Estimating instrumentation data acquired during flight test from a helicopter engine using predictor models

Flight test of a helicopter is the most dangerous phase of its development project. Any minor mistake or miscalculation during a test could lead to a catastrophic accident. In order to protect the health of a helicopter during a test, lots of sensors are placed on critical parts and equipment of the helicopter. Data acquired from these sensors are monitored by the experts from different fields through the test and in any kind of unexpected or out of limit data, these experts interfere with the test procedure. Acquiring miscalibrated sensor data or even losing the sensor completely would increase the potential risk and delay the test although there is no problem on helicopter. Also having miscalibrated or no data from a sensor makes a part of the helicopter invisible for analysis. Due to all these reasons, we propose a hybrid system composed of decision tree that calculates the noise ratio of the sensor readings and neural network that replaces miscalibrated ones with accurate predictions based on the other sensor readings. In this paper, we present the models and how we construct them, test them on semi-synthetic data and show that these models can be used in production and testing systems in real life.