Vehicle Motor Noise Prediction Using Transfer Learning Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle noise analysis systems struggle to accurately isolate and predict motor-specific noise due to overlapping noise patterns, leading to unreliable diagnostic accuracy and inefficiencies in performance optimization.
Innovation Solution
A method and device utilizing deep learning models through transfer learning, employing SNR-based data mixing and time domain audio source separation, to isolate and predict vehicle motor noise by generating specialized models for each vehicle type, ensuring accurate separation and prediction of motor noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If general vehicle noise analysis is performed without isolating motor noise, then overall noise monitoring is simple, but diagnostic accuracy for motor-specific issues deteriorates
Solution Approach 1:
The patent segments the mixed vehicle noise signal into distinct motor noise and general vehicle noise components using deep learning models. The audio source separation model divides the composite noise signal into isolated motor noise signals, enabling precise diagnostic analysis while maintaining system feasibility through automated processing.
Solution Approach 2:
The patent introduces deep learning models as intermediary tools to facilitate the separation of motor noise from general vehicle noise. These models act as mediators that process the mixed noise signal and output isolated motor noise components, resolving the contradiction between diagnostic precision and analysis complexity.
2Measurement precision
If motor noise is isolated using traditional methods, then analysis time is reduced, but separation accuracy deteriorates due to overlapping noise patterns
Solution Approach 1:
The patent replaces traditional mechanical or signal-processing-based noise separation methods with deep learning-based audio source separation models. This substitution enables the system to handle overlapping noise patterns more effectively by learning complex noise characteristics and separation patterns that traditional methods cannot capture.
Solution Approach 2:
The patent transforms the noise separation problem from a traditional signal processing task into a machine learning parameter optimization problem. By changing the approach from fixed algorithmic parameters to adaptive learning parameters, the system achieves higher separation accuracy for overlapping noise patterns through transfer learning and model training.
3Measurement precision
If separate deep learning models are trained for each vehicle type, then prediction accuracy improves, but training time and computational cost increase
Solution Approach 1:
The patent performs preliminary training of deep learning models on a reference vehicle to extract transferable features and patterns. This preliminary action creates a pre-trained model that captures general motor noise characteristics, which can then be efficiently adapted to specific vehicle types with minimal additional training, reducing overall training time while maintaining high accuracy.
Solution Approach 2:
The patent creates a universal deep learning model through transfer learning that can be applied across multiple vehicle types. The model trained on reference vehicle data serves multiple purposes and can be adapted to different vehicle types by adding specialized layers or fine-tuning, eliminating the need to train completely separate models for each vehicle type.
Data Source
AI summary
A method and a device for predicting vehicle motor noise that predicts motor noise separated from vehicle noise are provided. The method may include acquiring motor noise data that does not include the vehicle noise; acquiring vehicle noise data that does not comprise the motor noise; generating training data by mixing the motor noise data and the vehicle noise data; providing a deep learning model built differently for each vehicle through transfer learning based on a pre-trained model that is pre-trained by the learning data; and predicting motor noise for each vehicle by using the deep learning model.