AI AOI for Medical Device PCB Assembly

What makes medical device PCBA inspection different

The regulation changed in 2026. FDA's Quality Management System Regulation (QMSR) took effect on February 2, 2026. It incorporates ISO 13485:2016 by reference into 21 CFR Part 820. According to the final rule, its scope is unchanged, which means it still does not extend to manufacturers of components and parts. In practice, requirements reach a PCBA line through the device maker's supplier controls and quality agreements, and some contract manufacturers hold their own ISO 13485 certification.

Software that makes quality decisions must be validated. ISO 13485:2016 clause 4.1.6 requires that software used in the quality management system "be validated prior to initial use and, as appropriate, after changes to such software or its application," with an approach "proportionate to the risk." An AI inspection system that keeps learning raises an obvious question from quality: when does the program change, and who approved it?

Records have to tie back to the unit. Device makers work under UDI (21 CFR Part 830) and, for some devices, tracking requirements (21 CFR Part 821). At the board level, that usually means being able to show what inspection a specific serial number went through and what was found.

Acceptance class depends on the product. Depending on the device and the customer's drawing, boards are built to IPC-A-610 Class 2 or Class 3. IPC describes Class 3 as products "where continued high performance or performance-on-demand is critical, equipment downtime cannot be tolerated."

High mix, frequent revisions. Medical PCBAs are often built in small lots, with engineering changes that each require an updated inspection program. Long programming cycles slow down every revision.

Confidential designs. Device makers protect their board designs closely, and many quality agreements restrict where design data and images can go.

How DaoAI fits these requirements

• Change control over the AI. Locked parameters keep the values you set. For each check you can define the range available to AI updates, or disable AI updates entirely. Auto-programming and model updates keep your manual settings. Permissions separate who can program, tune and train from who can only inspect (full control).
• Inspection records per unit. Every inspection is stored with its images and verdict. Exports include CSV results, defect crops and the whole-board image for each unit, with defect codes mapped to your MES vendor codes (production integration). OCR reads serial numbers, barcodes and 2D codes on the board.
• Measurement-system data. Built-in tools run a repeatability study on placement measurements (X offset, Y offset and rotation) against AIAG guidelines and export Cp and Cpk to Excel. Attribute agreement analysis on pass/fail calls is not built in. See AOI Gauge R&R and Cpk.
• Data stays in your plant. All AI runs on the machine; inspection data never leaves your factory. Model training runs offline, with no image upload (data security).
• Fast revisions. Every DaoAI AOI programs itself from one golden board in 30 seconds to 5 minutes, which keeps small lots and engineering changes moving (AOI for NPI and prototype lines).
• Fewer false calls. With tuning, the false-call rate can reach about 1%.

Recommended DaoAI configurations

Compare all models on the machines page.

What DaoAI does and does not claim

DaoAI does not claim ISO 13485 certification, FDA registration or any medical-device compliance determination. Validating an AOI for its intended use is the manufacturer's responsibility. We can give your quality team what it needs for that work: how programs are locked, what data the machine stores and exports, and how AI updates are controlled. Sales and engineering support come from DaoAI's North American team in Vancouver, plus authorized partners.