SUBREL foundation
- Substation configuration comparison
- Component-outage representation
- Switching and restoration actions
- Continuity-of-supply indices
- Power-industry applications
← Tools & Methods · Data Centers hub →
Reliability Analysis Platform
InfraRel: From Component Failures to Infrastructure Consequences
InfraRel is an emerging reliability-analysis platform for evaluating interconnected electrical and supporting infrastructure systems. It extends the foundation of SUBREL, which has been used in power-industry applications to evaluate and compare alternative substation configurations.
Under active development
Why InfraRel
Reliability Does Not Stop at the Electrical Connection
Critical infrastructure depends on more than the availability of utility power. A data center, industrial facility or other essential service may also depend on local generation, batteries, switching systems, cooling, communications, controls, buildings and operational processes. A failure in any required subsystem can interrupt the service even when other parts of the system remain available.
InfraRel expands the modeled boundary so that failures, dependencies, restoration actions and their consequences can be evaluated across the complete service-delivery chain.
- Grid and local sources
- Facility power
- Cooling, communications and controls
- Critical load
- Useful service
Power is also required by cooling, communications, controls, offices and supporting facilities — so electrical availability and supporting-system availability are coupled.
Foundation
A New Extension Built on an Established Foundation
InfraRel is relatively new, but its underlying reliability methodology is not. It extends SUBREL, a reliability-analysis program developed and used in power-industry applications to compare alternative substation configurations.
The established foundation includes the representation of component outages, system configurations, switching and restoration actions, and their effects on continuity of supply. InfraRel extends this approach to interconnected infrastructure systems and broader operational consequences.
InfraRel extension
- Electrical and supporting infrastructure
- Cross-system dependencies
- Critical-load and service consequences
- Workload restoration and lost computation
- Outage-cost evaluation
- Scenario and parameter-matrix studies
InfraRel is emerging; it does not yet share SUBREL’s application history.
System boundary
An Integrated System Boundary
Utility supply and grid connections
External supply paths and interconnection points that feed the facility.
Local generation
On-site generation that can support ride-through, islanded or backup operation.
Batteries, UPS and stored energy
Short-term continuity and bridging between sources.
Automatic and manual transfer arrangements
Transfer paths that change the active supply configuration.
Switchgear and protection
Isolation, protection and switching that shape outage and restoration sequences.
Cooling and thermal support
Thermal systems required to keep critical loads operable.
Communications and network systems
Networks that support monitoring, control and operational continuity.
Monitoring and control systems
Supervisory and control functions that govern response and recovery.
Buildings, offices and supporting services
Supporting facilities that may be required for continued operation.
Critical computational or operational loads
The loads whose interruption defines service failure for the study.
The model can be simplified to match the study objective. Not every system or component must be represented at the same level of detail, but material dependencies and exclusions should be identified explicitly.
Methodology
From Failures to Service Consequences
- Define configurations, operating states and system boundaries.
- Represent components, connections and required dependencies.
- Generate applicable outage events and failure combinations.
- Evaluate switching, isolation and restoration actions.
- Determine interrupted loads, services and restoration durations.
- Calculate reliability, consequence and economic indices.
The objective is not merely to determine whether power is restored. For a data-center application, the relevant endpoint is whether useful computation continues or is restored within an acceptable time and cost.
Indices & consequences
Beyond a String of Nines
A single availability percentage is insufficient for many infrastructure decisions. InfraRel organizes complementary measures so frequency, duration, operational impact and cost can be examined separately.
Expected interruption frequency
How often disruptive events are expected to occur.
Physical interruption duration
How long physical supply or support is interrupted.
Workload interruption and recovery duration
How long useful work is lost, including restart and recovery.
Expected unserved energy
Energy not delivered because of interruptions.
Expected unserved computation
Useful computational work not completed because of interruptions.
Lost accelerator-hours
Accelerator or compute capacity-time lost to disruption and recovery.
