Services
Platform
Industries
EngagementsPricing
Resources
Company
Book a discovery call (470) 784-4045

Lifecycle

Models and alerts that stay sharp

Perflens Care looks after your inspection models and monitoring alerts after go-live — drift monitoring, scheduled retraining, analyst verification, fleet health and over-the-air updates — in three tiers with published prices and response times.

Tweezers add a tiny orange weight to a brass balance scale to bring it back into perfect balance.
FIG. 18Instruments drift; so do models — a small weight brings the scale back to level.

The problem

Go-live is where a model's working life begins

An inspection model is trained on a snapshot: one season, a few material lots, one state of the tooling. A condition-monitoring baseline is learned from a machine as it was in the weeks after installation. Then the plant keeps changing. A new film supplier, an LED panel that has dimmed, a regreased bearing that shifts the vibration signature — each moves the system a little further from what it learned.

Most plants don't have machine-learning engineers on staff, and they shouldn't need them. But a system nobody owns fades quietly until an escape or a missed failure makes the point. Our field note on why inspection models get worse in July describes how that usually unfolds.

Perflens Care makes us responsible for keeping what we installed as good as the day it passed acceptance.

Tiers

Three tiers, published

Uptime target for on-device inference is 99.9% per month, measured on the Gatepoint. Severity 1 means a Perflens system is stopping production or has stopped detecting; Severity 2 is degraded detection or monitoring; Severity 3 is a question, configuration change or report. Full terms are on the support page.
What you getCare EssentialsCare VerifiedCare Dedicated
PriceIncluded with subscriptions; $650 per station per month or $18 per asset per month on purchased systems+$650 per station per month or +$18 per asset per monthFrom $6,800 per site per month
Drift monitoring and monthly health report✓✓✓
Model retrainingTwice a yearQuarterlyContinuous, as needed
Alert verification by a Perflens analyst—Within 4 hours, business daysWithin 1 hour, 24/7
Severity 1 response4 business hours1 hour30 minutes, 24/7
On-site (contiguous US)ScheduledWithin 2 business daysNext business day
Advance hardware replacement3 business daysNext business dayNext business day + on-site spares kit
Named reliability engineer——✓

Drift monitoring

What drift monitoring watches

Everything below is computed on the Gatepoint and reported to Perflens Console. Production images do not need to leave your site.

Image statistics

Brightness, contrast and focus for every camera, trended daily — the first place a dirty window or an aging light shows up.

Confidence distribution

The share of borderline decisions and the average confidence on good parts, compared with the baseline recorded at site acceptance.

Reject mix by class

One defect class climbing while the others stay flat usually points to a process change, a new material lot or a model that has drifted.

Sensor baselines

Vibration, ultrasound and current baselines for every monitored asset, re-learned deliberately after maintenance instead of drifting silently.

Alert precision

The share of alerts that analysts confirm, tracked per asset and per rule, so noisy alerts get fixed rather than tolerated.

Device health

Gatepoint temperature, storage, uptime against the 99.9% target, Pulsenode battery levels and certificate expiry dates.

Retraining

How a retraining cycle runs

  1. Trigger

    Essentials: twice a year · Verified: quarterly · Dedicated: as needed

    A drift threshold agreed at commissioning is crossed, a known change is coming — a new SKU, supplier or tool — or the scheduled date arrives.

  2. Curate

    With your quality engineer

    We gather borderline, disputed and misclassified frames from the period, and your quality engineer confirms their labels in the Console.

  3. Challenge the current model

    Perflens ML team

    A candidate must match or beat the live model on the frozen acceptance set and on a fresh set from current production. If it regresses anywhere, it does not ship.

  4. Shadow, then switch

    Signed A/B update

    The candidate runs silently beside the live model, then takes over through a signed A/B update that can be rolled back in one step.

  5. Record

    Model card and monthly report

    The model card is updated with what changed and why, and the change appears in your next monthly report and in the audit log for your change control.

People and fleet

The desk, the fleet and the report

Alert verification desk

Reliability analysts on the Perflens Care desk, led by Grace Nakamura-Hill, review the data behind each alert before it reaches you — within 4 business hours on Care Verified, within 1 hour around the clock on Care Dedicated.

Fleet health and OTA updates

The Fleet module tracks every device, certificate and software version. Signed over-the-air updates are scheduled into your maintenance windows, with automatic rollback if a device fails its health check.

Monthly health report

One report per site — what drifted, what was retrained, alerts raised and verified, uptime against target, updates applied and open actions.

A named reliability engineer

On Care Dedicated, one engineer knows your site, attends your reliability reviews and keeps an on-site spares kit current.

Honest fit

Choosing a tier

A good fit when

  • Care Essentials suits stable products and processes where twice-yearly retraining and scheduled visits are enough
  • Care Verified suits teams that want only confirmed alerts and quarterly retraining
  • Care Dedicated suits sites where a missed detection stops a line or a customer, and an answer is needed at 3 a.m.
  • Any tier suits plants without in-house machine-learning staff

Probably not when

  • You have an in-house ML team that wants to run retraining itself — take the ONNX models and retraining recipe, and keep Care Essentials for fleet health and reports
  • Systems that will be retired within a few months; a final health check is enough

FAQ

Questions we hear often

Is Perflens Care required?

Care Essentials is included with every subscription. On purchased systems it is optional, from $650 per station per month or $18 per monitored asset per month.

How do we reach the Care team?

Through support@perflens.com or (470) 784-4045, Monday–Friday, 8:00 a.m.–6:00 p.m. Central Time. Care Dedicated customers use the same number for the 24/7 critical line after hours.

Does Care need remote access to our systems?

Remote monitoring uses an outbound, customer-controlled connection that you can switch off at any time. If you keep it off, reports and retraining run from data exports and scheduled site visits instead.

How does retraining fit our change-control process?

Every retrained model comes with an updated model card, before-and-after results on the same validation sets, and an audit-log entry. Regulated sites can hold the switch until their own approval is complete.

What happens if a Gatepoint fails?

A replacement ships in advance — within 3 business days on Essentials and next business day on Verified and Dedicated. Configuration and models are restored from the Fleet module, so the swap is a hardware job, not a re-commissioning.

Make someone responsible for the model

Tell us what you run and what a missed detection costs. We'll recommend the lowest tier that covers it.