The Integrator's Perspective on Edge Computing: Why Plant Floor Experience Matters

Edge computing in manufacturing isn't a technology problem, it's an architecture problem wrapped in organizational change. After deploying edge solutions across automotive plants, pharmaceutical facilities, and chemical refineries, one truth emerges: vendors selling edge platforms don't have to live with the consequences of their architectural decisions. You do.

Plant floor reality diverges sharply from PowerPoint diagrams. Legacy PLCs running proprietary protocols, network segmentation existing more in policy than practice, operators needing real-time data without latency in safety-critical systems, this is where edge computing must actually work. The gap between vendor promises and operational reality is where projects die.

This article draws on lessons from deploying edge infrastructure across seventeen distinct industrial archetypes, from FDA 21 CFR Part 11 compliance environments to autonomous mobile robot orchestration, to provide practical guidance on what creates value versus friction. IDC forecasts global edge computing spending will reach $380 billion by 2028, with manufacturing representing a quarter of investment (IDC, 2025). Yet market projections tell only part of the story. Successful implementation requires understanding the architectural realities vendors often gloss over.

The Reality of IT/OT Convergence: The Air Gap Myth

The air gap between IT and OT networks is a myth, now supported by extensive research (Seme, 2026; ISC2, 2024). Every manufacturer pursuing digital transformation has systematically demolished whatever isolation existed. The question isn't whether your networks are converged; it's whether you've architected for that reality.

The Purdue Enterprise Reference Architecture, conceived in the 1990s, has been repurposed as a security framework for which it was never designed (Seme, 2026). Its hierarchical layers assume strict segmentation and unidirectional data flow. This fails catastrophically when IIoT sensors stream to cloud platforms and remote vendor access is contractual (Zscaler, 2024; Claroty, 2024).

The Colonial Pipeline ransomware attack in May 2021 demonstrated this failure with brutal clarity. Attackers entered through a compromised VPN credential lacking multi-factor authentication and moved laterally through porous IT/OT boundaries (Insurica, 2024; Seme, 2026). The company shut down 5,500 miles of pipeline, not because control systems were directly compromised, but because poor segmentation meant they couldn't be certain. The impact: fuel shortages, $4.4 million ransom, and national security crisis.

For integrators, edge architecture must start from Zero Trust: assume breach, enforce least-privilege access, eliminate implicit trust. Layering edge compute onto flat OT networks with VPN-based access is professional malpractice. ISA/IEC 62443 standards provide the framework for defense-in-depth and zone-based segmentation that makes Zero Trust possible (ISA, 2024).

Architecture Decisions That Actually Matter

Edge computing spans deployment models with distinct requirements for latency, security, compliance, and resilience. Three architectural decisions determine success: compute placement, data persistence strategy, and network segmentation.

Compute Placement: When Edge Isn't Really Edge

The term 'edge' has become marketing pollution. True edge computing, where compute colocation with physical processes is non-negotiable, applies in specific scenarios: sub-10ms control loops, high-bandwidth vision systems generating terabytes, and safety-critical systems requiring network-outage operation.

Consider high-speed vision inspection. An automotive stamping line producing 12 parts per minute generates 2GB of image data hourly. Transmitting raw images introduces unacceptable latency and bandwidth consumption. The solution: NVIDIA GPU-accelerated inference at the line with local SSD buffering and selective defect image upload for retraining. This is true edge computing, computation where the data is born.

Contrast with OEE tracking, where aggregating machine states every 100ms has no real-time requirement. The 'edge gateway' can be a virtualized workload in a central K3s cluster. Gartner confirms manufacturers should optimize edge investments for applications genuinely requiring low latency, avoiding edge-washing every workload (Gartner, 2024).

Data Persistence: Compliance Drives Architecture

Data persistence isn't a database problem, it's regulatory compliance disguised as architecture. FDA 21 CFR Part 11 environments require immutable audit trails: every modification generates a new versioned entry, not updates. This demands transactional relational databases with cryptographic tamper evidence. The regulation ensures electronic data remains authentic, accurate, complete, and reliable throughout its lifecycle (Unifize, 2024).

NERC CIP standards mandate 90 days of high-resolution telemetry with cryptographic timestamps. Recent 2024 updates permit cloud and edge storage with specific security precautions (Microsoft, 2024). Edge architectures must include store-and-forward buffering with backfill logic preserving temporal continuity during outages.

