"Computer Vision" in Electronic Manufacturing Quality Management (QA/QC) – Eliminate Human Error and Increase Productivity 4 Times

Project Background and Pain of Manual Quality Inspection (Manual Inspection):

In the Printed Circuit Board (PCB) and electronic components manufacturing industry, where each product contains thousands of microscopic components and solder joints smaller than a hair, factory quality assurance (QA/QC) is the vital line between profit and bankruptcy. If a circuit board with defective solder joints passes inspection and is assembled into a car or medical device, the cost of recall and reputational compensation will be a colossal disaster. This case study delves into the operating process of a large-scale electronics assembly factory that is facing a serious crisis in terms of defect rate.


Before the AI ​​project was applied, the factory used a team of hundreds of skilled workers standing by the conveyor belt, using magnifying glasses to check the appearance (Visual Inspection) of each circuit board. However, human biological limits cannot be overcome. After 4 hours of continuous eye strain under bright light, workers' eyes begin to get tired, their concentration decreases, leading to missing microscopic cracks on components (False Negatives), or misjudging a normal circuit board as defective due to shadows (False Positives). The speed of manual inspection also creates a large bottleneck, slowing down the production pace of the entire line. The factory needs a technology solution capable of "seeing" with absolute accuracy and never getting tired.


Technical Solution 1: Super-fast Industrial Camera Network (Machine Vision):

The team of technology architects has deployed an automated Computer Vision system. At the Inspection Stations along the conveyor belt, ultra-high resolution industrial Cameras (Line Scan Cameras) are installed. Unlike conventional cameras, these devices are combined with a multi-angle strobe lighting system.


When an electronic circuit board passes through a testing station at high speed, the lighting system will flash dozens of times a second at different angles, and the camera will simultaneously capture images of the surface, barcodes and slots. Using omnidirectional light (Dome Light, Low Angle Light) helps to completely highlight hidden defects such as scratches on the chip surface, lack of thermal paste, or pitted solder joints that are normally invisible to the human eye. This huge volume of 4K images is immediately pushed to an edge computer (Edge Server) equipped with a powerful GPU located right at the factory (On-premise) to avoid network transmission latency.


Technical Solution 2: Detecting Defects using Deep Learning Anomaly Detection:

Instead of using traditional image processing algorithms (Rule-based Vision) - which are very hard Prompt and susceptible to interference due to changes in lighting or board placement - the system has applied the power of Deep Learning, specifically Convolutional Neural Networks (CNN) architecture.


Data engineers have collected tens of thousands of pictures of perfect boards (Golden Samples) and boards containing all kinds of different errors to train the AI model. The AI ​​system not only learns to look for known defects (such as misaligned components, missing screws, open welds), but through the Anomaly Detection algorithm, it has the ability to "flag" any structural feature that deviates from the standard model, even if it is a type of defect that has never appeared before. AI's inference process on each photo takes only 15 milliseconds. If the system detects an error, it immediately sends a signal to a Pneumatic Ejector Robot Arm to push the damaged circuit board straight off the conveyor belt, put it in a tray waiting for reprocessing, and at the same time display the error location in red on the screen (HMI) for technical staff to fix.


Operational Results and Zero Defect Production Era (Zero Defect):

After 4 months of implementation Deploying reality and continuously refining the model, the result for the factory is a leap in productivity. Product inspection speed increases 4 times compared to manual work, completely eliminating bottlenecks on the line. The error detection rate reached 99.98%, helping the company completely eliminate complaints from partner customers due to defective goods. Notably, the rate of mistakenly evaluating good goods as defective (False Reject Rate) dropped sharply by 75%, saving millions of dollars in unnecessary scrap costs (Scrap Cost).


The previous team of QA/QC workers were not fired, but were retrained (Reskilling) to switch to the role of controlling and fine-tuning the AI ​​system. This case study is clear evidence that: The application of Artificial Intelligence and Computer Vision in the production environment is not just an automation investment to cut costs, but it is the ultimate weapon to raise product quality to the level of machine perfection, creating a new standard of Smart Factories in the Industry 4.0 era.