<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Perflens Field Notes</title><description>Practical writing on computer vision, edge AI hardware, reliability and facility operations from the engineers at Perflens.</description><link>https://perflens.com/</link><language>en-us</language><item><title>Vision-language models on the production line: where they help and where they hallucinate</title><link>https://perflens.com/blog/vision-language-models-on-the-production-line/</link><guid isPermaLink="true">https://perflens.com/blog/vision-language-models-on-the-production-line/</guid><description>Where vision-language models help on a production line (search, triage, reject explanations, shift reports) and why they should never make pass/fail calls.</description><pubDate>Wed, 28 Oct 2026 00:00:00 GMT</pubDate><category>Computer vision</category><category>Edge AI</category><author>hello@perflens.com (Mara Lindqvist)</author></item><item><title>What a minute of downtime really costs: a calculation you can defend</title><link>https://perflens.com/blog/what-a-minute-of-downtime-really-costs/</link><guid isPermaLink="true">https://perflens.com/blog/what-a-minute-of-downtime-really-costs/</guid><description>How to calculate the cost of a minute of downtime using contribution margin, idled labor and restart scrap, with a worked example finance will accept.</description><pubDate>Tue, 20 Oct 2026 00:00:00 GMT</pubDate><category>Operations</category><author>hello@perflens.com (Grace Nakamura-Hill)</author></item><item><title>An IEC 62443 hardening checklist for edge devices</title><link>https://perflens.com/blog/iec-62443-hardening-checklist-edge-devices/</link><guid isPermaLink="true">https://perflens.com/blog/iec-62443-hardening-checklist-edge-devices/</guid><description>A practical IEC 62443 hardening checklist for edge devices — credentials, secure boot, signed updates, logging, SBOMs — plus EU Cyber Resilience Act dates.</description><pubDate>Tue, 06 Oct 2026 00:00:00 GMT</pubDate><category>Security</category><author>hello@perflens.com (Alicia Brandt)</author></item><item><title>Outcome-based pricing for industrial AI: how it works and when it fits</title><link>https://perflens.com/blog/outcome-based-pricing-industrial-ai/</link><guid isPermaLink="true">https://perflens.com/blog/outcome-based-pricing-industrial-ai/</guid><description>Outcome-based pricing ties part of an industrial AI fee to measured results. How baselines, metrics and guardrails work, and when a subscription is better.</description><pubDate>Tue, 29 Sep 2026 00:00:00 GMT</pubDate><category>Strategy</category><author>hello@perflens.com (Daniel Okafor)</author></item><item><title>Choosing edge AI compute in 2026: Jetson, Hailo or Qualcomm?</title><link>https://perflens.com/blog/choosing-edge-ai-compute-2026/</link><guid isPermaLink="true">https://perflens.com/blog/choosing-edge-ai-compute-2026/</guid><description>How to choose edge AI hardware in 2026: NVIDIA Jetson Orin and Thor, Hailo and Qualcomm Dragonwing compared by workload, power, toolchain and lifecycle.</description><pubDate>Tue, 15 Sep 2026 00:00:00 GMT</pubDate><category>Edge AI</category><author>hello@perflens.com (Theo Ramírez)</author></item><item><title>Continuous thermal monitoring vs periodic IR surveys for electrical assets</title><link>https://perflens.com/blog/continuous-thermal-vs-periodic-ir-surveys/</link><guid isPermaLink="true">https://perflens.com/blog/continuous-thermal-vs-periodic-ir-surveys/</guid><description>Periodic IR surveys miss load-dependent and intermittent faults. When continuous thermal monitoring pays off, how to set alarms, and when periodic is enough.</description><pubDate>Tue, 01 Sep 2026 00:00:00 GMT</pubDate><category>Maintenance</category><category>Facilities</category><author>hello@perflens.com (Theo Ramírez)</author></item><item><title>Why your inspection model got worse in July</title><link>https://perflens.com/blog/why-your-inspection-model-got-worse-in-july/</link><guid isPermaLink="true">https://perflens.com/blog/why-your-inspection-model-got-worse-in-july/</guid><description>Inspection models rarely fail suddenly — they drift as light, materials and machines change. How to spot model drift early and fix it before escapes rise.