AOI Machine Product

Overview

Automated optical inspection (AOI) is the primary defect-detection method in modern SMT assembly. An AOI machine uses multiple high-resolution cameras (at 0°, 45°, and 90° viewing angles) with programmable LED lighting to inspect populated PCBs immediately after reflow soldering. The system captures images of every solder joint and component, then runs defect-detection algorithms to classify each connection as acceptable, marginally acceptable, or defective. Defects are logged with high-resolution images and coordinates, enabling rapid root-cause analysis.

AOI systems are essential for yield control and traceability. Defect rates from stencil printing, pick-and-place, and reflow reach 1–5% initially; AOI catches 95–99% of defects (depending on type and tuning), preventing defective boards from reaching the customer. For high-reliability applications (automotive, medical), all defects are documented and traced back to the process step responsible, driving continuous improvement.

Imaging and Optical Path

The Camera Head Assembly includes three or more cameras positioned at different angles. The 0° (coaxial) camera captures top-down images of solder joints and components; the 45° and 90° side cameras detect solder fillet height and geometry from oblique angles, essential for identifying insufficient solder or dry joints. Some advanced systems add additional angles (135°, 180°-backlit) for comprehensive coverage.

Each camera is equipped with macro optics (1×–10× zoom) allowing inspection of passives (0402 size, 1 mm × 0.5 mm) through large packages (100+ mm BGA). The Focus Motor automatically adjusts focus as the board moves on the conveyor, compensating for height variation due to component height and board flatness (±1–2 mm typical range).

The Lighting Array provides multiple light sources: (1) coaxial LEDs ring-lighting from above (0°), illuminating solder joint tops; (2) oblique LEDs at 45°, creating shadowing that reveals fillet geometry and height; (3) backlighting (180°), illuminating component outlines and detecting missing or misplaced parts. LED intensity is independently controlled via LED Controller, allowing recipe-based tuning for different board types.

Defect Detection Algorithms

The Vision Processor executes defect detection at real-time speed (>30 FPS on 600 mm × 300 mm boards). Algorithms include:

1. Solder joint inspection: Template matching and machine learning models compare captured images against reference images of acceptable joints. Deviations in solder area, height, fillets, or shape trigger defect flags.

2. Solder bridge detection: Morphological image processing identifies connections between adjacent pads where solder has bridged (causing electrical shorts).

3. Open circuits: Insufficient solder or missing paste detected by solder-area measurement or geometry analysis.

4. Solder balls: Spherical solder particles detected as isolated bright spots outside intended pad areas.

5. Component inspection: Vision verifies component presence, orientation (polarity), and placement accuracy using edge detection and barcode/marking validation.

6. Lifted or missing components: Backlighting detects missing components; coaxial images reveal partial lift-off (component partially detached post-reflow).

Reference databases (Reference Database) store "golden" images of correctly assembled boards and known defects. The system is trained during first-article inspection (FAI), capturing good and bad assemblies and building the inspection recipe.

Machine Learning and Adaptive Detection

Modern AOI systems integrate machine learning for continuous improvement. The ML Training Module analyzes field failures and escapes (defects not caught by AOI), retraining the defect classifiers to improve detection. Active learning techniques flag borderline images for human review; as operators classify them, the model updates automatically.

Transfer learning (pre-training on generic defect patterns, then fine-tuning on specific products) accelerates recipe development. Some systems require <10 minutes of board exposure to tune; others require 1–2 hours of manual image labeling for high confidence.

Integration with SMT Line and Defect Reporting

The AOI machine is typically positioned after reflow, before final inspection or rework. Defect boards are either flagged for automatic rejection (conveyed to a rework station) or logged for statistical analysis. The Main Controller interfaces with the factory MES (manufacturing execution system) via Ethernet, reporting defect rates, locations, and types in real-time.

Defect images are stored on the Storage system, archived with batch identifiers for traceability. When a field failure occurs, engineers can query the AOI database, viewing the historical inspection image of that board, confirming whether the defect was present at assembly or developed in field.

Limitations and False Positives

AOI is not 100% accurate. False positive rates (incorrectly rejecting good boards) of 2–5% are typical; false negatives (missing actual defects) of 1–5% also occur. Process variation, lighting shadows, and component surface finish can trigger false positives.

Reducing false positives requires careful tuning during recipe development. Machine learning models trained on inadequate data (e.g., <100 boards) tend to overfit, classifying minor acceptable variation as defects. Best practice is to train on 500–1000 boards, a significant investment but necessary for mature production.

Some defect types are difficult to detect optically: buried voids (internal cavities in solder balls), cold joints (poor wetting that appears superficially acceptable), and latent opens (hairline cracks not yet separated). X-ray or acoustic inspection (C-SAM) are complementary methods for detecting these subsurface defects, used selectively on high-reliability products.

Rework and Defect Disposition

Boards flagged by AOI flow to a rework station where technicians manually inspect, analyze the defect, and repair (re-solder, replace components, etc.). Rework stations are equipped with fine-tipped soldering irons, solder wick, and component removal tools. Complex repairs (BGA rework) require specialized equipment (rework ovens, X-ray for visibility).

Rework cost is typically $5–$50 per board depending on defect complexity. For high-volume products, even a 0.5% rework rate translates to significant cost; yield improvement initiatives target <0.2% rework.

Solder Joint Quality Metrics

AOI systems measure solder joint parameters:

• Solder area: Total solder coverage on pad, typically 60–85% of pad area (100% coverage is rare and may indicate excess solder).
• Fillet height: Solder fillet rise above the component lead, typically 0.2–0.8 mm for fine-pitch, 1–3 mm for larger packages.
• Fillet shape: Concave (good wetting) vs. convex or domed (poor wetting, risk of cracking).
• Solder ball count: Number of balls; >3 is typically unacceptable.

Advanced systems include 3D height measurement via structured-light or laser triangulation, enabling quantitative fillet geometry analysis. These systems are slower (~0.1 m/sec) but provide high-confidence joint quality assessment.

Calibration and System Maintenance

Regular calibration keeps AOI accurate. Monthly or quarterly calibration using Calibration Board (reference board with known good and bad joints) verifies system detection rates. Cameras are cleaned regularly (dust degrades image quality), and LED arrays are tested for intensity drift.

Image processing recipe maintenance is critical. As products age, component variations accumulate (different manufacturers, lot variations). Recipes must be updated periodically; static recipes drift and accumulate false positives. Best practice is quarterly recipe audits, comparing recent inspection data against specification and retraining models if drift is detected.

Throughput and Production Planning

AOI throughput is typically 0.2–0.5 m/sec conveyor speed. For a 600 mm × 300 mm board with 500 components:

• 0.2 m/sec → 3000 mm / 200 mm/sec = 15 seconds per board = 240 boards/hour
• 0.4 m/sec → 7.5 seconds per board = 480 boards/hour

Faster inspection requires simpler boards or lower resolution; complex boards with fine-pitch BGAs require slower speeds for accurate detection. Multi-camera systems can increase speed: while one camera scans the top side, a second scans the reverse side in parallel, nearly doubling throughput.

Competitive Landscape and Variants

High-end AOI systems (from vendors like Koh Young, CyPhy, Mirtec) offer advanced features: 3D height measurement, machine learning models, multi-camera parallelization, and integration with AI-powered defect prediction. These systems cost $1–$3M but are justified in high-volume, high-complexity production.

Budget systems ($300k–$700k) offer basic optical inspection, sufficient for low-complexity boards (simple passives, no BGA) or research environments. These lack advanced features but cover fundamental defects adequately.

Inline X-ray AOI systems, while more expensive ($2–$5M), complement optical AOI for detecting solder voids in BGAs and high-aspect-ratio components, critical for automotive and aerospace products.

Continuous Improvement and Data Analytics

The defect logs from AOI drive process improvement. Pareto analysis of defect types identifies the top 3–5 defect modes, focusing efforts on root-cause elimination. If solder bridges are 40% of defects, investigate stencil aperture design, solder paste volume, or reflow profile. If component lifts are 30%, focus on adhesive curing or thermal profile.

Statistical process control (SPC) charts track defect rate over time, detecting process excursions. A sudden increase in a particular defect (e.g., opens), correlated with a reagent lot change or equipment maintenance, enables rapid diagnosis and correction. This data-driven approach is foundational to modern high-yield assembly.