Concept

Engineering proxies & shorthands

Overview

An engineering proxy is a simplified stand-in — a typical value, a rule of thumb, a factored estimate, a generic scaling exponent — used in place of a rigorously derived figure when better data is unavailable or not worth the effort to obtain. Filling a model’s many unmeasured parameters with defensible proxies is the normal way an early-stage TEA gets built; the discipline is to use a class-appropriate proxy, state it as a modeling choice, and track the error it carries.

Body

What a proxy is. A proxy is a defensible placeholder for a quantity there is no first-principles data for. The corpus already runs on them: the six-tenths cost exponent ~0.6, the installation factor that turns purchased into installed cost, maintenance as a percentage of ISBL, a typical per-pass conversion, an equipment design margin of ~10–20%, a residence-time rule that fixes a vessel volume. Each replaces a calculation that would otherwise need data the estimate does not have.

Why proxies are legitimate here. The estimate’s accuracy class is wide (an order of ±30%, FEL-1/2 grade), so deriving every parameter rigorously buys precision the overall number cannot carry. A good proxy lands inside that band in minutes rather than weeks — which is exactly the deterministic, proxy-grade modeling the maturity anchor calibrates to.

The discipline that keeps a proxy honest. A proxy is not a guess. Using one well means:

• drawing it from a class-appropriate source — a reference text, a comparable process — not inventing it;
• stating it as a modeling choice with its provenance (“taking ~X as a round anchor”), never asserting it as fact;
• knowing its error band and whether the answer is sensitive to it.

Where they live. Proxies pervade both the physical and the cost sides: sizing scalars and exponents, installation and Lang factors, factored opex, typical conversions, efficiencies, and split fractions. They are the raw material of parameterizing a unit operation and of the capex build, and the decision to omit non-binding pressure and temperature is itself a proxy choice.

Limits & typical error

• A proxy on a sensitive parameter imports its full error. A proxy’s error enters the answer in proportion to that parameter’s leverage; on a parameter the result is sensitive to, a generic value can be the single largest error in the estimate, while on a non-dominant parameter the same generic value barely registers.
• A proxy used outside its class is just a wrong number. A factor or exponent borrowed from a different equipment class, service, or operating regime is no longer a defensible stand-in (the like-for-like requirement of the six-tenths rule); it imports an unrelated cost step disguised as a typical value.
• Stacked proxies multiply their error bands. A chain of factored estimates — installation × Lang × indirects × contingency — compounds, so the product’s uncertainty is wider than any single factor’s, and a string of individually “reasonable” proxies can land well off.
• A proxy frozen as fact hides its own assumption. Carrying a placeholder into a headline number without flagging it as a modeling choice removes the signal that it is uncertain — the provenance failure that makes a number look more solid than it is.
• Over-refining a non-driver while a driver stays generic. Effort spent sharpening a proxy the answer is insensitive to does not improve the estimate, and can leave a genuinely load-bearing proxy untouched — a misallocation that the magnitude of each parameter’s leverage, not its convenience, should govern.

See also

• Parameterizing a unit operation — the modeling step built largely from proxies for conversions, efficiencies, and split fractions.
• When pressure / temperature matter — the proxy decision of omitting operating conditions where they do not bind.
• Equipment sizing — where residence-time rules, design margins, and class scalars stand in for rigorous design.
• Six-tenths rule — the generic ~0.6 exponent, an archetypal cost proxy with a like-for-like limit.
• Installation factor / Lang factor — the factored-estimate proxies that turn equipment cost into plant cost.
• Mass balance — the accounting whose reaction and split parameters are commonly proxy values.

Mini-example

Sizing the ammonia synthesis reactor with proxies instead of a kinetic model: take a typical per-pass conversion (~20%, the running-example value), a nominal catalyst residence time to fix the vessel volume, and a generic vessel cost exponent (~0.6 from sizing scalars). Three class-appropriate proxies, each stated as a modeling choice rather than a derived value, put the reactor’s size and cost in the right range in minutes — without a reaction-engineering study the estimate’s accuracy would not justify.

Separately, to show where a proxy bites: because per-pass conversion sets the recycle flow and therefore the loop equipment, the answer is sensitive to it — so the generic ~20% carries its whole error into the loop sizing, whereas the generic ~0.6 exponent on a non-dominant vessel passes only a small error through. The same quality of proxy matters very differently depending on what it feeds.

See also

• Parameterizing a unit operation
• When pressure / temperature matter
• Equipment sizing
• Six-tenths rule
• Installation factor / Lang factor
• Mass balance