PCB Defect Detection Using Neural Networks Under Variable Lighting

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

AOI systems face limitations such as single-task dedication, high expense, limited processing speed, lack of portability, difficulty in training, limited camera angles, and inability to operate under multiple lighting conditions.

Innovation Solution

The use of neural networks, specifically Siamese convolutional neural networks, trained on diverse image data to detect defects and count components on PCBs, utilizing multiple cameras and flexible lighting conditions, enabling real-time edge inferencing and cloud-based augmentation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If traditional AOI systems are used with dedicated single-task design, then task specialization is improved, but system flexibility and adaptability deteriorate

Engineering Contradiction:

Improvetask specializationVSAvoidsystem flexibility

Core Design Contradiction:

Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent implements a universal AOI system using neural networks that can perform multiple inspection tasks (defect detection, component counting, verification) across different PCB types and manufacturing stages. The system uses a single configurable platform with adjustable parameters to handle various inspection requirements, eliminating the need for dedicated single-task systems while maintaining high specialization through software configuration.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Measurement precision

If multiple cameras and lighting conditions are implemented, then detection accuracy is improved, but system complexity and cost increase

Engineering Contradiction:

Improvedetection accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent changes the parameters of existing cameras and lighting systems dynamically through neural network processing. Instead of adding multiple physical cameras and lighting setups, the system adjusts imaging parameters (focus, exposure, contrast enhancement) and lighting conditions through software control, achieving improved detection accuracy while minimizing hardware complexity.

Inventive Principle:

Principle #35Parameter changes

3Productivity

If real-time processing is implemented, then productivity is improved, but processing speed requirements and computational resources increase

Engineering Contradiction:

Improveinspection throughputVSAvoidprocessing speed

Core Design Contradiction:

ProductivityVSSpeed

Solution Approach 1:

The patent performs preliminary actions by pre-training neural networks with extensive defect data and component libraries before actual inspection. This pre-processing allows the system to make rapid real-time decisions during production, as the computational heavy lifting has already been done during training. The system also performs preliminary image preprocessing and feature extraction to reduce computational load during real-time inference.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12573023B2Defect detection using one or more neural networks

Publication Date: 2026.03.10 NVIDIA CORP

AI summary

Apparatuses, systems, and techniques to facilitate feature detection of a manufactured object such as a PCB using combined images of said manufactured object. In at least one embodiment, an automated optical inspection system (AOI) comprising one or more neural networks can infer based, at least in part, on combined images of a PCB the existence of defects on said PCB.