Machine Learning Network for NVH Signal Prediction Analysis

The patent applies a universal machine learning framework that can be applied across multiple applications without requiring custom physics-driven models for each. The system uses a general-purpose ML model that processes various signal types (vibration, sound, current, voltage) and can be adapted to different product types through data training, eliminating the need to develop separate custom models for each application while maintaining accuracy.

The patent replaces complex custom physics-driven models with a machine learning-based approach. Instead of manually developing and tuning physics models for each application, the system uses ML algorithms that automatically learn relationships from data, substituting the complex manual model development process with an automated data-driven approach that reduces overall system complexity.

The patent implements a feedback mechanism where the machine learning model continuously refines its predictions by comparing predicted signal characteristics against actual measured values. The system uses prediction errors to adjust and improve future predictions, ensuring that the desired metrics (such as sound, torque, or vibration predictions) are accurately captured rather than losing information in the estimation process.

The patent changes the output parameters of the machine learning model from generic mutual information estimates to specific, application-relevant predictions. By training the model to predict specific signal characteristics (sound levels, torque values, vibration amplitudes) directly, the system transforms the vague ML output into concrete, actionable predictions that match the desired metrics for each application.