Aumento da confiabilidade de Air Data Systems através do uso de sensores de fluxo térmico e votação de sinais baseada em detecção de novidades

Abstract

The measurement of air parameters and airspeed in aeronautical applications has a high criticality because failures in the Air Data Systems (ADS) for such parameters can cause the death of hundreds of people, as was the case of Air France flight 447 with 228 fatalities. To meet this demand, the present work seeks to increase the reliability of Air Data Systems by exploring two aspects: the use of thermal flow sensors to measure airspeed and the development of a new redundant signal voter to be used in the ADS. The use of thermal flow sensors is studied by integrating the model of such sensors into a simplified ADS model with 4-way redundancy and comparing several established simple voting techniques. The new signal voting proposal uses novelty detection based on pattern recognition and machine learning. The performance of the proposed voter is compared to a standard voter in simulations of an ADS using actual input data. The results indicate that for thermal flow sensorss characteristic failures the ADS performs best when using a simple algorithm based on the calculation of the output as the average of the nearest redundant inputs that respect a defined interval. In relation to the proposed voter with novelty detection the results indicate that it performs better than the standard voter for cases in which the failure is constant, increasing or related to the derivative of the measured value. For cases in which the failure has behavior proportional to the measured value the proposed voter has inferior performance. The results obtained are a step toward improving the reliability of Air Data Systems and show applicability of pattern recognition for fault identification, as well as pave the way for other related research.