Scaling Vision Without Compromise: Multi-Instance PAQi on High-Performance PCs

As manufacturers continue pushing for higher throughput, shorter cycle times, and more comprehensive quality control, AI vision systems are under pressure to scale—not just in accuracy, but in architecture and execution.

At LM3 Technologies, we’ve answered that demand by evolving our PAQi vision controller to run multiple parallel software instances on high-performance industrial PCs, enabling full-featured inspection across dozens of cameras—without sacrificing reliability, speed, or traceability.

The Challenge: More Cameras, Tighter Cycles, No Room for Error

Modern inspection stations often require:

• 12–30+ cameras running simultaneously

• Sub-2s cycle times

• Multiple part zones or inspection regions

• Mixed models or part variants moving through a shared cell

These are not edge cases—they’re becoming the norm. But running multiple vision systems in parallel, or trying to jam them into a single application, introduces major limitations:

• Software crashes or memory overload

• Frame drops and mis-timed acquisition

• Network congestion from raw image streams

• Cumbersome handoffs between PLCs and vision servers

• Limited flexibility when scaling to new inspections or parts

The LM3 Approach: Parallelized PAQi Architecture

Instead of overextending one oversized vision application, we run multiple self-contained PAQi instances, each monitoring a logical group of cameras. These instances run in parallel on a dedicated, rack-mounted industrial PC—pre-configured by LM3 for maximum performance and stability.

✅ Why Multi-Instance PAQi Works So Well:

• Independent processing pipelines mean each instance operates autonomously, reducing system risk.

• Tuned scheduling ensures camera acquisition, inference, and result generation are tightly managed per task.

• Shared SQL backend allows all instances to contribute to the same traceability pool, creating one unified database of inspection data.

• Modular job loading lets new part types or zones be spun up in parallel without downtime.

The Hardware: High-End Industrial PCs

For these high-demand use cases, LM3 utilizes ruggedized industrial PCs from leading suppliers, typically featuring:

• Intel Xeon or 13th-gen i9 processors (16+ threads)

• 64GB DDR5 RAM

• 1TB+ NVMe SSD for local caching

• Dual-GPU (optional) for hybrid AI workloads or redundancy

• Intel i225-V 2.5GbE and 10GbE LAN cards to handle high-resolution camera streaming with low latency

• Onboard PLC IO (via Beckhoff or WAGO) or industrial fieldbus cards to interface directly with your line

We also preload the system with:

• Optimized Windows 10 LTSC or Windows 11 Industrial OS

• Preinstalled PAQi software with service monitors

• Integrated SICK driver stacks for devices like the Ruler, PLB520, and Inspector830 AI

• Pre-configured networking for camera IP ranges and secure remote support

Camera Onboarding: From Chaos to Clean Configuration

One of the biggest bottlenecks in scaling multi-camera systems is simply getting all the devices online, synched, and coordinated. We’ve automated that with:

• PAQi’s auto-discovery tool for GigE and USB3 vision devices

• Camera grouping logic that lets integrators assign cameras to specific PAQi instances

• Pre-validation scripts that check frame rate stability, trigger health, and acquisition latency

• Structured annotation management, allowing training datasets to be organized by camera groups

Whether you’re running 6, 12, or 36 cameras, onboarding is smooth, and job logic remains modular.

Real-Time Control + Traceability

Each PAQi instance pushes data to a central SQL database with:

• Inspection images (OK/NG)

• Object detection overlays

• Model versions

• Timestamps, part numbers, and user IDs

• Pass/fail metrics and zone scores

Results are then passed upstream through industrial communication protocols such as Ethernet/IP, PROFINET, or Modbus, depending on customer standards.

At the operator level, our Results Viewer interface provides live feedback from every active PAQi instance—no matter how many cameras or jobs are running.

The Bottom Line: PAQi Isn’t Just Scalable. It’s Built for It.

With multi-instance capability, LM3 is pushing the boundary of what compact, affordable AI vision systems can do. We’re proving that:

• You don’t need a data center to run industrial AI.

• You don’t need dozens of separate controllers for 30 cameras.

• You don’t need to compromise traceability, communication, or speed.

You just need smart software architecture and the right hardware.

If you're scaling your inspection goals—or stuck trying to make an overloaded system work—let’s talk.

We’re building vision systems that don’t flinch at complexity.