YKK Window Sealent Inspection System
End Effector Design
Introduction
Delamination in window sealant—defined as separation between the sealant and aluminum frame due to wear or aging—can compromise waterproofing and structural integrity. To address this, the project aimed to design and fabricate an end effector capable of probing the sealant surface and revealing delamination for high‑accuracy sensor scanning within an automated robotic inspection system .
Methods
The initial end effector assembly combined a Zivid high‑precision surface scanner, a force‑sensing module for feedback protection, an optimized testing probe geometry, and a RealSense RGB‑D camera to capture both color and depth data during point inspections . Building on this, a spring mechanism was added to stabilize the probe before activation, and a metal‑rod roller was integrated for continuous, faster inspection—trading off some camera visibility for throughput . After identifying drift and creep issues with the original force‑sensitive resistor (FSR), the design replaced it with a robust load cell (0–30 N range) paired with an HX711 amplifier and Arduino Nano, delivering more consistent force measurements with minimal viscoelastic creep effects . The robotic routine was implemented on a UR5e arm under ROS control: raw load‑cell readings are published, filtered by a PID controller for admittance control, and translated into motor commands to maintain target probing force .
Results
Comparative testing showed that the roller assembly examined surfaces significantly faster than point probing, with both methods equally effective at exposing delamination but the roller reducing camera visibility . Load‑cell measurements averaged 25–35 N during probing, though maintaining uniform pressure proved challenging due to assembly dynamics .
Future Work
Key next steps include implementing and tuning the force‑feedback controller to stabilize applied pressure, redesigning the roller assembly—potentially using sheet‑metal components—to improve camera line‑of‑sight, and enhancing the ROS force‑topic pipeline and electronic packaging. Ultimately, the goal is a compact, robust end effector capable of uniform, high‑throughput delamination inspection.
Automatic Sample Scan Procedure
Sample Scan On-board Camera View
Improved Sample Scan with Roller End Effector
Handheld Inspection Tool Design
Introduction
Delamination in window sealants—defined as the loss of adhesion between silicone and aluminum frames from aging or environmental stress—can severely compromise waterproofing and structural integrity. Manual inspection remains slow, inconsistent, and risky for workers. This project developed an automated delamination inspection tool: a robotic end effector combining controlled-force sensing, roller probing, and machine vision to detect hidden defects efficiently and safely.
Methods
The inspection tool integrates a force-sensing assembly with a stepper-driven lead-screw actuator, suspension spring, and load cell (0–50 N, ±0.5% F.S.), ensuring stable force application near the 35 N threshold needed to expose delamination without damage. A linear guide rail constrained motion and improved accuracy. A modular housing mounted the actuator, rollers, and camera with vibration damping and serviceability in mind.
Electronics included an Arduino Nano, HX711 amplifier, and SERVO42C stepper driver for closed-loop force control. A proportional feedback loop maintained contact within a 33–37 N tolerance band, publishing data through ROS. For imaging, an Intel RealSense D405 RGB-D camera paired with diffused LED backlighting captured surface defects at 5 fps. Data were pre-processed (cropping, marker-based ROI detection) and segmented using SAM2 multi-mask workflows with morphological refinement to highlight delamination regions.
Results
The tool successfully revealed delamination and crater-like defects with high fidelity under controlled lighting. The feedback system applied stable probing forces within ±2 N of target. The imaging and segmentation pipeline accurately isolated delamination zones, minimizing false positives through multi-mask refinement. Compared to manual probing, the tool improved cycle speed by over 200% while preserving detection sensitivity. However, challenges included roller misalignment, variable lighting reflections, and occasional marker misidentification, which sometimes propagated errors in segmentation.
Future Work
Next steps include automating roller alignment via closed-loop or vision guidance, improving pre-processing robustness for varying colors/lighting, and incorporating deep learning methods (CNNs) for more resilient defect detection. Expanding the algorithm to classify defect types (delamination vs. craters vs. irregularities) would further enhance utility. Ultimately, the goal is a compact, industrial-grade inspection system capable of high-throughput, operator-independent detection in real-world environments.
