•

•

GAMP 5 Validation Guide for Industrial Vision and Metrology Systems

GAMP 5 Validation Guide for Industrial Vision and Metrology Systems

GAMP 5 Validation Guide for Industrial Vision and Metrology Systems

Matt Wilton

Director

GAMP 5
COMPUTERISED SYSTEM VALIDATION
MACHINE VISION
IQ OQ PQ
GAUGE R&R
21 CFR PART 11
PHARMACEUTICAL MANUFACTURING
MEDICAL DEVICES
GAMP 5
COMPUTERISED SYSTEM VALIDATION
MACHINE VISION
IQ OQ PQ
GAUGE R&R
21 CFR PART 11
PHARMACEUTICAL MANUFACTURING
MEDICAL DEVICES
GAMP 5
COMPUTERISED SYSTEM VALIDATION
MACHINE VISION
IQ OQ PQ
GAUGE R&R
21 CFR PART 11
PHARMACEUTICAL MANUFACTURING
MEDICAL DEVICES
Machine vision camera inspecting medical device product

Table of contents

Share

GAMP 5 comes up in almost every conversation we have with pharmaceutical and medical device manufacturers about automation. Usually it arrives as a single line in a purchase specification, and the questions follow soon after. What documents do we need? What can you provide? Who is meant to do the testing?

The answers are simpler than the acronyms suggest. This article covers GAMP 5 validation for vision and measurement systems, from the first specification through to the day the system is released to production, and sets out what falls to the supplier and what falls to the manufacturer. It draws on decades spent installing and validating these systems for major global medical device and pharma companies.

What is GAMP 5 validation for a vision system?

GAMP 5 validation for a vision system is the documented, risk-based life cycle that proves a camera, its lighting, its software and its reject mechanism do their intended job on a GxP production line, and keep doing it until the system is retired.

GAMP stands for Good Automated Manufacturing Practice. The guide, A Risk-Based Approach to Compliant GxP Computerized Systems, is published by the International Society for Pharmaceutical Engineering (ISPE) and is now in its Second Edition. Its logic is proportionate: test effort concentrates on the functions that can release a defective unit, and good supplier testing can be reused as evidence.

The regulations around it have moved on in 2026. FDA's Quality Management System Regulation took effect on 2 February, FDA reissued its Computer Software Assurance guidance the following day, and a rewritten EU GMP Annex 11 and a new Annex 22 on artificial intelligence are being finalised.

Which regulations apply to vision systems in the UAE and GCC?

GAMP 5 is industry guidance. The legal duties sit in the GMP and device rules of each market a plant supplies, and five texts do most of the work on a vision project:

  • EU GMP Annex 11: computerised systems, covering risk management, supplier assessment and periodic evaluation.

  • EU GMP Annex 15: qualification and validation, including URS, FAT, IQ, OQ and PQ.

  • EU GMP Annex 1, paragraph 8.32: automated inspection of sterile products must be validated to detect known defects, be equal to or better than manual inspection, and be challenged with representative defects before start-up and at regular intervals through the batch.

  • 21 CFR Part 11: electronic records and signatures for FDA markets.

  • 21 CFR Part 820: US medical device rules now incorporate ISO 13485:2016, whose clauses 4.1.6, 7.5.6 and 7.6 cover software validation.

The Saudi Food and Drug Authority has been a PIC/S participating authority since 1 July 2023, and the UAE's Emirates Drug Establishment applied for PIC/S pre-accession on 5 March 2026. The PIC/S GMP Guide is harmonised with the EU text, so Annexes 1, 11 and 15 set the direction for GMP inspection across the region.

Serialisation adds local scope. The UAE's Tatmeen and Saudi Arabia's RSD track and trace systems depend on a GS1 Data Matrix read by an industrial code reader on each pack, so that station produces GxP records. The same applies to OCR and OCV on batch codes and expiry dates.

What GAMP category is a vision system?

Most vision systems are Category 4 configured products with a few Category 5 parts. GAMP 5 Appendix M4 sorts software by how much of it is written for the job, and Category 2 no longer exists.

  • Category 1, infrastructure: Windows and camera drivers. Record the version and verify the installation.

  • Category 3, non-configured: a code reader running fixed factory functions. Test it against the requirement.

  • Category 4, configured: inspections built from standard tools in packages such as NeuroCheck or HIKROBOT VisionMaster. Test the regions of interest, thresholds, tool order and pass limits.