Restoration and restart consequences
Operational effects of switching, restart and return to service.
Expected outage cost
Economic consequences associated with interruptions and recovery.
Two systems with the same availability can experience very different interruption frequencies, restoration times, lost computation and financial consequences. InfraRel therefore separates the frequency of disruptive events from their physical and operational duration.
Related reading: outage duration clocks and cost boundaries →
Related methodology: Data-Center Power Architecture & Reliability → Configuration, switching and common-mode structure on that page can inform InfraRel case definition where scoped.
Related methodology: Grid Capacity & Flexibility Assurance Framework → InfraRel may support configuration and consequence analysis within that broader assurance workflow where scoped.
Related methodology: Load Behavior & POI → Operating-state, ramp, rebound and model-requirement framing can inform InfraRel scenarios where scoped.
Uncertainty & ranges
From Individual Cases to Ranges and Tail Risk
Each InfraRel run presently produces a single set of reliability and consequence indices for one defined system configuration and set of input parameters. These point estimates are useful for comparing alternatives, but they do not by themselves describe the complete range of possible outcomes or uncertainty in the assumptions.
Range-based assessments can be developed through a structured matrix of InfraRel cases in which selected failure rates, restoration times, dependency assumptions, demand levels, recovery times and cost parameters are varied. Because individual cases can generally be evaluated independently, parallel processing can make large sensitivity and scenario studies practical.
Sensitivity studies
Vary one parameter at a time to identify influential assumptions.
Coordinated scenarios
Compare defined normal, adverse and extreme combinations of conditions.
Uncertainty sampling
Evaluate sampled parameter combinations to estimate ranges, percentiles, exceedance probabilities and tail exposure.
Running more cases cannot compensate for incomplete system boundaries, omitted dependencies or poorly supported input assumptions. Results should be interpreted together with their assumptions and data limitations.
Use cases
Potential Applications
The following are potential applications where the methodology is being developed and demonstrated. Validation and case studies are still in progress.
Comparing alternative data-center power architectures
Evaluating grid–facility–computational-load dependencies
Identifying single points of failure
Assessing cooling and communications dependencies
Comparing redundancy and restoration strategies
Evaluating UPS, BESS and local-generation roles
Estimating interruption and outage costs
Testing configuration changes and resilience investments
Supporting utility and large-load reliability discussions
Exploring other interconnected critical infrastructures
Limitations & development
Transparent About What the Model Can—and Cannot—Answer
InfraRel can organize complex reliability questions, compare defined alternatives and expose the consequences of failures and assumptions. It does not eliminate uncertainty or provide universally applicable answers. Data-center architectures, workloads, controls, recovery processes and economic consequences differ considerably among facilities.
| Limitation | Practical response |
|---|---|
| Uncertain failure and repair data | Sensitivity ranges and improved field data |
| Single-point result per run | Structured matrices of parallel cases |
| Difficult-to-quantify common-mode events | Explicit scenarios and bounded assumptions |
| Changing load and workload behavior | Multiple operating and recovery states |
| Simplified supporting systems | Selective detail with explicit dependencies |
| Facility-specific outage costs | Cost ranges and scenario analysis |
| Limited rare-event evidence | Stress cases and tail-risk reporting |
The objective is not to claim precision beyond the available evidence. It is to make assumptions explicit, identify which uncertainties influence the decision and test whether conclusions remain valid across credible conditions.
Collaboration
Case Studies and Technical Paper in Development
GR is developing comparative case studies to demonstrate the methodology, required data, reliability indices and configuration tradeoffs. Detailed assumptions, results and limitations will be presented in a forthcoming technical paper.
We welcome opportunities to explore pilot applications and collaborate with utilities, data-center operators, equipment manufacturers, engineering organizations, software developers and researchers who can contribute system knowledge, operational data and validation experience.
Data-center power & grid reliability → · From Analysis to Decision → · Utilities: grid–data center reliability →