Optimizing paper filler automation with industrial AI vision computers


In today’s packaging industry, efficiency and sustainability are
closely linked, with automation playing a pivotal role in minimizing material waste. To achieve precise void fill in every carton, a high-volume operation turned to machine vision and AI for smarter, real-time decision-making. By pairing smart cameras with Premio’s rugged machine vision computer at the edge, the system delivered inference speeds that kept pace with conveyor throughput, supported eco-friendly goals, and integrated seamlessly with wide-range power input and front-facing I/O for streamlined deployment.
 


Challenges

  • Requirement for a FHFL GPU to accelerate vision inference
  • Additional PCIe slot needed for high-bandwidth Ethernet and field networking 
  • All I/O to face one direction for clean cable routing and maintenance
  • Wide power input to handle 9–48VDC with protections in industrial settings
  • Rugged design to endure harsh environments 

Solution 

  • Premio’s industrial AI vision computer (VCO-6000-RPL)
  • 4-slot PCIe Gen4 expansion supporting FHFL GPUs plus dedicated lanes for add-in cards 
  • Extra PCIe for network adapters to add multi-port 2.5/10GbE  
  • Consolidated front-facing I/O for simplified cable management in tight enclosures
  • 9–48VDC input with OVP/OCP and ignition-sensing options for stable edge deployment
  • Built for shock, vibration, and temperature 

Benefits 

  • Accessible engineering support and technical expertise
  • Reliable long-term deployment in critical environments 

 

Company Overview

A global packaging innovator develops recyclable, paper-based protective materials and end-of-line automation for e-commerce, retail, and industrial supply chains. Known for machine-vision-driven systems that right-size cartons and optimize void fill, the organization helps reduce waste while improving throughput. It is now expanding AI-enabled automation to help major brands scale sustainably and exploring future opportunities in predictive analytics, advanced robotics, and fully autonomous packaging lines. 


The Challenges 

High-performance GPU acceleration 

Vision models for carton measurement and content recognition demanded a full-length GPU to meet frame-rate targets without dropping boxes from the stream. CPU-only inference introduced latency spikes during peak hours that risked line slowdowns. A platform was needed that mechanically and electrically supported FHFL GPUs with adequate cooling headroom. 

Additional PCIe for networking 

Beyond GPU acceleration, the system required a second PCIe slot to support high-performance Ethernet for multi-port connectivity with cameras, PLCs, and MES systems. Relying only on onboard LAN limited both segmentation and redundancy in critical packaging operations. By adding a dedicated NIC, the platform gained the flexibility to expand ports, increase speeds, and adapt seamlessly as production layouts evolved. 

One-direction I/O for clean cabling 

Mixed-side connectors created bulky harnesses that were hard to service on mezzanines and within machine frames. Maintenance teams asked for a single service edge so cables could be bundled and labeled logically. A front-facing I/O layout reduced accidental disconnects and shortened swap times. 

Wide-range protected power 

Lines run from diverse DC busses and can experience sags, surges, and reverse wiring during service. The compute platform needed 9–48VDC tolerance with built-in protections to avoid brownouts during motor starts. Ignition-sensing and AT/ATX modes simplified coordinated power-up with conveyors. 

Ruggedization for the edge 

Equipment near conveyors sees constant vibration, occasional shocks, dust, and variable ambient temperatures. Fan performance and card retention become critical for uptime. A design validated for 20G shock, 3 Grms vibration, and wide temperature operation was required to meet reliability goals. 

 

The Solution

Premio’s industrial AI vision computer (VCO-6000-RPL) 


The system harnesses Intel® 13th/12th Gen hybrid-architecture CPUs, combining powerful Performance-cores for intensive inference with Efficient-cores that manage parallel background tasks. Intel Thread Director intelligently balances workloads across cores, ensuring smooth responsiveness even under demanding conditions. With dual DDR5 SODIMMs supporting up to 64GB of ECC or non-ECC memory, the platform easily handles high-throughput image buffers, while triple-display capability simplifies setup and monitoring. Advanced thermal regulation sustains peak performance during continuous edge operation, delivering reliability where it matters most. 


4-slot PCIe Gen4 for FHFL GPUs 

A 4-slot configuration provides the mechanical clearance and power delivery needed for full-length, dual-slot GPUs used in real-time vision. Gen4 lanes ensure high bandwidth between GPU and CPU for low-latency inference. Locking brackets and active cooling stabilize performance in vibration-prone areas. 

Dedicated PCIe for networking growth 

Beyond the PCIe slots dedicated to GPUs, additional expansion options support multi-port 2.5/10GbE or TSN NICs, ensuring camera traffic remains isolated from broader plant networks. This separation preserves deterministic performance as both camera counts and image resolutions continue to grow. With modular flexibility, field teams can upgrade or swap NICs as needed without re-architecting the core compute system. 

Front-facing I/O for serviceability 

Consolidated I/O along one face simplifies harnessing in cabinets and machine frames, making service and integration far more efficient. The system provides multiple USB 3.2 ports, dual 2.5GbE LAN, RS-232/422/485 serial connections, and isolated digital I/O, all easily accessible from a single side. This layout allows technicians to probe or replace connections without disturbing GPU cabling, resulting in faster MTTR and cleaner, safer cable management throughout the line. 

Industrial power flexibility 

Wide 9–48VDC input with over-voltage and over-current protection tolerates noisy plant power. Optional secondary GPU power input keeps accelerators stable under load without impacting the CPU rail. AT/ATX modes and ignition-sensing align startup with upstream controllers. 

Built for shock, vibration, and temperature 

The chassis and card-retention design are validated for 20G shock and 3 Grms vibration with SSDs, supporting reliable operation near conveyors and sealers. Smart, hot-swappable fans sustain airflow around GPUs and NVMe drives. A -25°C to 70°C operating window covers unconditioned spaces and warm enclosures. 

 

The Benefits 

Accessible technical expertise

Engineering and support resources based in Los Angeles ensured smooth integration and rapid response. Having direct access to experts minimized disruption and kept operations running efficiently. 

Dependable deployment over time

The rugged, standards-compliant design delivers consistent reliability in demanding environments. Its durability allows teams to expand production with confidence, free from concerns about system failure. 

 

Conclusion 

By integrating smart cameras with a rugged, PCIe-rich industrial AI vision computer, the team automated paper filler decisions with accuracy and speed. The result is cleaner cabling, stable power, and GPU-class performance that scales as models and camera counts grow. With accessible engineering support and edge-ready reliability, packaging lines can now run faster and greener.