「英中对照」霍尼韦尔联合NVIDIA与Dell推出Forge Cognition设施智能体运维与数字孪生平台

smartbuildingmag.com

15 {{ "2026-10-15T00:00:00+00:00" | date "MMM" }} '26

Written on {{ "2026-10-15T00:00:00+00:00" | date "longDate" }} Modified on {{ "2026-10-15T00:00:00+00:00" | date "longDate" }}

Honeywell Technologies introduces Forge Cognition with NVIDIA and Dell Technologies to resolve facility alarm fatigue, predictive maintenance, and energy inefficiencies.

  www.honeywell.com

Facility engineering teams overseeing commercial campuses, high-density real estate, and industrial plants face severe operational bottlenecks caused by fragmented building management systems, recurring false-positive alarms, and uncoordinated equipment control loops. Sifting manually through thousands of disparate telemetry data streams from heating, ventilation, and air conditioning (HVAC) systems, electrical distribution panels, and physical access controls slows root-cause isolation and drives up utility costs. Resolving these operational challenges requires converging information technology with operational technology (IT/OT) through physical AI models, continuous digital twin simulations, and autonomous multi-agent orchestration.

Agentic Architecture and Physical World Context
Honeywell Technologies launched Forge Cognition as a self-contained agentic artificial intelligence and simulation platform engineered for commercial buildings, university campuses, and industrial operations. Rather than functioning as a passive reporting dashboard, the platform overlays existing building automation infrastructure to contextualize equipment behaviors and map complex dependencies across multi-vendor assets.

The software foundation integrates domain expertise with telemetry streams into a structured operational knowledge base. Autonomous software agents query this operational repository to discover systemic anomalies, interpret shifting ambient conditions, and coordinate adjustments across connected supervisory control systems. Designed to support human-in-the-loop governance, the software offers manual, semi-autonomous, and fully autonomous operational modes, allowing facility managers to establish supervisory guardrails and override automated decisions at any time.

Operational Capabilities Across Facilities
The platform targets three primary operational bottlenecks across facility management environments:

• Alarm Management and Triage: Specialized software agents classify and verify incoming building management alerts, filtering transient sensor spikes and correlating related faults to reduce nuisance alarms and minimize control room alert fatigue.

• Predictive Asset Health: Autonomous models continuously monitor real-time thermodynamic and electrical metrics across HVAC chillers, air handlers, and pumps to identify mechanical degradation, guiding condition-based maintenance schedules prior to catastrophic component failure.

• Energy Intelligence: Real-time digital twin simulations assess proposed operational adjustments, modeling thermodynamic building loads to pinpoint opportunities for energy reduction without compromising indoor air quality or occupant thermal comfort standards.

Underpinning AI Compute and Simulation Infrastructure
Forge Cognition incorporates hardware and physical AI architectures developed in collaboration with NVIDIA and Dell Technologies. On-premises and edge computing tasks are driven by the Dell AI Factory with NVIDIA, deploying Dell PowerEdge XE servers alongside Dell AI Data Platform storage engines and integrated cyber resiliency frameworks.

The AI execution stack combines multiple NVIDIA technologies:

NVIDIA Nemotron open models power reasoning agents, including a dedicated True Alarm agent that classifies building events and refines inference accuracy based on facility operator feedback.

NVIDIA Omniverse libraries provide physically accurate digital twin simulations, allowing facility engineers to simulate thermodynamic changes and test control sequences virtually prior to shop-floor deployment.

NVIDIA Cosmos open models and the NVIDIA Metropolis Visual System Solutions (VSS) Blueprint enable computer vision agents to parse, index, and summarize visual and sensor telemetry feeds.

Local inference workloads execute on NVIDIA RTX PRO GPUs, with planned architectural expansion to NVIDIA DGX Spark hardware to accelerate far-edge processing directly within distributed plant and mechanical rooms.

Additional Context
This section details technical specifications and competitive benchmarking not included in the original product announcement.

Traditional building management systems (BMS) adhere to ANSI/ASHRAE Standard 135 (BACnet), Modbus, or LonWorks communication protocols, capturing point-to-point telemetry at predefined intervals without semantic reasoning. In standard commercial facilities, alarm flooding regularly exceeds the ISA-18.2 / IEC 62682 industrial alarm management benchmark, which recommends fewer than one alarm every ten minutes per operator console during normal operation. Independent field pilot deployments of Forge Cognition—including validation at NVIDIA’s 750,000-square-foot Voyager campus in Santa Clara, California—demonstrated false-alert reductions of up to 90 percent and identified potential utility consumption decreases of up to 10 percent while maintaining occupant thermal comfort scores above 95 percent.

Within the smart building and industrial digital twin market, Honeywell Forge Cognition competes with platforms such as Siemens Desigo CC with Building X, Johnson Controls OpenBlue, and Schneider Electric EcoStruxure. While conventional supervisory platforms rely on static rule-based fault detection and diagnostics (FDD) engines hosted primarily in centralized cloud environments, Forge Cognition differentiates by deploying world foundation models (WFMs) and local agentic reasoning on-premises via high-density GPU server clusters. This edge-native approach reduces dependency on continuous cloud connectivity, satisfies strict operational security requirements, and minimizes latency during critical HVAC and life-safety fault interventions.

Edited by Romila DSilva, Induportals Editor, with AI assistance.

www.honeywell.com

  Ask For More Information…