Advanced combinatorial optimization techniques with applications to fault diagnosis and multiuser detection
Date of Completion
January 2003
Keywords
Engineering, Electronics and Electrical
Degree
Ph.D.
Abstract
In this thesis, we develop efficient combinatorial optimization algorithms in the areas of fault diagnosis and multiuser detection in wireless communications. First, we solve multiple fault diagnosis problems (MFD) for systems. In order to alleviate the computational explosion of the current MFD approaches, we present three algorithms to efficiently diagnose multiple faults on the basis of system failure source probabilities, and the dependency matrix (fault dictionary). For systems with AND nodes (denoting redundancies or fault-tolerance), the concept of minimal cut sets is utilized to generate an augmented fault dictionary matrix for diagnosis. We also apply our algorithms developed to the QMR-DT (Quick Medical Reference, Decision-Theoretic) network for medical diagnosis; this network is dense, highly loopy and includes large numbers of diseases and findings. ^ Next, we consider the problem of generating near-optimal test sequences by employing concepts from information theory and rollout strategies. We apply rollout strategies, which can be combined with the one-step or multi-step looka-head heuristic algorithms, to obtain near-optimal solutions in a computationally efficient manner compared to the intractable optimal strategies. ^ Finally, we present a multiuser detection algorithm which employs rollout strategies and achieves significantly lower Joint Error Rate (JER) than its underlying base policy—decision feedback detector. ^ The increasing complexity of modern systems is putting new demands on system maintenance. The potential hazards that may arise from faults within these systems range from system inefficiency, loss of mission, system damage to loss of crew. When the system fails, the job of maintenance is to detect and diagnose the failure causes as rapidly and accurately as possible. Given efficient fault detection and diagnosis, it becomes easier and quicker to apply necessary actions to prevent further damage. The MFD algorithms developed in this research can be used in a variety of applications, including the real-time diagnosis of complex systems and medical diagnosis based on a set of imperfect manifestations (symptoms). The test sequencing algorithm can be applied as an interactive maintenance aid that minimizes the expected testing cost/time by adaptively sequencing tests based on the observed symptoms, available resources (humans and instrumentation) and system state. The rollout algorithm for multiuser detection has the potential to improve the quality of service in wireless communications. ^
Recommended Citation
Tu, Fang, "Advanced combinatorial optimization techniques with applications to fault diagnosis and multiuser detection" (2003). Doctoral Dissertations. AAI3101717.
https://digitalcommons.lib.uconn.edu/dissertations/AAI3101717