Vibration detection method aiming at minimizing defects in manufactured products or problematic conditions during production

Abstract

The invention relates to an innovative method designed for detecting vibrations ( chatter phenomenon) in machining operations, combining the extraction of specific frequency characteristics from signal analysis, feeding these last into an Deep Neural Ordinary Differential Equation Networks (Deep Neural ODEs) architecture, and training the model using a pioneering cost function (Dynamic StabilityLoss), in parallel to the dynamic integration of new samples monitoring aimed at early notification in case of vibration detection. Said method employs advanced techniques to enrich the data with critical latent, dynamic frequency information, removing noise and revealing significant vibration patterns. The enriched data are fed into the Deep Neural ODE Network which models the continuous dynamics of the vibrations, detecting microscopic invisible instability-indicating deviations.