TL;DR
This paper presents a novel robot-driven 3D shape measurement system designed for real-time quality inspection of hot, thermally radiating objects on a forging production line. By integrating a specialized 3D scanner with an industrial robot and employing advanced heat-resistant and data processing techniques, the system achieves accurate and efficient inspection in a challenging industrial environment.
What Problem Does This Paper Address?
Automated quality inspection of forged parts is crucial for maintaining production efficiency and product quality. However, inspecting these parts while they are still hot on a production line presents significant challenges. The intense heat radiation can interfere with optical sensors, and the high temperatures prevent the use of traditional artificial markers for aligning 3D data. Furthermore, the dynamic nature of the production line, with moving parts and potential background clutter, complicates accurate 3D reconstruction and registration. Existing 3D measurement systems are often not robust enough to handle these extreme conditions, leading to manual inspections or delays in quality control, which can result in the production of defective parts.
Methods
The proposed system tackles these challenges through several innovative approaches:
- Heat Resistance: A dual-filter set (blue light filter and IR cut filter) is employed to mitigate the effects of thermal radiation on the cameras and projector. Additionally, a closed-loop air circulation cooling system with a thermoelectric cooling device is integrated into the sensor's end-effector to maintain a stable operating temperature.
- Accelerated Measurement: To improve speed and reduce exposure to heat, a time-division multiplexing trigger is used to synchronize the projector and cameras, maximizing bandwidth utilization. An improved multi-frequency phase-shifting method is also introduced, reducing the number of projected patterns required for 3D reconstruction by 38.5%.
- Robust Data Alignment: A view integration method is used to align non-overlapping 3D data from different views. This relies on the robot's repeatable motion for coarse alignment. For fine registration, a novel noise-insensitive adaptive distance threshold-based ICP (Iterative Closest Point) algorithm is proposed. This algorithm dynamically adjusts its distance threshold based on statistical analysis of nearest neighbor distances, making it more resilient to irrelevant background data often present in industrial settings.
Datasets
The paper does not explicitly mention training datasets. The system was evaluated using real-world data collected from a forging factory production line, specifically inspecting thermal axles. The "datasets" used for evaluation were the 3D point cloud data generated from these real-time inspections.
Key Insights and Ideas
- Heat is a Major Hurdle: The paper strongly emphasizes that high temperatures are not just an inconvenience but a fundamental obstacle to accurate optical metrology, requiring dedicated engineering solutions beyond standard sensor design.
- Robot Repeatability is Key: Recognizing that absolute robot accuracy might be insufficient, the system leverages the repeatability of industrial robots for robust multi-view data alignment, a clever workaround for precise positioning.
- ICP Needs a Tune-Up for Industry: The standard ICP algorithm struggles with the "noise" of industrial environments (e.g., conveyor belts, grippers). The proposed adaptive threshold ICP demonstrates that fine-tuning established algorithms can yield significant improvements in robustness for real-world applications.
- Speed vs. Accuracy Trade-off Management: The paper shows that by intelligently combining techniques like time-division multiplexing and optimized phase-shifting, significant speedups can be achieved without compromising the required accuracy for quality inspection.
Key Results
- Speed Improvement: The improved multi-frequency phase-shifting method reduced the time for a typical four-step, three-frequency phase-shifting operation by 38.5%.
- Data Alignment Accuracy: The system demonstrated stable online 3D quality inspection for thermal axles. In experiments, the maximum deviation for the total length of a vehicle axle was 0.11 mm, well within the expected tolerance of 1.0 mm.
- Robustness to Noise: The proposed noise-insensitive data registration algorithm successfully removed irrelevant background data (occupying 22% of the original scanned data) and accurately aligned the processed data to the design model.
- System Performance: The entire scanning procedure for an axle took 27 seconds, with 3D data processing taking 12-13 seconds, allowing for parallel execution with the production line.
Limitations and Future Work
- View Integration Flexibility: The current view integration method is noted as being less flexible, requiring separate configurations for different plans and potentially needing recalibration with environmental temperature changes.
- Absolute Accuracy Estimation: The authors acknowledge that determining the absolute accuracy of the system is an area that needs further investigation.
- Long-Term Repeatability: A long-term quantitative assessment of robotic repeatability under thermal conditions is also suggested for future work.
- Future Directions: The team plans to use targets with accurately obtainable thermal dimensions for future accuracy evaluation and is exploring other precision assurance and verification methods.
Summary updated