Network Segmentation: Beyond VLANs

Traditional VLAN segmentation creates isolation illusion while leaving lateral movement paths open. An attacker compromising an 'OT network' workstation has implicit access to every PLC and historian on that segment.

Zero Trust Networking inverts this: default deny all traffic, explicitly allow only authenticated application-to-application connections. Research demonstrates how combining advanced connectivity with edge computing enables ultra-low latency, high-resilience IIoT (Deloitte, 2024).

Implementation uses software-defined overlays with microsegmentation at the container level. Each workload exists in isolated segments; traffic between segments requires explicit policy. The blast radius of any breach is exactly one application.

What Creates Friction (And How to Avoid It)

Edge projects fail not from technical limitations but organizational friction: teams working at cross-purposes, vendor lock-in, and operational processes assuming centralized control.

Vendor Lock-In Masquerading as Integration

Industrial vendors sell 'integrated solutions' integrated only within their walled gardens. A PLC vendor's 'edge gateway' speaks fluently to their hardware but forces expensive protocol translation for competitors. The antidote: deploy edge compute on open platforms (Linux-based, container-native) and enforce protocol neutrality. Ignition Edge, CODESYS runtimes, and open-source stacks (Modbus, OPC-UA, MQTT Sparkplug B) provide vendor-agnostic connectivity. Initial integration is higher, but five-year TCO is dramatically lower.

Security Theater vs. Actual Security

Corporate IT policies written for office environments create friction in industrial systems. Mandatory 90-day password rotation breaks operator muscle memory. Forced Windows Updates during production cause HMI reboots. Antivirus scanning PLC traffic triggers timeouts.

Effective industrial security is fundamentally different. Immutable operating systems where root filesystems are read-only eliminate malware persistence without performance-killing agents. The conversation must shift from 'apply our policies' to 'achieve equivalent security outcomes with OT-appropriate controls.' EmberOS reduces attack surface by 70% compared to traditional systems (Seme, 2026).

Centralized Operations in a Distributed World

Edge computing distributes compute, but organizations fail to distribute operational authority. Remote sites generate alerts only corporate SOC can investigate. The solution: role-based operational delegation with centralizeReferencesClaroty. (2024). ICS Security: The Purdue Model. https://claroty.com/blog/ics-security-the-purdue-model

Comparitech. (2024). Ransomware attacks on manufacturing companies. https://www.comparitech.com/blog/information-security/ransomware-manufacturing-companies/

Deloitte. (2024). Connected Edge: 5G and edge computing solutions for industrial IoT. https://www.deloitte.com/us/en/services/consulting/5g-edge-computing-solutions.html

Gartner. (2024). Edge computing for manufacturing excellence. https://www.gartner.com/en/documents/5995503

IDC. (2025). Global spending on edge computing solutions to reach $380 billion by 2028. https://my.idc.com/getdoc.jsp?containerId=prUS53261225

Insurica. (2024). Colonial Pipeline ransomware attack: Lessons learned. https://insurica.com/blog/colonial-pipeline-ransomware-attack/

ISA. (2024). ISA/IEC 62443 series of standards. https://www.isa.org/standards-and-publications/isa-standards/isa-iec-62443-series-of-standards

ISC2. (2024). IT/OT convergence: Managing cybersecurity risks in industrial environments. https://www.isc2.org/Insights/2024/08/IT-OT-Convergence

McKinsey and Company. (2024). Preparing for the next normal via digital manufacturing. https://www.mckinsey.com/capabilities/operations/our-insights/preparing-for-the-next-normal-via-digital-manufacturings-scaling-potential

Microsoft. (2024). Navigating NERC CIP compliance in the cloud. https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/final/en-us/microsoft-product-and-services/industry/pdf/Microsoft-P-U-NERC-CIP-compliance-2024.pdf

Seme, E. (2026). The literal cost of seconds: Quantifying ROI in the era of Industry 5.0. Fireball Industries White Paper.

Seme, E. (2026). The Purdue Model is dead: EmberNet and the Zero-Trust imperative for industrial operations. Fireball Industries Technical Report.

Unifize. (2024). The definitive guide to 21 CFR Part 11. https://www.unifize.com/guide/the-definitive-guide-to-21-cfr-part-11

Zscaler. (2024). What is the Purdue Model for ICS security? https://www.zscaler.com/resources/security-terms-glossary/what-is-purdue-model-ics-security