  • Category 5, custom: scripts, plug-ins and PLC logic written for the job. Add a design specification and code review.

Hardware has two classes: standard components such as cameras and lenses, and custom-built items such as frames and reject mechanisms.

GAMP treats these categories as a continuum. A Category 4 inspection containing a 20-line script takes Category 5 treatment for those 20 lines only.

Which documents does validation need?

These deliverables take a vision project from purchase order to release, in this order:

  1. Validation plan. Scope, responsibilities, deliverables and acceptance criteria.

  2. Supplier assessment. Sets how much supplier testing can be reused.

  3. User requirement specification (URS). Every defect by name and minimum size, product range, line speed and record retention.

  4. Risk assessment. An FMEA under ICH Q9(R1), weighted on patient impact. A missed critical defect earns full scripted testing. A display colour earns a glance.

  5. Functional design specification (FDS). Produced on every project. It describes what the system does, screen by screen and signal by signal, and how it is built to do it.

  6. Hardware and software design specifications (HDS and SDS). Optional, and added where custom hardware or code justifies them.

  7. Requirements traceability matrix (RTM). Links each URS line to its design item and test, in both directions.

  8. Protocols and reports. FAT, IQ, OQ and PQ, each pre-approved, with every deviation closed.

  9. Validation summary report. The QA signature on this document releases the system to production.

What happens at FAT, IQ, OQ and PQ?

The vision-specific evidence is what makes a protocol useful on the shop floor.

Stage

Question answered

Vision-specific evidence

DQ

Does the design meet the URS?

Pixels across the smallest defect, depth of field, lighting proven on real samples

FAT

Does the built system meet the FDS before it ships?

Agreed set of good and defective samples run at the supplier, with results recorded

IQ

Is it installed as designed?

Optics part numbers, firmware and software versions, working distance, first backup

OQ

Does every function work, including at its limits and when things fail?

Reject confirmed at maximum line speed, trigger timing, light output at both limits, user levels, audit trail, power loss recovery

PQ

Does it hold in routine production?

Challenge set of known defects mixed into live or placebo runs, across shifts and changeovers, with detection and false reject rates against agreed criteria

The FAT is where a project saves line time. EU GMP Annex 15, paragraph 3.6, lets FAT tests stand without repetition at IQ or OQ where transport and installation do not affect the function. Software logic and audit trail testing usually qualify. Optical alignment and reject timing are proven again on your line.

A separate site acceptance test (SAT) is optional, and most manufacturers fold their site checks into IQ and OQ. Injectable lines usually add a Knapp study at PQ, comparing automated and manual detection unit by unit.

Who is responsible for what?

The supplier documents and proves the system, and the manufacturer owns the requirement and the decision to release it.

Supplier

Manufacturer

FDS, plus HDS and SDS where needed

URS (AIET Group can draft it)

FAT protocol and execution

Validation plan and SOPs

IQ and OQ protocols and execution

Risk assessment ownership, with supplier input

Traceability matrix

Supplier assessment

Manuals, training and version records

Approval of every protocol

PQ support

PQ ownership, samples and release signature

When do you need a Gauge R&R?

You need a Gauge R&R whenever a system reports a dimension that someone will act on, which makes it part of validating most measurement and metrology systems.

A Type 1 gauge study comes first, checking repeatability and bias against one reference part through Cg and Cgk. The Gauge R&R then adds real parts and operators. Under AIAG guidance, %GRR below 10% is acceptable and 10 to 30% may be acceptable depending on the application. Medical device customers often run this as test method validation under ISO 13485.

AIET Group runs these studies on your own parts, on OGP video measuring systems and pharmaceutical bottle and vial measurement systems as well as inline vision gauging.

How do you keep a validated vision system compliant?

The controls are proven at OQ and run every day after release. Data integrity under ALCOA+ comes first:

  • Individual logins, with operators locked out of thresholds and recipes

  • An audit trail on every recipe and threshold change, showing who, when, the old value and the new value

  • Reject images stored against batch, time and recipe version

  • Every inspection attempt recorded, so a unit cannot be re-run until it passes

  • Backup and restore proven at OQ

Change control then protects the validated state:

  • Camera or light replacement: change record, targeted OQ tests and a challenge set run

  • Firmware or Windows patch: risk assessment, then regression testing of affected functions

  • New product: PQ extended to the new format, with worst-case bracketing across sizes

  • Routine challenge: test kit at start-up and at set intervals through the batch

  • Periodic evaluation: a scheduled Annex 11 review of changes, deviations and performance

  • Retirement: images and records kept readable for the full retention period

Can a deep learning inspection model be validated?

