Multi-Auto Associative Neural Network based sensor validation and estimation for aero-engine

Aircraft gas turbine engine, being a complex system, uses a wide sensor network to monitor its performance for control and Engine Health Management (EHM) purposes. Both applications necessitate accurate functioning of all sensors, however due to harsh operating conditions, life and accuracy of sensors is affected. Early detection of drift in measurement or fault in sensors is important as it can help in avoiding false alarms in the EHM system. It is equally important to predict the measurement, that the sensor failed to measure, till the time sensor is replaced. An Auto Associative Neural network (AANN) based sensor validation module is an analytically-redundant sensor network, which provides continuous sensor status information and estimates the measurement value in place of faulty measurements during both online and offline data validation. The number of sensors used to monitor engine are large and it is not viable to monitor all the sensors using a single AANN. Hence in this work a novel approach is adopted for sensor validation and Estimation (SVE) where sensors are grouped into smaller sets based on their location and physical relationships between them. By breaking network into smaller groups dual benefit is achieved; first it reduces complexity arising from higher dimensionality, secondly it ensures multiple-validation of each sensor through various networks. The network is trained using data generated from a validated twin spool turbojet engine simulation model. Presented approach is validated through a simplified experiment and results show prompt fault identification and prediction of sensor value with satisfactory accuracy.