Neural-Network-Based Sensor Data Fusion for Multi-Hole Fluid Velocity Probes

For measuring three components of velocity in unknown flow fields, multi-hole pressure probes possess a significant advantage. Unlike methods such as hot-wire anemometry, laser-Doppler velocimetry and particle-image velocimetry, multi-hole pressure probes can provide not only the three components of local velocity, but also static and stagnation pressures. However, multi-hole probes do require exhaustive calibration. The traditional technique for calibrating these probes is based on either look-up tables or polynomial curve fitting, but with the low cost and easy availability of powerful computing resources, neural networks are increasingly being used. Here, we explore the possibility to further reduce measurement uncertainty by implementing neural-network-based methods that have not been previously used for probe calibration, including supervised and unsupervised learning neural networks, regression models and elastic-map methods. We demonstrate that calibrating probes in this way can reduce the uncertainty in flow angularity by as much as 50% compared to conventional techniques.