Deep Learning Road Surface Classification for Vehicle Terrain Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for controlling vehicle terrain modes based on road surface conditions are inaccurate due to reliance on dynamics-based models and fail to determine road surface states with high precision, leading to suboptimal terrain mode settings that affect travel stability and comfort.
Innovation Solution
A deep learning-based system that identifies road surface types (normal, sand, mud, snow) by processing vehicle signals through a neural network model, calculating energy ratios in frequency bands, and determining road surface state values to set optimal terrain modes, while maintaining current settings under certain conditions to ensure stability and comfort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a dynamics-based model is used to determine road surface state, then the system complexity is reduced, but the measurement precision of road surface state deteriorates
Solution Approach 1:
The patent replaces the dynamics-based mechanical model with a deep learning-based intelligent model. The deep learning model processes multiple terrain surface signals (acceleration, steering angle, brake pedal position, engine torque) to accurately determine road surface state without relying on complex mechanical dynamics calculations, thereby maintaining low system complexity while achieving high measurement precision.
Solution Approach 2:
The patent transforms the road surface determination approach from using physical dynamics parameters to using data-driven parameters from deep learning. By changing the fundamental parameters from mechanical model outputs to neural network predictions based on multiple sensor signals, the system achieves both simplicity and accuracy.
2Device complexity
If a combined probability value calculation scheme is used to determine sub-system control mode, then the device complexity is reduced, but the measurement precision of road surface state deteriorates
Solution Approach 1:
The patent replaces the simple combined probability calculation scheme with a deep learning-based determination system. The deep learning model comprehensively analyzes multiple terrain surface signals to identify road surface state with high accuracy, avoiding the limitations of probability combination methods while maintaining manageable system complexity through integrated processing.
3Measurement precision
If deep learning-based road surface model is used to identify road surface type, then the measurement precision of road surface type identification is improved, but the device complexity increases
Solution Approach 1:
The patent implements a multi-functional deep learning model that simultaneously processes multiple terrain surface signals (acceleration, steering angle, brake pedal position, engine torque) and performs comprehensive road surface state determination. This universal approach achieves high identification accuracy for various road surfaces while consolidating multiple functions into a single system, preventing excessive complexity increase.
Data Source
AI summary
A device for identifying a road surface includes: storage for storing a deep learning-based road surface model; and a controller configured to identify a type of a road surface on which a vehicle is currently traveling, using the road surface model. The device for identifying a road surface can identify a type of a road surface on which the vehicle is traveling based on deep learning and control the terrain mode of the vehicle based on the identified type of the road surface. The type of the road surface on which the vehicle is traveling may be identified with a high accuracy and an optimal terrain mode may be set, thereby improving not only travel stability but also riding comfort of the vehicle.