Deep Learning Sound Prediction Using Accelerometer Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current virtual sensing methods for estimating sound in electric drive systems are limited by their reliance on physics-based models, which are cumbersome and difficult to adapt, and lack data-driven approaches, making it challenging to predict sound from vibrational data effectively in noisy environments like end-of-line testing.
Innovation Solution
A deep learning-based virtual sensing method that uses machine learning models, such as U-Net and Transformer, to learn the relationship between vibrational data and sound, allowing for the prediction of sound emissions from electric drive systems using accelerometer data, even in environments where high-quality sound recordings are not feasible.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If physics-based models are used for virtual sensing, then sound estimation can be achieved, but the models become cumbersome and difficult to adapt
Solution Approach 1:
The patent replaces physics-based mechanical models with data-driven machine learning models (specifically deep neural networks). This substitution eliminates the need for complex analytical physics models while achieving accurate sound predictions through learned patterns from training data, directly resolving the contradiction between reliability and device complexity.
Solution Approach 2:
The patent transforms the approach from using fixed physics-based parameters to using learned parameters from training data. The machine learning model learns optimal parameter relationships from labeled training data, allowing accurate sound estimation without the complexity of explicit physics model parameter management.
2Measurement precision
If high-quality sound recordings are obtained, then accurate NVH analysis can be performed, but dedicated recording environments are required which increase cost and reduce productivity
Solution Approach 1:
The patent creates a virtual copy of the sound signal from accelerometer data using machine learning. Instead of requiring actual high-quality sound recordings in controlled environments, the system learns the mapping from vibration to sound and generates predicted sound signals that can be used for NVH analysis in any environment, including noisy factory floors.
Solution Approach 2:
The patent introduces accelerometer data as an intermediary that bridges the gap between physical vibration and perceived sound. The machine learning model learns to translate accelerometer measurements into sound predictions, eliminating the need for direct sound recording while maintaining measurement precision.
3Reliability
If physics-based models are used, then theoretical accuracy can be achieved, but adaptation to different systems becomes difficult
Solution Approach 1:
The patent makes the system adaptive by training machine learning models on system-specific data. Each electric drive system can have its own trained model that adapts to its unique characteristics, allowing the system to maintain high prediction accuracy across different applications without requiring complex physics model adjustments.
Solution Approach 2:
The patent enables easy adaptation by changing the training data rather than the model structure. When applying to a new electric drive system, the same machine learning framework can be retrained with data from that specific system, allowing rapid adaptation without redesigning the underlying physics models.
Data Source
AI summary
A system includes a processor in communication with one or more sensors, wherein the processor is programmed to receive data including one or more of real-time current information, real-time voltage information, or real-time vibrational information from a run-time device, wherein the run-time device is an actuator or electric dive, and utilize a trained machine learning model and the data as an input to the trained machine learning model, output a sound prediction associated with estimated sound emitted from the run-time device.