</description><pubDate>Wed, 19 Aug 2026 00:00:00 GMT</pubDate><category>Computer vision</category><category>Operations</category><author>hello@perflens.com (Priya Venkataraman)</author></item><item><title>Hyperspectral imaging for foreign-material detection in food</title><link>https://perflens.com/blog/hyperspectral-foreign-material-detection-food/</link><guid isPermaLink="true">https://perflens.com/blog/hyperspectral-foreign-material-detection-food/</guid><description>What hyperspectral imaging adds to X-ray for foreign-material detection in food, and how band selection, lighting, calibration and ejection make it work.</description><pubDate>Tue, 04 Aug 2026 00:00:00 GMT</pubDate><category>Computer vision</category><author>hello@perflens.com (Samuel Achterberg)</author></item><item><title>Seeing without surveillance: privacy-preserving sensing in buildings</title><link>https://perflens.com/blog/seeing-without-surveillance-facility-sensing/</link><guid isPermaLink="true">https://perflens.com/blog/seeing-without-surveillance-facility-sensing/</guid><description>How to measure occupancy without identifying anyone — on-device processing, counts instead of images, sensor choice, data minimization and governance.</description><pubDate>Tue, 21 Jul 2026 00:00:00 GMT</pubDate><category>Facilities</category><category>Security</category><author>hello@perflens.com (Alicia Brandt)</author></item><item><title>Synthetic defect data: when generated images help and when they hurt</title><link>https://perflens.com/blog/synthetic-defect-data-when-it-helps/</link><guid isPermaLink="true">https://perflens.com/blog/synthetic-defect-data-when-it-helps/</guid><description>Synthetic defect data can fill rare-defect gaps or teach a model the wrong thing. Procedural vs generative methods, domain gap, label leakage and validation.</description><pubDate>Wed, 08 Jul 2026 00:00:00 GMT</pubDate><category>Computer vision</category><author>hello@perflens.com (Priya Venkataraman)</author></item><item><title>Compressed-air leaks: the utility nobody is watching</title><link>https://perflens.com/blog/compressed-air-leaks-the-unwatched-utility/</link><guid isPermaLink="true">https://perflens.com/blog/compressed-air-leaks-the-unwatched-utility/</guid><description>Compressed-air leaks waste energy every hour a plant is pressurized. Where they hide, how ultrasound finds them, and how compressor kW data proves savings.</description><pubDate>Tue, 23 Jun 2026 00:00:00 GMT</pubDate><category>Facilities</category><category>Maintenance</category><author>hello@perflens.com (Grace Nakamura-Hill)</author></item><item><title>Air-gapped AI: running inspection and monitoring without the cloud</title><link>https://perflens.com/blog/air-gapped-ai-inspection-without-the-cloud/</link><guid isPermaLink="true">https://perflens.com/blog/air-gapped-ai-inspection-without-the-cloud/</guid><description>How to run AI inspection and monitoring without the cloud — what must live on site, how to update and retrain models offline, and the honest trade-offs.</description><pubDate>Tue, 09 Jun 2026 00:00:00 GMT</pubDate><category>Edge AI</category><category>Security</category><author>hello@perflens.com (Mara Lindqvist)</author></item><item><title>Anomaly detection or supervised classification? Picking the right defect model</title><link>https://perflens.com/blog/anomaly-detection-vs-supervised-classification/</link><guid isPermaLink="true">https://perflens.com/blog/anomaly-detection-vs-supervised-classification/</guid><description>When to use anomaly detection or supervised classification for defect inspection: data needs, thresholds, false rejects, heat maps and new defect types.</description><pubDate>Tue, 26 May 2026 00:00:00 GMT</pubDate><category>Computer vision</category><author>hello@perflens.com (Priya Venkataraman)</author></item><item><title>Microstops and the hidden factory your OEE number hides</title><link>https://perflens.com/blog/microstops-and-the-hidden-factory/</link><guid isPermaLink="true">https://perflens.com/blog/microstops-and-the-hidden-factory/</guid><description>Short stops rarely reach the downtime log, so OEE quietly hides them. How to capture microstops automatically, rank their causes and act on them daily.</description><pubDate>Tue, 12 May 2026 00:00:00 GMT</pubDate><category>Operations</category><author>hello@perflens.com (Daniel Okafor)</author></item><item><title>How to write an inspection spec your integrator can&apos;t misread</title><link>https://perflens.com/blog/how-to-write-an-inspection-spec/</link><guid isPermaLink="true">https://perflens.com/blog/how-to-write-an-inspection-spec/</guid><description>How to write a machine vision inspection spec: defect catalogue, limit samples, resolution math, false-reject budget, traceability and acceptance tests.</description><pubDate>Tue, 21 Apr 2026 00:00:00 GMT</pubDate><category>Computer vision</category><category>Strategy</category><author>hello@perflens.com (Mara Lindqvist)</author></item><item><title>Retrofit or replace? Adding AI vision to a 20-year-old line</title><link>https://perflens.com/blog/retrofit-or-replace-ai-vision-on-legacy-lines/</link><guid isPermaLink="true">https://perflens.com/blog/retrofit-or-replace-ai-vision-on-legacy-lines/</guid><description>Machine vision retrofit or replace? A framework for old lines covering cameras, mounting, encoders, PLC capacity, guarding, validation and downtime windows.</description><pubDate>Tue, 07 Apr 2026 00:00:00 GMT</pubDate><category>Strategy</category><category>Computer vision</category><author>hello@perflens.com (Jonah Whitfield)</author></item><item><title>Encoders, handshakes and reject gates: integrating vision with PLCs</title><link>https://perflens.com/blog/vision-plc-integration-encoders-reject-gates/</link><guid isPermaLink="true">https://perflens.com/blog/vision-plc-integration-encoders-reject-gates/</guid><description>A vision system is only as good as its reject. How hardware triggers, encoder tracking, FIFOs, heartbeats and timing budgets make PLC integration reliable.</description><pubDate>Tue, 24 Mar 2026 00:00:00 GMT</pubDate><category>Edge AI</category><category>Computer vision</category><author>hello@perflens.com (Jonah Whitfield)</author></item><item><title>Lighting is the lens: choosing illumination geometry for machine vision</title><link>https://perflens.com/blog/lighting-geometry-for-machine-vision/</link><guid isPermaLink="true">https://perflens.com/blog/lighting-geometry-for-machine-vision/</guid><description>How to choose machine vision lighting: bright and dark field, dome, coaxial, backlight, structured light, polarization and wavelength, matched to the defect.</description><pubDate>Tue, 10 Mar 2026 00:00:00 GMT</pubDate><category>Computer vision</category><author>hello@perflens.com (Samuel Achterberg)</author></item><item><title>From pilot to fleet: why industrial AI pilots stall</title><link>https://perflens.com/blog/from-pilot-to-fleet-industrial-ai/</link><guid isPermaLink="true">https://perflens.com/blog/from-pilot-to-fleet-industrial-ai/</guid><description>Industrial AI pilots rarely stall because the model fails. Six causes, from missing baselines to per-site customization, and a checklist for scaling.</description><pubDate>Tue, 17 Feb 2026 00:00:00 GMT</pubDate><category>Strategy</category><author>hello@perflens.com (Daniel Okafor)</author></item><item><title>Five vibration spectra every maintenance manager should recognize</title><link>https://perflens.com/blog/five-vibration-spectra-worth-knowing/</link><guid isPermaLink="true">https://perflens.com/blog/five-vibration-spectra-worth-knowing/</guid><description>Unbalance, misalignment, looseness, bearing defects and electrical faults each leave a distinct vibration spectrum. What each looks like, and what to do.</description><pubDate>Tue, 03 Feb 2026 00:00:00 GMT</pubDate><category>Maintenance</category><author>hello@perflens.com (Jonah Whitfield)</author></item></channel></rss>