Fuel mass estimation in aircraft tanks using neural nets

Determination of fuel mass in aircraft tanks is a nonlinear estimation problem, in which a set of noisy sensor readings is transformed into a single estimate. In the typical commercial aircraft currently in service, this model-based transformation is approximated by a sum of one-dimensional look-up tables representing contributions from individual fuel quantity probes. Significant improvements in accuracy may be achieved by replacing this decomposed approach with a multi-dimensional optimal estimator In the paper feedforward neural nets are trained to approximate the unknown optimal estimator A simulation example demonstrates the benefits offered by the neural net technique compared with the traditional method. Issues related to certification of safety-critical software containing neural net modules are also addressed