Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.317827
Title: Signal processing in local sensor validation
Author: Yung, Sheung Kai
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 1992
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
Sensor integrity plays a crucial role in automatic control and system monitoring, both in achieving performance and guaranteeing safety. Conventional approaches in sensor failure detection demand precise process models and abundant central computing power. This thesis describes the development and the evaluation of a novel local sensor validation scheme which is independent of the underlying process and is applicable to a wide variety of sensors. A signal-based in-situ sensor validation scheme is proposed. Typical sensor failures are classified according to their signal patterns. To avoid the ambiguity between genuine failures and legitimate measurand variations, a pair of decomposition filters are designed to partition the sensor output; and attention is focused on characteristics beyond the measurement signal bandwidth, which is the only essential process-related variable required. In addition, the application of decimating filters is explored, both as a relief to the analog anti-aliasing filter and as an enhancement in signal discretization. An expression is derived relating the oversampling rate and the attainable improvement in signal resolution. Based on a period of failure-free observation, a whitening filter is identified by modelling the decomposed sensor signal as a stochastic time-series. Significant progress is achieved by a deliberate injection of bandlimited random noise to ensure signal stationarity and to avoid inadmissible leakage of measurement signal into the innovation sequence. The adopted failure detection strategy is primarily innovation-based. Pertinent sensor signal information is extracted recursively by a collection of efficient and robust signal processing algorithms. Its validity is continuously monitored by statistical tests on which a series of precursory failure alarms are formulated. Any aberration detected is then diagnosed under the supervision of a simple rule-based system. The practicality, efficacy and flexibility of the proposed scheme are successfully demonstrated by a bench-top thermocouple experiment and extensive synthetic simulations.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.317827  DOI: Not available
Keywords: Control systems & control theory
Share: