Orama.AOI and BrainChip push neuromorphic edge AI into production inspection

Orama.AOI is partnering with BrainChip to retrain and optimize Akida neuromorphic models on Orama’s industrial inspection datasets, targeting high-accuracy defect detection directly on fanless industrial PCs at the edge. The collaboration positions Akida as a production-ready platform that can be updated over the air, improving performance without hardware swaps in brownfield environments.[embeddedcomputing]

For factories, this is a concrete step toward low-power AOI that can sit inside existing HMIs, gateways, or compact IPCs near the line instead of in a central server room. Over the next 12–24 months, that makes it more realistic to retrofit visual defect detection (for surfaces, tires, castings, etc.) on legacy lines without re-racking your compute layer—provided you have enough labeled defect imagery, where platforms like Klyff can help keep datasets and labels consistent before you export models to Akida.

POSCO DX launches NPU-based Vision AI platform for unstructured plant data
POSCO DX unveiled an unstructured data analysis platform that uses a domestically produced NPU to power its Vision AI stack, focusing on video data from industrial sites. The platform standardizes recurring functions such as model management, performance metrics, and application history while using GPUs for training and NPUs for real-time inference in equipment control systems. Plants should read this as a blueprint for hybrid GPU+NPU architectures that cut inference infrastructure costs by around 50% and power by roughly 90%, making always-on visual inspection or safety analytics more economical at scale.[biz.chosun]

Advantech IWS brings “agentic” AI onto the factory floor
Advantech announced its Integrated WISE Solution & Station (IWS) platform at Taipei Automation 2026, describing it as an AI software–hardware stack that tightly couples AI agents with edge computing to support smart production, operational decisions, and AI infrastructure in one architecture. IWS is designed so AI agents can sense, analyze, and assist with tasks directly at the line while keeping data local for security and latency reasons—summed up as “agents go to the shop floor, intelligence doesn’t leave the factory.” For manufacturers, this signals that next-wave digital assistants (LLM-style copilots for maintenance, quality, or scheduling) will increasingly run beside PLCs and MES; you’ll still need clean, contextualized data and well-governed prompts, where platforms like Klyff can help standardize event and label streams before models are deployed into IWS cells.[money.udn]

Intozi and Axelera push higher-density Vision AI at the edge
Intozi Tech is porting its Ikshana AI video analytics platform onto Axelera’s Metis AI Processing Unit, reporting 2.7× higher throughput per stream versus its previous baselines. The companies say this allows more cameras per device and Vision AI deployments across manufacturing, oil and gas, and other industrial sectors without cloud dependence or high-cost GPUs. For AOI and safety analytics, this effectively raises the ceiling on how many feeds you can process per edge box, which matters if you’re trying to cover entire lines or plants with visual inspection and behavior monitoring while keeping inference close to the process.[einpresswire]

TDK SensEI edgeRX Pro adds acoustic and magnetic sensing to PdM nodes
TDK introduced the SensEI edgeRX Pro, a new sensor node that extends its edgeRX predictive maintenance platform with vibration, acoustic, magnetic, temperature, and rotational motion sensing in an IP67 housing. The device supports both long-life battery power (up to about 10 years) and wired USB power, enabling higher sampling rates and more complex on-device AI models when mains power is available. For maintenance teams, this means you can move beyond vibration-only monitoring to combine sound and magnetic signatures for applications like compressed air leak detection, alignment issues, and bearing anomalies, feeding richer signals into your edge models or platforms like Klyff for labeling and feature curation before those models are rolled out fleet-wide.[tdk-electronics.tdk]

Basler and Syslogic harden multi-camera vision for harsh environments
Basler’s ace 2 GMSL cameras are now tightly integrated with Syslogic’s rugged computers, forming a combined platform for demanding edge vision applications such as mobile robots, agricultural vehicles, and construction equipment. The joint solution is built around Jetson-based Syslogic systems that can ingest up to eight GMSL cameras and fuse data from LiDAR, radar, and GNSS sensors while running AI inference directly at the edge. Although much of the marketing targets off-highway and mobile machines, the same pattern—rugged, multi-camera edge boxes plus a consistent camera stack—is directly applicable to intralogistics vehicles, automated warehouses, and flexible inspection cells where cabling, vibrations, and temperature fluctuations make standard IPC + USB-camera setups fragile.[embeddedcomputing]

Primax Tymphany showcases Vision + Acoustic Edge AI for inspection and AMRs
At Automation Taipei 2026, Primax Tymphany Group is demonstrating Edge AI Sensor Fusion that combines vision, acoustics, and edge computing across autonomous delivery robots, factory AMRs, and smart manufacturing inspection systems. The company’s Smart Manufacturing Inspection solution uses vision to monitor product appearance and workflows while acoustic edge AI analyzes equipment sounds and vibrations to turn experience-based judgments into quantifiable indicators and early-warning signals. Over the next 12–24 months, expect to see more vendors pair microphones with cameras on inspection rigs and mobile robots, allowing you to detect both visible defects and “sounds wrong” conditions with a single edge box—again shifting the bottleneck toward managing and labeling multimodal data, where tools like Klyff can keep the vision and acoustic streams aligned.[finance.yahoo]

Memory supply is now part of the edge computing decision — Why RAM and supply terms now gate your edge AI plans.
This analysis argues that edge computing has effectively become the default answer on the factory floor, but that many deployments now stall on memory constraints and long lead times rather than CPU choices. For manufacturing AI teams, it is a reminder to treat DRAM and flash capacity—and vendor supply commitments—as first-class inputs when sizing edge boxes for computer vision, predictive maintenance, or local analytics.[iottechnews]

On-Board Computers: Intelligence for Autonomous Vehicles and Mobile Machines — Lessons for factory AMRs.
Syslogic’s August tech brief explains how on-board edge computers orchestrate sensor fusion, AI inference, and control for autonomous vehicles and mobile machines in harsh conditions. The same architectural choices—ruggedized compute close to sensors, deterministic networking, and tight power envelopes—map directly to AMRs and forklifts in warehouses and plants that need reliable on-vehicle AI for navigation, safety, and inline inspection.[syslogic]

Where to Put Intelligence: Edge AI Architecture for Factory Maintenance — A practical layer model for PdM.
This IIoT World article breaks down a four-layer architecture—sensor, gateway, cloud, and enterprise—for placing AI workloads, with concrete examples of which maintenance and quality tasks belong at each layer. It’s a useful mental model if you’re wrestling with whether to run anomaly detection on sensors, in gateways, or in the cloud, and how to push trained models back down to the edge for fast, contextual PdM decisions.[iiot-world]

Azilen expands Industrial AI and IoT Engineering — Turning fleet signals into operational intelligence.
Azilen’s announcement describes how it is bundling sensorization, industrial IoT, edge computing, and AI engineering to transform raw machine signals into real-time operational insights, including an edge platform deployed across a fleet of over 1,000 commercial dishwashers. While the example is non-factory, the architecture—retrofit hardware plus edge analytics and secure connectivity—is directly analogous to how you might modernize legacy process equipment, with data quality and failure-mode labeling still being the critical path where something like Klyff can shorten setup time.[aol]

Send up details on what you would like us to cover in Manufacturing Intelligence

TWIMI is published weekly. The scope covers developments from the prior 7 days or earlier if that ties into the stories for this week. No vendor relationships influence coverage. Forward to a colleague in ops, quality, or IT/OT — the more disciplines reading from the same page, the faster deployments happen.

Team twimi