Discrete event development framework for highly reliable sensor fusion systems

• Main Author: Rokonuzzaman, Mohd., 1965-
• Other Authors: Memorial University of Newfoundland. Faculty of Engineering and Applied Science
• Format: Thesis
• Language: English
• Published: 1999
• Subjects: Fault tolerance (Engineering); Multisensor data fusion; Petri nets; Newfoundland studies; University of Newfoundland
• Online Access: http://collections.mun.ca/cdm/ref/collection/theses5/id/5380

Thesis (Ph.D.)--Memorial University of Newfoundland, 1999. Engineering and Applied Science Bibliography: leaves 131-137. Intelligent Systems are being deployed increasingly in safety and mission critical applications. This thesis has synthesized a novel engineering methodology for developing highly reliable sensor fusion systems (SFS) of multi-sensori intelligent systems for the applications in the safety and mission critical environments. This methodology includes both the avoidance of faults during the development phase and the tolerance of sensor failures during the operation phase. Petri net based novel discrete event framework has been proposed to model SFS as discrete event dynamic system. This intuitive mathematical framework abstracts the SFS as a hierarchically finite state machine. The intuitive graphical nature of this framework has the potential to enhance the communication between the developer and the client to capture sensing requirements resulting in avoidance of requirement errors. The mathematical attribute enables the developer to analyze different attributes of the modeled SFS to ensure logical and temporal correctness of the performance of the system. This proposed discrete event framework has been verified by simulating the design of an example sensor fusion system. The reasoning basis of the architecture of the underlying computing system from this Petri net model of the SFS has also been developed to ensure the temporal correctness during the operation phase. The use of redundancy to tolerate failure of sensors has been experimentally verified. Overheads have been identified to incorporate hardware fault-tolerance in this proposed SFS framework to tolerate sensor faults during the operation phase. A novel scheme has been developed to manage these overheads in a predictable manner. A fault-tree based novel scheme has been proposed to measure the probability of failure of different levels of fusion due to the failure of different sensors. A computationally simple scheme to detect transients present on the sensor data stream has been proposed with extensive simulation results to enhance system performance in operation phase. The loss of time sensitive data during the fault clearance intervals compromises the effectiveness of fault-tolerance in the SFS. A parallel sensing based novel scheme has been proposed to restore sensor data lost during the fault clearance intervals. The effectiveness of this proposed scheme has been experimentally verified by restoring data lost during fault clearance intervals of a triple modular redundant optical sensor.