Brake Disc Quality Evaluation With CNN-Based Surface Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing manual methods for evaluating the quality of vehicle brake discs, particularly those with FNC surfaces, are inefficient, labor-intensive, and prone to inaccuracies due to the uniform grayscale and subtle feature differences, making it difficult to assess characteristics like infiltrated layer thickness accurately.
Innovation Solution
A deep learning-based convolutional neural network model is employed, incorporating a feature extraction network, anchor box detection, and classification and regression networks with attention mechanisms and double fully connected layers, to automatically detect and evaluate brake disc quality by identifying features such as infiltrated layer thickness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual evaluation methods are used for brake disc quality assessment, then operational simplicity is maintained, but evaluation efficiency is low and processing time is long
Solution Approach 1:
The patent replaces manual visual inspection with an automated convolutional neural network system that processes images of brake disc surfaces. The CNN model automatically detects and measures infiltrated layer thickness, eliminating the need for manual measurement while significantly improving evaluation efficiency and reducing processing time from minutes to seconds.
Solution Approach 2:
The patent creates a digital copy of the brake disc surface through image capture and uses this copy for analysis. The system processes multiple images of the brake disc surface, extracting features and measurements from these digital copies rather than requiring direct manual measurement of the physical object, thereby improving efficiency and consistency.
2Measurement precision
If manual evaluation methods are used, then device complexity is low, but measurement precision is insufficient due to uniform grayscale and subtle feature differences
Solution Approach 1:
The patent transforms the image data by adjusting parameters such as color space conversion (RGB to HSV), contrast enhancement, and feature extraction thresholds. These parameter changes enable the system to detect subtle variations in infiltrated layer thickness that are not visible to the human eye, significantly improving measurement precision for uniform grayscale surfaces.
Solution Approach 2:
The patent applies different processing weights and detection thresholds to different regions of the brake disc surface images. The system identifies and emphasizes specific local features such as boundary lines and texture variations that indicate infiltrated layer thickness, while suppressing background noise and uniform regions, thereby improving local measurement precision.
3Reliability
If deep learning-based CNN model is implemented, then evaluation accuracy and robustness are improved, but computational resources and model complexity increase
Solution Approach 1:
The patent divides the CNN model into distinct functional modules: image preprocessing module, feature extraction modules (convolutional layers), anchor box detection module, classification module, and regression module. This segmentation allows each module to perform specific tasks efficiently and enables independent optimization of each component, improving overall reliability while managing computational complexity.
Solution Approach 2:
The patent performs preliminary image processing and feature extraction before the main detection and classification processes. By pre-processing images to enhance contrast, normalize data, and extract preliminary features, the system reduces the computational burden on subsequent processing stages, thereby improving evaluation accuracy while optimizing resource utilization.
4Reliability
If extensive training data is used to improve model accuracy, then evaluation reliability increases, but data collection time and processing requirements increase
Solution Approach 1:
The patent generates synthetic training data by creating virtual copies of brake disc surfaces with varying characteristics. These synthetic images, generated through computer graphics and image processing, provide diverse training examples without requiring physical sampling and manual annotation, significantly reducing data collection time while maintaining model training quality.
Solution Approach 2:
The patent applies data augmentation techniques that transform existing training images by adjusting parameters such as rotation, scaling, color variation, and noise addition. These parameter changes create varied training samples from a limited set of original images, effectively increasing the training data volume without requiring additional physical data collection, thereby improving model generalization performance while reducing time loss.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for establishing a deep learning based convolutional neural network model, in which the convolutional neural network model is used to evaluating the quality of a vehicle's brake disc, and includes a feature extraction network, an anchor box detection network and a classification and regression network; and the classification and regression network further includes a first fully connected layer, a second fully connected layer, and a regularization layer between the first fully connected layer and the second fully connected layer.