研究目的
To propose a real-time algorithm for generating differential topographic images of product surfaces using laser scanning to detect and measure defects for quality control in manufacturing industries.
研究成果
The proposed algorithm successfully generates differential topographic images in real time, enabling efficient surface defect detection and measurement for quality control. It meets real-time constraints with processing times below 2 ms per profile and achieves a mean error of 0.14 mm, comparable to sensor noise. The use of parallel processing and caching strategies enhances performance, making it suitable for industrial applications. Future work could focus on improving robustness to environmental variations and extending to other product types.
研究不足
The algorithm requires accurate placement of sensors perpendicular to the product surface to avoid occlusions and ensure precision. It is sensitive to vibrations and flatness defects, which may affect registration accuracy. The method is specific to long products with definable profiles and may not generalize to other shapes. Computational resources (e.g., multiple cores) are needed for real-time processing, and the error is influenced by sensor noise.
1：Experimental Design and Method Selection:
The algorithm uses laser scanning with profilometers to acquire 3D point clouds of product surfaces, followed by registration to a mathematical model using the iterative closest point (ICP) algorithm, differential map calculation, and image generation. It employs parallel processing with multiple threads for real-time performance.
2：Sample Selection and Data Sources:
Point clouds are acquired from long products like rails using four profilometers (LMI Technologies Gocator 3250), with tests conducted on 10 rails of 100m length (approximately 500,000 profiles). Synthetic vibrations are applied to simulate industrial conditions.
3：List of Experimental Equipment and Materials:
Profilometers (LMI Technologies Gocator 3250), computers (System A: Intel Core I7 4790 3.60 GHz, 16GB RAM; System B: Intel Xeon E5-2687w v4 3.00 GHz, 32GB RAM), and mathematical models of product profiles defined by arcs and segments.
4：60 GHz, 16GB RAM; System B:
4. Experimental Procedures and Operational Workflow: For each profilometer, point clouds are acquired, registered to the model using coarse and fine registration (ICP), differential maps are calculated by comparing points to the model, and images are generated by converting deviations to grayscale values. The process is parallelized with threads for acquisition and processing.
5：00 GHz, 32GB RAM), and mathematical models of product profiles defined by arcs and segments.
Experimental Procedures and Operational Workflow:
5. Data Analysis Methods: Performance is evaluated based on processing times (e.g., registration time, differential map calculation time), mean distance from point cloud to model, and error analysis using histograms and percentiles. Subsampling rates are optimized to balance speed and accuracy.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
