert, 1977), (signed) directed graphs (Bhushan & Rengaswamy, 2000b;Raghuraj, Bhushan, & Rengaswamy, 1999), structural relationship models (Chang, Mah, & Tsai, 1993), detailed mathematical models (Alpay & Shor, 2000;Patton, Frank, & Clark, 1989;Watanabe, Sasaki, & Himmelblau, 1985), and data based models such as principal component analysis (Musulin, Bagajewicz, Puigjaner, 2003;Wang, Song, & Li, 2002;, have been used. In our previous work on sensor network design based on digraph (Raghuraj et al, 1999) and signed digraph (Bhushan & Rengaswamy, 2000b) representations, notions of fault observability and resolution were used to select optimal sensors. Later, optimization formulations incorporating various constraints on the sensor network design problem, as well as utilizing quantitative information such as fault occurrence and sensor failure probabilities, and cost constraints for sensor location were presented (Bhushan & Rengaswamy, 2002a, 2002b.…”