Inline Deep Learning PCB Inspection System

Project Overview

An electronics contract manufacturer was struggling to maintain quality while meeting customer demands for faster turnaround. Their manual inspection process—requiring trained operators to visually check every component and solder joint—was the production bottleneck on the SMT line.

With densely populated boards, the inspectors could not keep pace with the line, and inspector fatigue meant solder defects were still reaching the customer despite every board being checked by hand.

The AI Advantage

Traditional automated optical inspection (AOI) systems rely on explicit programming of acceptable and defective conditions. This approach struggles with the natural variation in electronics manufacturing and generates high rates of false calls that require human review.

AMD Machines implemented a deep learning-based approach:

Supervised Learning

The system was trained on images of both good and defective boards from actual production. The neural network learned to distinguish acceptable variation from actual defects without explicit programming.

Multi-Angle Imaging

Eight cameras positioned at different angles capture each board, enabling inspection of:

• Components hidden under larger parts
• Solder joints shadowed by tall components
• Reflective surfaces that confuse single-camera systems

Continuous Learning

As new defect types emerge, additional training images can be added and the model retrained.

Integration and Deployment

The system was designed for integration into the existing SMT line without a separate offline inspection step:

Phase 1: System installation during a planned maintenance window Phase 2: Training data collection from production boards Phase 3: Model training and validation Phase 4: Parallel operation with manual inspection for verification Phase 5: Full production deployment

Beyond Defect Detection

The real-time data collection supports proactive quality improvement:

Trend Detection

Statistical process control charts track defect rates by type. Operators can use the trends to adjust stencil alignment, reflow profiles, or paste volume.

Traceability

Every inspection result is linked to the board's serial number, creating a complete quality record for warranty analysis and customer inquiries.

Supplier Quality

Increases in defect rates by type can point to incoming component quality issues, giving the client data to take back to suppliers.