CircuitVision is an automated optical inspection system for PCB (Printed Circuit Board) boards, utilizing a Raspberry Pi 4 and a KL25 microcontroller. The system, integrated with a CNC machine, automates the process of capturing images of PCB boards and evaluating the placement and presence of components.
CircuitVision offers two modes of operation, selectable via a keypad with a menu displayed on an LCD:
-
Manual Configuration and Picture Taking
- Allows the user to manually control the camera mounted on the CNC machine using a joystick.
- The user can capture images at any moment, triggering an automatic evaluation of the PCB board based on the selected design.
-
Automatic Inspection of Preconfigured PCB Boards
- Recommended mode for inspecting multiple PCB boards.
- The CNC machine automatically captures images of all PCB boards in the test bed, saving and displaying the results on a monitor.
The KL25 microcontroller manages the interaction and control of the following hardware components:
- LCD Display
- Joystick
- 3x3 Keypad
- Stepper Motor Driver Control
- UART Communication with Raspberry Pi 4
The KL25 orchestrates the system's operations and communicates with the Raspberry Pi 4 only when PCB evaluation is required.
The Raspberry Pi 4 handles image processing and evaluation using OpenCV in C++. The main tasks include:
- Capturing images of the PCB board on the CNC bed.
- Processing and evaluating the images.
- Sending results back to the KL25 microcontroller for display on the LCD or monitor (via HDMI).
The automatic optical evaluation process is inspired by the following research papers:
- PCB Defect Detection Using OpenCV with Image Subtraction Method by Fa Iq Raihan and Win Ce
- Automatic PCB Inspection Algorithms: A Survey by Madhav Mogonti and Fikret Ercal
The evaluation pipeline involves the following steps:
- Capture and save the image.
- Create a mask to filter the white background.
- Identify the largest contour (the PCB itself).
- Correct the perspective and resize the image.
- Apply preprocessing (blurring, edge detection, conversion to a single channel).
- Perform XOR operation between reference and evaluation images.
- Remove noise from the XOR result.
- Read component bounding box coordinates from a CSV file.
- Generate a box image for each component.
- Calculate the percentage of lit pixels and compare it to the allowed maximum.
- Display and save the results.
Below are some temporary reference images from the project:
