Instrument VI · Applied systems
A power grid, re-posed as a facility-location problem.
Most engineering problems aren’t entirely new. Their underlying structures may already be understood in another discipline.
PowerFlow investigates what happens when a domain-specific engineering problem is recognized, translated, and solved through an existing mathematical framework.
What begins as transformer placement becomes an exploration of optimization, system behavior, and resilience.
Operate the instrument ↓The proposition
A problem’s apparent complexity can depend on the disciplinary framework through which it is understood. Changing that framework can reveal solutions that were previously difficult to see.
Fault response uses nearest-live-transformer reassignment. The simulation illustrates capacity and voltage constraints under disruption; it does not model switching topology, protection coordination, or full distribution-network restoration.
Engineering teams were spending days manually evaluating electrical circuit configurations to determine viable transformer placements. Each configuration required consideration of electrical loads, distances, current, and voltage-drop constraints.
The task involved repeatedly testing possible arrangements until an acceptable configuration could be found.
The difficulty wasn’t a lack of engineering knowledge. It was the amount of manual iteration required to evaluate the possibilities.
The challenge
Could this process be represented mathematically and evaluated computationally, rather than through repeated manual experimentation?
The demonstration case is a streetlight distribution circuit: a set of lighting loads, a set of candidate transformer locations, and one rule that decides everything — no node may lose more than five percent of its voltage on the way.
| Parameter — from the workbook | Value |
|---|---|
| System phase | Single-phase (K = 2) |
| Nominal voltage | 120 V |
| Power factor | 0.95 |
| Load per light | 1.14 A |
| Feeder conductor | 1/0 AL XLPE TPX UG · 0.20 Ω per 1000 ft |
| Streetlight conductor | #4 AL XLPE DPX UG · 0.42 Ω per 1000 ft |
| Maximum allowable voltage drop | 5% (6.0 V) |
The electrical engineering problem had a recognizable mathematical structure. In operations research, facility-location problems ask where facilities should be positioned to serve distributed demand while satisfying defined constraints.
The terminology changed. The underlying optimization structure was familiar.
This recognition changed the direction of the problem. Instead of manually testing electrical configurations, the problem could be translated into an established optimization framework.
The insight
Domain-specific complexity can conceal a more general mathematical structure.
A precision worth keeping: the correspondence is structural. It does not claim that voltage drop and transportation cost are the same quantity — only that they occupy the same position in the same kind of problem.
A repeatable method, not a project timeline
Identify the repetitive manual process.
Separate domain terminology from the underlying relationships.
Map electrical components onto optimization concepts.
Define variables, objective, and constraints.
Build the computational environment.
Examine solutions, behavior, and resilience.
The critical operation wasn’t replacing electrical engineering with operations research. It was identifying which relationships could be preserved while changing the problem’s representation.
The electrical system was abstracted into a set of demand points, candidate facilities, assignment decisions, and feasibility constraints. Once formalized, candidate configurations could be evaluated systematically rather than through manual trial and error.
The result was an optimization infrastructure that made the search process computational.
Two models appear on this page, and they are kept apart on purpose. The workbook is the working tool. The instrument above is a simplified demonstration built on the workbook’s electrical formula, with a few additions so that failure can be studied.
A · Decision variables
In the workbook these are the Open_Flag column and the binary assignment matrix, up to ten candidate sites.
B · Objective
In the workbook the engineer decides which candidate sites are open: Open_Flag is an input, not something the model may switch on. With the sites fixed, the objective cell is the worst voltage drop on the circuit, so it cannot be improved simply by opening every site. The template also carries a fixed cost per site and optional distance and voltage-drop cost terms, for extending the model to site selection; they are left empty.
The instrument asks the other question — how few transformers will do — and answers it by search, not proof.
C · Constraints
Every node is assigned exactly once, only to an open site, and its voltage drop must stay inside the limit. Transformer capacity is an addition of the instrument.
The instrument treats each load as its own run from transformer to node on the #4 AL streetlight conductor (0.42 Ω per 1000 ft). It does not model the 1/0 AL feeder (0.20 Ω per 1000 ft) as a separate segment.
D · Evaluation
Each node is graded PASS or FAIL against the limit. A configuration is feasible when every node passes; it is infeasible when some node cannot be brought within 5% from any open site.
The instrument’s search is a weighted clustering heuristic: it tries each transformer count in turn and keeps the first configuration that meets the targets. The result is a feasible candidate configuration. It is not shown to be the global optimum.
The human decision-maker
The optimization infrastructure evaluates configurations against formalized objectives and constraints. The engineer retains the ability to exercise judgment, override assignments, and document the rationale. In the workbook this is literal: beside the model’s answer, every node carries a Designer_Selection and a Designer_Justification.
A computational model is a representation of a problem, not the entirety of the problem. The engineer may know about site conditions, operating requirements, or physical constraints the model was never told. In the instrument, a transformer you place by hand is marked DESIGNER and stands beside the ones the search found.
The facility-location objective does not capture everything about electrical distribution. Each load here is evaluated on its own straight-line run to its transformer; real circuits share conductors, current accumulates along them, and feeder and branch segments have different impedances. The formulation is a faithful statement of this model, not of the whole discipline.
Mode I — Configuration
Move the dial and watch the search open and close transformers. Tap empty ground to place one yourself, and see what your judgment changes.
Mode II — Disruption
Inject a shock, or tap any transformer to fault it. Loads reroute to whatever is still alive. Watch whether the system holds, sags, or cascades.
Mode III — Recovery
Heal the faults, search again, or generate a new site. Compare the configuration that survives with the leanest one the search found.
Finding an optimal configuration solves a particular problem under a particular set of assumptions. But what happens when those assumptions change?
Optimization can improve utilization while reducing excess capacity. A configuration that performs efficiently under normal conditions may have less capacity to absorb unexpected disruptions.
Optimization sharpens local performance and quietly erases the slack that used to absorb a shock. Resilience never lived in the parts — it lived in the space between them.
This introduces a second problem. Not simply where to place infrastructure, but how to configure infrastructure that can continue functioning when individual components fail.
Efficiency asks
How little does the system need in order to operate?
Resilience asks
What does the system need in order to survive disruption?
These objectives can conflict. The challenge isn’t to maximize either one, but to understand their relationship and decide what trade-offs are appropriate.
Underneath is a distinction between local optimization and system-level performance. A system can improve its individual components, or its operating efficiency, without improving its capacity to absorb disruption.
This is not a universal law. Slack is one contributor to resilience; redundancy, network topology, component reliability, and recovery mechanisms matter too. The instrument isolates slack so its effect can be seen — it does not claim slack is the whole story.
Everything below is computed from the site currently in the instrument, and it updates as you operate it. It is evidence about this model’s behavior — not a report of project outcomes.
| Slack | Transformers | Max Vdrop | Min headroom | Aggregate reserve | Single faults survived | Worst single fault |
|---|
| Node_ID | Lights | I_Amps | Assigned_Facility | Distance_ft | Vdrop_V | Vdrop_Pct | Within_Limit? |
|---|
PowerFlow began with a practical engineering problem. Its development revealed two broader questions.
First: how many seemingly specialized problems are actually instances of mathematical structures that have already been explored elsewhere?
Second: when we solve a problem through optimization, are we improving the system as a whole, or merely improving the objective we’ve chosen to measure?
The first question concerns recognition and translation. The second concerns the assumptions embedded in our solutions. Both point toward a broader practice: understanding systems through their underlying relationships rather than treating their visible components as isolated problems.
The opportunity isn’t always to invent something new. Sometimes it’s to recognize what is already there, see what it makes possible, and understand what its solution leaves behind.