Yes, a deep learning model can be validated when it is frozen at release and tested on data it has never seen. ISPE published a GAMP guide on artificial intelligence in July 2025, and EU GMP draft Annex 22 shows what inspectors are preparing to ask for. The draft limits critical GMP applications to static, deterministic models, which describes a locked inspection classifier exactly. AIET Group builds to it now:

  • Model weights frozen, with version and file hash recorded in the recipe

  • Independent test data, kept away from everyone who trained the model

  • Confusion matrix and sensitivity reported for each defect class

  • Low-confidence units sent to reject or human review

  • Retraining only through change control, with the independent test repeated

  • Input drift monitored, including image brightness and product appearance

What AIET Group offers

AIET Group supplies all of the supplier validation documents and executes them: FDS, HDS and SDS where needed, traceability matrix, FAT, IQ and OQ, with PQ protocols and support. AIET Group's engineers bring decades of experience installing and validating systems for major global blue-chip medical device and pharma manufacturers, and the document set is built on those deployments. All of those skills sit in-house, alongside the machine vision systems themselves, for manufacturers across the UAE, Saudi Arabia and the wider GCC.

Speak to AIET Group

Send AIET Group your URS if you have one, your product list, samples of each defect class, line speed and the markets you supply. If you do not have a URS yet, AIET Group will write it with you. You will receive a validation scope covering the GAMP category of each component, the deliverables list and who does what. Start with the AIET Group sales team.

FAQ

What is GAMP 5 in machine vision?

GAMP 5 in machine vision is the ISPE risk-based framework used to validate cameras, lighting, software and reject mechanisms on GxP production lines. It sets how the system is specified, tested, released and controlled through its life, with test effort scaled to the risk a missed defect poses to patients.

Is GAMP 5 a legal requirement?

No, GAMP 5 is industry guidance, and the legal duties come from GMP and medical device rules such as EU GMP Annexes 11 and 15, 21 CFR Part 11 and 21 CFR Part 820. Regulated manufacturers adopt GAMP 5 because it gives inspectors a recognised, structured way to see those duties have been met.

What is the difference between FAT, IQ, OQ and PQ?

FAT proves the system meets its specification at the supplier, while IQ, OQ and PQ are the formal qualification stages on site. IQ confirms correct installation, OQ proves every function across its limits, and PQ shows the system performs in routine production against a challenge set of known defects.

Do we need a SAT as well as IQ and OQ?

Usually no, most manufacturers go from FAT to IQ and OQ on site and fold any site checks into those protocols. EU GMP Annex 15 says a FAT may be supplemented by a SAT, so a separate SAT is a project decision. It is most common where a large system is rebuilt on site after shipping.

Is automated inspection always better than manual inspection?

Automated inspection is better when validation proves it, and for sterile products EU GMP Annex 1 requires it to be shown equal to or better than manual inspection. A well-designed system applies the same criteria to every unit at line speed. The Knapp method compares the two directly, unit by unit, using a characterised challenge set.

Does CSA replace CSV for vision systems?

CSA replaces part of traditional validation practice for medical device production software, and current GAMP 5 guidance applies the same thinking to pharmaceutical systems. FDA finalised its Computer Software Assurance guidance on 24 September 2025 and updated it on 3 February 2026. Both concentrate test effort on high-risk functions and accept digital records as evidence.

Who is responsible for validating a vision system?

The supplier is responsible for the FDS, FAT, IQ and OQ documents and their execution, and the manufacturer owns the URS, PQ and final release. AIET Group writes the supplier documents and executes the tests, so the customer's QA team reviews and approves. EU GMP Annex 15 and GAMP 5 both expect this division of work.

How much validation does a simple vision station need?

A simple vision station needs the same life cycle as a large one, with far less testing where the risk is low. A presence check on a non-critical feature can take a short combined IQ and OQ protocol, while a critical defect check needs scripted OQ and a defect challenge set. The risk assessment sets the difference.

Does GAMP 5 apply to medical device manufacturing?

Yes, GAMP 5 is widely used by medical device manufacturers, although their legal duty comes from ISO 13485 and, in the US, 21 CFR Part 820. Production and inspection software must be validated for its intended use under ISO 13485 clauses 4.1.6, 7.5.6 and 7.6, and GAMP 5 gives vision projects a practical way to do it.

Looking for more? Dive into our other articles, updates, and strategies