| Sponsor |
|
Epoxy Offers Ultra High Heat Transfer
Bond EP5TC-80 is a NASA low outgassing rated epoxy that achieves a thermal conductivity of 3.3-3.7 W/(m·K), while also retaining its electrical non-conductivity.
Master Bond
|
|
Machine Vision-based Defect Inspection for Plated Through Hole Components
This study delineates the conceptualization of a machine vision-based defect inspection model, meticulously engineered to confront the intricacies of high-throughput PCB assembly.
Technical Paper
DOWNLOAD
|
Authored By:
Jinal Prajapati, Soujanya M N, Darshil Patel, Daryl Santos
Department of Systems Science and Industrial Engineering, Binghamton University
NY, USA
Pranav Rastogi, Snehal Shindadkar
Manufacturing Engineering, Foxconn Industrial Internet
WI, USA
Summary
In contemporary Printed Circuit Board Assembly (PCBA) operations, to ensure high-quality and efficient manufacturing processes is critical. The optimization of process conformance in PCBA lines presents significant challenges to production efficiency. Detailed process mining analyses of actual PCBA lines have revealed substantial deviations from ideal processes at the Plated Through Hole (PTH) stage, resulting in elevated rework rates post-wave soldering due to component placement issues. The primary root causes include component misalignment, missing components, and incorrect polarity at the manual insertion station preceding wave soldering.
This research paper introduces an innovative machine vision-based inspection framework to address the above-mentioned critical issues for large-scale PCBA operations. The developed framework employs a three-phase computational approach wherein Phase 1 includes calibrated image acquisition with reduced radial distortion. Next, Phase 2 defines Region of Interest (ROI) using advanced template extraction algorithms that precisely isolate component templates, locations, and sizes. Finally, Phase 3 consists of a multiscale template matching algorithm that employs a hybrid multichannel approach to enhance light and color sensitivity. The proposed multiscale template matching algorithm employs a hybrid multichannel approach to enhance light and color sensitivity, integrating color-based matching with luminance information to improve defect detection.
This approach utilizes a three-tiered strategy incorporating Hue, Saturation, Value (HSV), which separates image color components for precise hue detection; Blue, Green, Red (BGR), which captures standard color information; and grayscale, which focuses on luminance for detailed texture and edge detection. Additionally, bilateral filtering and Gaussian blur are applied to reduce noise and enhance edge detection, further maximizing the accuracy and reliability of the defect identification process.
By systematically addressing these root causes identified through process mining, the system significantly enhances process conformance by reducing the number of defective PCBs entering the wave soldering stage and subsequently lowering the rework rate. The proposed methodology provides a robust solution to common manufacturing challenges, ensuring higher process adherence and operational efficiency.
Conclusions
In the dynamic and increasingly complex domain of PCB assembly, the assurance of PTH component integrity is paramount to upholding rigorous production standards. This study delineates the conceptualization and implementation of a machine vision-based defect inspection model, meticulously engineered to confront the intricacies of high-throughput PCB assembly environments. The framework of the proposed system is underpinned by three phase computational framework, which synergizes precision image acquisition, exacting region of interest delineation, and an innovative multiscale template matching methodology.
This integration facilitates a robust and comprehensive defect detection process, significantly improving the system’s diagnostic acuity. The primary phase includes the acquisition of high-resolution PCB imagery under rigorously controlled lighting conditions to mitigate the potential confounding effects of reflections and shadows, which could compromise the accuracy of defect identification. The subsequent phase is devoted to ROI definition, where targeted regions, particularly those encompassing PTH components, are isolated for focused analysis. The final phase implements multiscale template matching, a method that juxtaposes multiple template scales with the target images, ensuring precise defect detection across varying component sizes and orientations.
The efficacy and robustness of the inspection system were validated through rigorous empirical evaluations, encompassing both main and system PCB assemblies. Iterative optimizations of the detection algorithms were conducted to refine system performance, culminating in an exceptional detection accuracy rate of 98%. This precision is attributed to a hybrid inspection approach that leverages the synergistic capabilities of multiple color spaces—HSV, BGR, and grayscale—coupled with advanced filtering techniques.
These methodological improvements enable the system to maintain heightened sensitivity to minute lighting variations and subtle color discrepancies, which are frequently overlooked by conventional inspection systems. The successful deployment of this system across diverse PCB assembly environments underscores its scalability and robustness, marking a significant advancement in PCB assembly industries. By substantially reducing rework rates and mitigating defect occurrences, the system not only optimizes production throughput but also elevates the overall quality of the manufactured products.
Looking ahead, the future of defect inspection for PCB manufacturing lies in embracing digital transformation and advanced AI technologies to push the current detection capabilities. One promising direction seems an integration of transformational AI, particularly attention-based models like Vision Transformers (ViTs), which focus on local patches, as it processes entire images which makes them ideal for identifying defects across PCBs. By learning both global and local dependencies, ViTs can dynamically optimize their detection accuracy in real-time as they deal with new defect patterns and adapt to production conditions.
Apart from that, incorporating self-supervised learning that allows the inspection models to continuously learn from new defect patterns and dynamically optimize the detection accuracy, even in the absence of labeled data. This adaptability, paired with continuous learning, must drive inspection systems toward achieving detection accuracy of 99.99%, significantly surpassing current industry standards. Further than AI, advancements in imaging technologies are also pivotal. The inclusion of hyperspectral imaging and exploration of color spectrums such as CIELAB color space, YUV (Luminance, Blue Projection, Red Projection), which separates brightness from color information can enhance the detection which can facilitate detection of some anomalies which might be not captured earlier.
Furthermore, the application of multi-dimensional (X-D) imaging, such as 4D or 5D techniques, could also provide insights into component structure by ensuring smallest deviations are identified from multiple perspectives. To accommodate holistic automation, Co-bots (collaborative robots) in conjunction with machine vision systems can improve real-time defect detection. Co-bots with high-precision actuators can assist in repositioning components.
By integrating sensor feedback loops with AI decision-making models, the system can respond dynamically to detected defects without halting production which is crucial for systematic integration. Such seamless, end-to-end integration of AI models, imaging systems, and robotic assistance into the PCB manufacturing process will streamline operations and improves the overall product quality.
Initially Published in the SMTA Proceedings
|
Comments
|
No comments have been submitted to date.
|
|
|
|
Free Newsletter Subscription
Circuitnet is built for professionals who bear the responsibility of looking ahead, imagining the future, and preparing for it.
Insert Your Email Address
|
|
