Tire Resonance Audio Analysis for Low-Cost Road Surface Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing road surface condition management systems, such as road weather information systems, face high installation and maintenance costs, limiting their widespread dissemination and effectiveness in providing real-time road surface condition information for safe driving and management.
Innovation Solution
An apparatus and method for analyzing road surface conditions using audio signals collected from tire resonance sounds and friction sounds, employing a frequency selection model and recognition model to identify dominant frequency bands and predict road surface conditions, thereby reducing reliance on costly equipment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If road weather information systems are deployed to provide real-time road surface condition information, then road safety and management efficiency are improved, but installation and maintenance costs become prohibitively high
Solution Approach 1:
The patent replaces complex mechanical road weather information systems with an acoustic-based detection system. By using microphones to capture tire-road friction sounds and resonance sounds, and processing these acoustic signals through frequency selection models and recognition models, the system achieves road surface condition detection without requiring expensive mechanical sensors and infrastructure.
Solution Approach 2:
The system utilizes sounds naturally generated by the vehicle-tire-road interaction as the detection source. The friction sounds and resonance sounds are byproducts of normal driving operations, eliminating the need for separate active sensing mechanisms. This self-service approach allows the system to leverage existing operational data for road condition monitoring.
2Measurement precision
If traditional road weather information systems are used, then comprehensive road surface condition monitoring is achieved, but equipment installation and maintenance costs are very high
Solution Approach 1:
The patent extracts specific frequency components from the complex acoustic signal that carry road surface condition information. By using frequency selection models to identify and isolate dominant frequency bands corresponding to friction sounds and resonance sounds, the system extracts meaningful data from the raw acoustic environment, achieving precise detection with minimal hardware.
Solution Approach 2:
The system transforms the road surface condition detection problem from analyzing physical parameters (temperature, moisture content) to analyzing acoustic parameters (frequency, amplitude, spectral characteristics). By changing the detection parameter from direct physical measurement to acoustic signal analysis, the system achieves comparable or superior precision with significantly reduced deployment costs.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Accurately determines road surface conditions and tire deterioration without the need for extensive infrastructure, enhancing safety and management efficiency while minimizing costs.
Implementation Method 1
audio acquired from a target such as a resonance sound inside a tire of a vehicle driving on a road
Implementation Method 2
generating, by the label generator, a label audio-of-interest signal by attenuating a band other than a frequency-of-interest band in the training audio signal
Data Source
AI summary
Analyzing the condition of a road surface by using a frequency-of-interest or a resonance sound in a tire. A training audio signal is sent to a learning unit, a label generator and a frequency selection model that does not complete learning. The training audio signal is obtained by collecting driving noise generated when a vehicle travels on a road. The label generator generates a label audio-of-interest signal by attenuating a band other than a frequency-of-interest band in the training audio signal. The frequency selection model derives a training imitated audio-of-interest signal imitating an audio-of-interest signal by performing a plurality of operations in which an unlearned weight is applied to the training audio signal. The learning unit calculates a generation loss that is a difference between the training imitated audio-of-interest signal and the label audio-of-interest signal, and performs optimization of updating the weight of the frequency selection model.