A general computational framework for distributed sensing and fault-tolerant sensor integration

This paper proposes an abstract framework to address the problem of fault-tolerant integration of information provided by multiple sensors. Extending our earlier results[lO], this paper presents a formal description of spatially distributed sensor networks, where i) clusters of sensors monitor (possible overlapping) regions of the environment; ii) sensors return measured values of a multi-dimensional parameter of interest; and iii) uncertainities associated with a sensor output are represented by a connected subset in the parameter space. A method to obtain interval estimates of components of the actual parameter vector is developed, wherein information from faulty sensors (whose uncertainty subspaces do not contain the true value of the parameter) are filtered out. The problem addressed involves combining interval estimates of sensor outputs into a best intersection estimate of outputs. The sensor fault model used assumes most faults are “tame.” That is, faults cluster in the neighborhood of the correct values. The procedure of this paper, assuming this fault model, is superior to earlier work by M a r d o . To test the theoretical analysis of the computational framework proposed in our paper, we have developed a modular parameter-driven simulator SIMDSN for the fault-tolerant integration of abstract sensor interval estimates. The simulator uses the well-known Monte-Carlo technique to generate random correct and tamely faulty intervals. The results of our algorithm and Manullo’s are compared. We conclude based on the simulation that the union of the disjoint intervals (if any are disjoint) determined by our algorithm and the gaps between these intervals together can never exceed the width of the final output of Marzullo’s method. Thus, while Marzullo’s method specilies a rather large interval as the final output of the integration process, our method subdivides this interval into (possibly) disjoint intervals each with an associated reliability measure.