Arc-Flash Hazards: Standards Development Framework in Facilities of National Importance in Healthcare and Data Centers

Overview

The question is whether, in facilities that serve critical infrastructure, field documentation (equipment front-panel photographs, as-builts and shop drawings) can be analysed by AI to generate arc-flash labels and spot missing data, so as to reduce the risk to electricians working on energised gear. The authors propose a standards development and implementation method, presented as an inventory of items to add to the data-collection items described in IEEE 1584.2-2025.

The method runs through two small examples: a 440 V, three-phase, 100 A residential service and a 480 V, three-phase, 1200 A commercial service. One-line diagrams were built in SKM software, with available utility fault current of 21,000 A for the residential case and 21,000-40,000 A for the commercial one. The paper also gives an AI perspective on fault prediction and early detection, fault categorisation and location, and self-healing grids, then tabulates the label-generation process and the extra data needed for training.

The main output is a data checklist: nameplate and overcurrent-protection data for switchgear, switchboards, panelboards and MCCs, trip unit model with long-time, short-time and instantaneous pickup and delay settings, conductor gauge and conduit type, and transformer impedance. It also stresses that in data centres and hospitals the arc-flash study follows the completed relay and breaker coordination study (TCCs). Listed mitigation strategies include readjusting breaker settings, current-limiting breakers, active arc-flash mitigation systems, compartmentalisation and deeper enclosures.

The limits are clear. The quantitative content is thin, the examples are small residential and commercial systems, and no measured performance of the AI model is reported. Some simplified fault-current statements, such as those about a 75 kVA transformer, should not be reused as calculation rules; real studies should rest on IEEE 1584 calculation and manufacturer data. The paper is best read as a position paper.

Key points

1. In data centres and hospitals, arc-flash studies are carried out only after the relay and breaker coordination study (TCCs) is complete.
2. Proposed additional data items: nameplate and OCPD data for switchgear, switchboards, panelboards and MCCs; trip unit model with long-time, short-time and instantaneous pickup and delay settings; conductor gauge and conduit type; transformer impedance.
3. Example studies assumed available utility fault current of 21,000 A for the 440 V residential service and 21,000-40,000 A for the 480 V/1200 A commercial service.
4. Listed mitigation strategies: readjusting breaker settings, current-limiting breakers, active arc-flash mitigation systems, compartmentalisation and deeper enclosures.
5. Table 4 gives indicative power densities of about 400-1000 W/m² for enterprise/colocation, about 2000 W/m² for hyperscale and about 2000-4400 W/m² for AI facilities.
6. The authors argue that AI analysis of field documentation could eliminate about 2% of arc-flash fatalities across all occupations and reduce the 37% of fatalities attributed to electrical hazards in 2023; this is an inference, not a measured result.

Using it in design

• Schedule the arc-flash study after the coordination study and make that order a mandatory step in the project plan and model.
• Add the paper's extra items (trip unit settings, conductor type, transformer impedance, generator and ATS data) to the site data-collection form, and flag missing photographs and as-built discrepancies.
• In high-density and AI facilities, assess low-voltage switchboard fault-current class and mitigation options (current-limiting breakers, active mitigation, compartmentalisation) at design stage.
• Do not use the simplified fault-current statements as calculation rules; base the study on IEEE 1584 methods and equipment manufacturer data.

Who should read it

Engineers and safety managers running electrical safety, coordination and arc-flash labelling studies in critical facilities.

Document contents

1. 1. Introduction
2. 2. Research Methods (2.1 Fault Prediction and Early Detection, 2.2 Fault Categorization and Location, 2.3 Self-Healing Grids)
3. 3. Implementation (440 V and 480 V, 3-phase residential and commercial systems; 3.2 Arc-Flash Label Generation)
4. 4. Standard for Training AI Model for Healthcare and Data Centers
5. 5. Discussion
6. 6. Conclusions

Related standards

• IEEE 1584-2018
• IEEE 1584.2-2025
• NFPA 70E

Read the document

Publication details

Authors

Pravin Sankhwar; Khushabu Sankhwar

Publisher

Standards (MDPI AG)

Year

2026

Volume

6

Issue

2

Pages

19

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/standards6020019

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation