Defects caught at line speed, not at the customer
Deep-learning defect detection on PCBs and assemblies at 10x manual speed with 99.5% accuracy.
Tired eyes miss defects; AI at line speed does not.
The Challenge
An electronics manufacturer running high-mix PCB assembly was catching only 85% of defects through manual visual inspection. The 30% false positive rate generated unnecessary rework, inflated scrap costs, and consumed engineering time on disposition. Customer escapes triggered warranty claims and strained key accounts. Inspection was the bottleneck: adding shifts was unsustainable, and existing AOI equipment could not keep pace with product changeovers.
The Innovoco Solution
We implemented a deep-learning computer vision system for inline PCB inspection—detecting surface defects, component placement errors, and soldering anomalies at 10x the speed of manual inspection, with configurable confidence thresholds and human review for borderline cases.
Phase 1 — Dataset and model baseline
Collected and labeled 25,000+ defect images across solder bridges, tombstones, missing components, and surface contamination. Established per-class accuracy targets with production engineering and benchmarked against existing AOI and manual inspection escape rates.
Phase 2 — Inline deployment with shadow mode
Deployed models in shadow mode alongside manual inspection for four weeks. Validated detection rates, tuned confidence thresholds by defect class, and then transitioned to inline gating with human review on low-confidence calls and new product introductions.
Key implementations
Multi-defect classification
Single-pass inference detects solder, placement, polarity, and contamination defects simultaneously—no sequential inspection stages.
Changeover adaptation
Transfer learning with small labeled sets (200–500 images) enables new product qualification in hours rather than weeks of AOI programming.
Confidence-tiered routing
High-confidence rejects go to rework; borderline cases route to human inspectors with annotated imagery and suggested defect class.
SPC integration
Defect counts, types, and locations feed statistical process control charts in real time, enabling upstream root-cause action before yield drops.
Model versioning and rollback
Every model version is tagged to production lots; rollback to previous version takes under five minutes with no line stoppage.
Technical Innovation
On-line AI scoring delivers sub-200ms per-board results at full production speed. An orchestration layer manages model versions per product, routes exceptions, and streams defect data to quality dashboards—keeping the vision system in sync with manufacturing execution without custom automation logic.
Impact
- 99.5% defect detection rate, up from 85% with manual inspection.
- 85% reduction in inspection time per board at equivalent or higher throughput.
- $2M annual savings from reduced scrap, rework, and warranty claims.
- New product qualification in hours instead of weeks of AOI programming.
Quality shifted from a bottleneck to a competitive advantage—defects are caught at line speed, customer escapes dropped, and engineering time redirected from disposition to process improvement.
Explore this outcome on your stack
We map scope, guardrails, and rollout to your data boundaries and teams—practical next steps, not a generic slide deck.
60 min · Free · No obligation
More case studies
Explore adjacent patterns from the same production playbook.
AI-Powered Business Intelligence
Ask questions in plain language, get governed answers with charts — no analysts, no ticket queue.
Read more
Customer support & voice
Triage, sentiment-aware routing, CRM-grounded answers at scale.
Read more
Regulated onboarding (KYC / AML)
Document intelligence, exception queues, immutable evidence for auditors.
Read more