Inspection specification
The defect catalog, acceptance limits and sample images that your quality team signs off before anything is built.
How we work
Five stages, each with a written exit
Every Perflens project moves from a free discovery call to a running, supported system through fixed-scope stages. Each ends with a deliverable you keep and a decision you make, whether or not you continue.
The idea
A machine vision project process built on gates
Industrial AI projects go wrong in predictable ways. The scope grows before anyone has proved the defect is visible. A pilot runs for months without agreed success criteria. A system goes live and quietly degrades because nobody budgeted for keeping it accurate.
We avoid those failures with gates. Each stage is small enough to stop after, has a fixed price or a fixed scope, and ends with a written result. You never commit to a deployment on the strength of a demo; you commit on a measurement taken on your own parts, assets or building. That is our principle of measured, not promised, in practice.
The five stages
From first call to a system that stays accurate
-
Discovery call
What happens: we ask about the defect, the asset or the building, and what it costs you when the current process gets it wrong. If we can't help, we say so on the call. Who's involved: your process owner from quality, maintenance or facilities, and a Perflens applications engineer. Your controls or IT lead is welcome too.
-
Feasibility Sprint
What happens: your samples come to our optics lab for a lighting and camera study, and we train a baseline model on them. For condition or facility monitoring, the same three weeks go into an asset criticality review and a site survey. Who's involved: our optics and machine learning engineers, with your quality engineer confirming defect labels.
-
Proof of Value
What happens: loaner hardware runs on your line in shadow mode against success criteria agreed in writing before we start. Who's involved: a Perflens field engineer and a machine learning engineer, your line or maintenance lead, and a controls engineer for a few hours to connect read-only signals. See Proof of Value.
-
Deployment
What happens: installation, PLC and plant-system integration, factory acceptance testing in our lab, site acceptance testing on your line, operator training and handover documentation. Who's involved: our field engineering team, working with your maintenance, controls and IT/OT security leads.
-
Care
What happens: drift monitoring, alert verification, retraining and signed updates keep the system as accurate as it was on the day you accepted it. Who's involved: the Perflens Care desk, plus a named reliability engineer on Care Dedicated. See Perflens Care.
Gates
Exit criteria and deliverables at each stage
| Stage | Exit criteria | Deliverable you keep |
|---|---|---|
| Discovery call | We agree there is a measurable problem worth testing, or we tell you there isn't | A written summary of the problem and, after a scoping call, a proposal within 5 business days |
| Feasibility Sprint | A written go/no-go — can the defect be imaged and detected reliably, and does the modeled return justify a pilot? | Go/no-go report, sample images per defect class, lighting and camera recommendation, baseline model results and modeled ROI |
| Proof of Value | A measured result against the success criteria signed before the start | Results report comparing baseline and measured performance, the collected data and a deployment proposal |
| Deployment | Site acceptance test passed against the agreed test cases; operators trained | Installed system, FAT and SAT reports, model card, as-built documentation and ONNX model export |
| Care | Continues while it earns its keep; reviewed with you at each renewal | Monthly health reports and an updated model card after every retraining |
Your side
What we need from you
Projects move at the speed of their slowest decision. These are the things that keep ours moving.
- A named owner with the authority to sign off success criteria and acceptance tests
- Good and defective samples for the Feasibility Sprint, shipped to our lab with a short description of each defect
- Your current figures for scrap, escapes, false rejects or downtime, so the baseline is honest
- Access to the line or site during agreed windows, including a planned stop for installation
- Time from a controls engineer for I/O and PLC changes, and from IT/OT security for the network review
- Operators and inspectors available for training and for labeling borderline cases
Documentation
Documentation you receive
Everything we write for your project is yours, and written to be used by the people who run the system after we leave.
Integration specification
Signal list, I/O map, timing budget, fail states and data feeds, agreed with your controls engineer. See integrations.
Network and security design
Every device, port, direction and zone, reviewed by your IT/OT security team, plus the SBOM for each software release. See security.
Test reports
Factory and site acceptance test reports with measured detection, false-reject and timing results.
Model card
What each model was trained on, how it was measured, its known limits and the trigger for retraining it.
As-built and training material
Drawings, wiring, the spares list, operator guides and maintenance procedures for the system as installed.
Timeline
A typical timeline for one line
| Stage | Duration | Typical elapsed time | Your team's time |
|---|---|---|---|
| Discovery call | 45 minutes | Week 0 | 45 minutes |
| Scoping call and proposal | Proposal within 5 business days | Week 1 | About an hour, plus review |
| Feasibility Sprint | 3 weeks | Weeks 2–4 | Shipping samples, confirming labels and a results review |
| Proof of Value | 6 weeks | Weeks 6–11 | A few hours a week, more during the shadow-mode weeks |
| Deployment | 6–12 weeks | Weeks 13–24 | Installation windows, integration reviews, acceptance tests and training |
| Care | Ongoing | From handover | A monthly health report review |
FAQ
Questions we hear often
Can we skip the Feasibility Sprint?
Sometimes. If you already have evidence the defect can be imaged — from an earlier study or an existing camera — or the project is condition or facility monitoring, we can start with a Proof of Value. If we think skipping the sprint is a risk, we'll say so.
What happens if the sprint says no-go?
You keep the report, the images and the baseline model results, and you owe nothing further. The report explains what would have to change — the optics, how the part is presented, or the defect definition — for the answer to become yes.
Are stage fees credited if we continue?
Yes. The Feasibility Sprint fee is 100% credited against a project or subscription started within 90 days, and 50% of the Proof of Value fee is credited to a subscription. Details are on the pricing page.
Can you work alongside our system integrator?
Yes. We can write the PLC-side logic ourselves or hand a tested integration specification to your integrator, and we share that specification with them from the start.
How do multi-line rollouts work?
After the first line is accepted, later lines reuse its specifications, models and documentation, so each one is planned as its own deployment of typically 6–12 weeks, and deployments can overlap.
Start with a 45-minute call
Tell us about the defect, the asset or the building. We'll tell you which stage to start with and what it will cost.