Neural network algorithm for evaluating wind velocity from pressure measurements performed on a train's surface

Starting from the knowledge of flush air data-sensing systems developed in the aerospace field, a system able to evaluate the crosswind velocity (absolute and angle) from pressure measurements performed on a train's surface has been studied and developed. The crosswind evaluation algorithm is based on a neural network and is created and tested using pressure measurements carried out during wind tunnel tests on a scale-model of a railway vehicle. The number and positions of pressure taps on the surface of the train's leading car are optimized, in terms of the sensitivity and robustness of the measurement system. The system is calibrated and validated using steady and quasi-steady wind tunnel tests on a scale model of a train. The designed neural network gives a maximum error of about 1 m/s for the modulus and 1° for the angle of the relative velocity.