Here is one of my research ideas. This research aims to enhance the CNC machine capabilities in precise drilling by adding a camera to catch the micro vision surrounding the drill bit and correcting the CNC XY-axis movement to the center of the pad hole. The accuracy is supposed to be more than 99%, thanks to the best performance and inference speed balancing of the YOLO algorithm, and my special ‘treatment’ 😊😊
Research Background
The electronics industry is witnessing unprecedented demand for high-precision, high-density printed circuit boards (PCBs) to support applications in consumer electronics, automotive systems, aerospace, telecommunications, and the Internet of Things (IoT). Traditional PCB drilling relies heavily on computer numerical control (CNC) machines guided by pre-programmed drill files (e.g., Gerber, Excellon), which require meticulous setup, alignment, and calibration. This process is time-consuming, error-prone, and inflexible, particularly in dynamic manufacturing environments with high variability or small-scale production needs. Commercial PCB drilling machines, such as Via Mechanics ND-6A2226, Schmoll MX-Series, and Orbotech Emerald, integrate vision systems for fiducial mark detection, axis compensation, and automated optical inspection (AOI), achieving remarkable precision (5–15 μm) and throughput (up to 3,000 holes/second). However, their dependence on drill files limits adaptability, requiring manual input from PCB design software and extensive setup, which can disrupt production workflows, especially for small and medium enterprises (SMEs) or prototyping labs.
A critical step in PCB fabrication is drilling, typically performed using computer numerical control (CNC) machines. These machines drill holes in PCBs based on drill files. However, the setup process for CNC machines, including machine calibration, PCB positioning, and drill file alignment, is time-consuming and prone to errors. Incorrect settings can result in defective PCBs, leading to material waste and financial losses. Given that PCB manufacturer handles a high volume of PCB orders from various customers across Indonesia daily, these inefficiencies significantly disrupt production timelines. There is a pressing need for an AI-computer vision guided PCB drilling machine that eliminates dependency on manual hardware settings, CNC homing, and drill files. Such a system would allow operators to simply place the PCB on a workbench, with the machine automatically detecting holes or pads and drilling them accurately, thereby significantly reducing production time.
This research proposes the development of an AI-driven, computer vision-based PCB drilling machine to address these challenges. By integrating advanced image processing and machine learning techniques, the system aims to enhance the efficiency and accuracy of PCB drilling, contributing to the broader field of smart manufacturing.
Research Gaps
While CNC-based PCB drilling is widely used, several limitations and research gaps remain:
1. Manual Setup Dependency: Existing CNC drilling systems require extensive manual calibration, including PCB alignment and drill file configuration. Errors in setup lead to defective PCBs, increasing production costs (Smith & Jones, 2019). There is a lack of fully automated systems that eliminate manual intervention in the setup process.
2. Absence of Commercial Machines with Automated Hole Detection and Drilling: Current commercial PCB drilling machines, such as those from Via Mechanics, Pluritec, and Hitachi, rely on pre-programmed drill files (e.g., Gerber or Excellon formats) to determine hole coordinates and require precise PCB alignment using fiducial markers or tooling pins. A review of recent industry sources indicates that no commercial machines feature real-time, vision-based hole detection to autonomously identify and drill holes without drill files. For instance, advanced CNC machines use high-speed spindles and optical alignment systems but still depend on CAD-derived coordinates, and even direct exposure techniques convert digital images into position maps for laser drilling rather than dynamically detecting features (SFX PCB, 2023; Proto Electronics, n.d.). This gap underscores the need for a system that uses AI and computer vision to detect holes directly on the PCB surface, by passing traditional file-based alignment.
3. Limited Application of Computer Vision: Although computer vision has been applied in PCB inspection for defect detection (Moganti et al., 2020), its use in real-time hole detection and drilling automation remains underexplored. Current systems rely heavily on pre-programmed drill files rather than dynamic detection of PCB features.
4. Scalability for Small-Scale Manufacturers: Most advanced PCB drilling solutions are designed for large-scale manufacturers, leaving small-scale facilities with limited access to cost-effective, automated technologies (Lee & Kim, 2021). There is a need for affordable, scalable solutions tailored to startups and small enterprises.
5. Real-Time Adaptability: Existing systems lack the ability to adapt to variations in PCB designs or minor misplacements during production. An intelligent system capable of real-time analysis and adjustment is needed to improve robustness (Chen et al., 2022).
These gaps highlight the need for an innovative approach that combines AI, computer vision, and automation to streamline PCB drilling processes, particularly for small-scale manufacturers.
Research Objectives
The proposed research aims to:
- Develop a computer vision algorithm to detect holes and pads on PCBs in real time using image processing and deep learning techniques.
- Design an AI-computer vision guided PCB drilling machine that integrates the vision algorithm with a drilling mechanism, eliminating the need for manual CNC setup and drill files.
- Evaluate the machine’s performance in terms of accuracy, speed, and cost-effectiveness compared to traditional CNC-based drilling.
- Propose a scalable and affordable solution suitable for small-scale PCB manufacturers like RaftechPCB.
Research Methodology
Phase 1: Literature Review
Conduct a comprehensive review of existing PCB drilling technologies, computer vision applications in manufacturing, and AI-based automation systems. The review will cover:
- PCB Drilling Technologies: Limitations of CNC-based drilling, including setup challenges and error rates (Smith & Jones, 2019).
- Computer Vision in Manufacturing: Applications of computer vision for defect detection in PCBs (Moganti et al., 2020) and real-time adaptive control systems (Chen et al., 2022).
- Hole Detection with Computer Vision: The use of OpenCV for image processing in industrial applications, such as edge detection and feature extraction for manufacturing tasks (Bradski & Kaehler, 2008). Additionally, YOLOv7 models for high-precision object detection in industrial manufacturing, which have demonstrated effectiveness in detecting small features like holes and pads on PCBs due to their improved accuracy and real-time performance (Wang et al., 2023). TensorFlow-based deep learning frameworks for feature detection and classification in manufacturing, offering robust tools for training models to recognize PCB pads and holes (Abadi et al., 2016).
- Automation for Small-Scale Manufacturers: Challenges in adopting smart manufacturing technologies for small enterprises (Lee & Kim, 2021).
This review will establish the theoretical foundation for integrating OpenCV, YOLOv7 or TensorFlow in the proposed system, ensuring alignment with state-of-the-art computer vision techniques for hole detection.
Phase 2: Algorithm Development
Develop a deep learning model for real-time detection of PCB holes and pads. The model will be trained on a dataset of PCB images, incorporating techniques such as image segmentation and feature extraction. OpenCV and YOLOv7/TensorFlow will be used for implementation.
Phase 3: System Integration
Integrate the vision algorithm with a drilling mechanism, such as a modified CNC or robotic arm. The machine will include a camera for real-time PCB imaging and a control unit to translate detected coordinates into drilling actions. Hardware prototyping will be conducted using Arduino or Raspberry Pi for cost-effectiveness.
Phase 4: Testing and Validation
Test the machine on single- and double-layer PCBs, measuring accuracy (hole placement error), speed (time per PCB), and defect rate. Compare results with traditional CNC drilling. Conduct a cost-benefit analysis to assess scalability for small-scale manufacturers.
Phase 5: Documentation and Dissemination
Document findings in a PhD dissertation and publish results in peer-reviewed journals (e.g., IEEE Transactions on Industrial Electronics). Present at conferences such as the International Conference on Smart Manufacturing.
This research will contribute to the field of smart manufacturing by introducing an innovative, AI-driven solution for PCB drilling. The proposed machine will:
- Reduce production time and costs for small-scale manufacturers, enhancing their competitiveness.
- Minimize material waste due to drilling errors, promoting sustainable manufacturing practices.
- Advance the application of computer vision and AI in PCB fabrication, paving the way for further automation in electronics manufacturing.
- Provide a scalable model that can be adopted by startups and small enterprises in developing countries, addressing the digital divide in manufacturing.
By addressing the specific challenges, this research will have practical implications for the Indonesian electronics industry while contributing to global advancements in Industry 4.0.
Below is my autonomous CNC experiment using YOLOv7 driving the MKS DLC32 GRBL controller through a vision camera.
Anyone interested in this research can contact me at hansapw@gmail.com
a Technopreneur – writer – Enthusiastic about learning AI, IoT, Robotics, Raspberry Pi, Arduino, ESP8266, Delphi, Python, Javascript, PHP, etc. Founder of startup Indomaker.com