space-rha.comShort courseChapter 3 · RHA requirements

Chapter 3 · Lesson 5 of 6 · about 10 minutes

Margin needs a reason

A design margin provides headroom against named uncertainties and against design choices. The value of the margin does not by itself express a survival probability or a confidence level.

Margin needs a reason.

Three separate records

Keep three things on separate lines: the statistic taken from the environment model, the variability in the device response, and the design margin the project chose. Once they are folded into one number, it becomes hard to tell which uncertainty has been covered and which has been counted twice.

No universal multiplier

A factor of two is used in this lesson for illustration only. Nothing here recommends that two is sufficient for a given mission or technology; the right value depends on which uncertainties the margin is meant to cover and on what evidence already covers them.

Do not count an uncertainty twice

A high-percentile environment case, a conservative transport assumption and a design factor can each be covering the same uncertainty. Keep a ledger that says what each choice is for, so that any overlap is visible.

Ledger. The ledger should have a line for each source of uncertainty: the variability of the environment itself, the difference between the model and the environment, the uncertainty in the shielding geometry and materials, the lot-to-lot and process variation of the device, the uncertainty of the test, and the definition of failure in the application. Against each line, record which choice covers it.

Tradeoffs. Better evidence on the geometry or on the device response can change the decision. Removing a factor without closing the gap it was covering is a different thing, and it does not give the same assurance.

3D scene

Margin is a design choice

The same three columns as lesson 3, read as a margin trade.

  1. With the factor set to one, the required level equals the predicted local dose: the statistic taken from the environment model, with the transport behind it. That statistic already carries a confidence level and whatever conservatism went into the transport, and both belong on their own lines of the ledger.

  2. A factor of two raises the required level to 24 krad(Si). The factor is used here for illustration only; nothing in this lesson says that two is sufficient for a given mission or technology. Its value should trace to the uncertainties it is meant to cover: the model against the environment, the shielding geometry and materials, lot-to-lot variation, the test, the definition of failure.

  3. At a factor of three the required level, 36 krad(Si), exceeds the 30 krad(Si) demonstrated. Before the factor is changed, the ledger should show whether a high-percentile environment case, a conservative transport assumption and the design factor are covering the same uncertainty twice. Removing a factor without closing the gap it was covering does not give the same assurance.

The original space-rha.com environment and mission renderer (NASA imagery and spacecraft models, credits in the view), with the course scenes drawn by the same renderer. Trails, clocks and counts are illustrative; the scene computes no dose or rate.

Exercise

Take the environment allowance, the response allowance and the design factor from one requirement and write down which uncertainty each one covers, and where two of them cover the same thing.

Every conservatism should trace either to a named uncertainty or to a policy choice the project made on purpose.

Going deeper

Match the uncertainty method to the claim

A percentile, a confidence bound and a tolerance bound describe three different things, and they cannot be combined without a model that says how, together with the assumptions that model makes.

Statistical meaning. An environment percentile describes the modeled variability of the exposure. A confidence interval describes how well a parameter has been estimated from the data. A tolerance bound describes what fraction of a population falls inside a limit, with a stated confidence. They are not interchangeable, and a number from one cannot be read as a number from another.

When to escalate. Where the simple ledger is not enough, use a sensitivity analysis and, where the data support it, a probabilistic method with explicit dependence between the inputs and explicit response distributions. A Monte Carlo calculation does not remove the uncertainty in the models it is built on, and it does not take the place of the project accepting the result.

Takeaway

Write down which uncertainty the margin is intended to cover.

Check yourself

A system has margins of 2 on dose, 10 on SEL rate and 3 on upset rate. What is the system probability of success?

It cannot be computed from the margins alone. Margins are ratios against individual criteria; a probability needs a model of how each effect leads to failure, the rates with their uncertainties and the dependence between them. The margins say each criterion is met with room; they do not combine into a number.

When is a 95 % confidence statement about a rate appropriate, and when is a scenario statement better?

A confidence statement fits a quantity with a statistical model behind it, such as a count-based cross section. A scenario statement fits a choice between environment cases (quiet, worst week, peak) that carry no probability of occurrence; attaching a confidence to a scenario invents a number.

A requirement stacks a high-percentile environment case, a conservative transport assumption and a design factor of two, and a reviewer concludes that the real margin is far more than two. What does the ledger have to show before that conclusion is drawn?

Which uncertainty each of the three choices covers, because they can each be covering the same one. The ledger has a line for each source of uncertainty (the variability of the environment itself, the model against the environment, the shielding geometry and materials, lot-to-lot and process variation, the test, the definition of failure in the application) and records which choice covers it, so that an overlap is visible and a gap is visible too. Conservatisms that cover the same uncertainty generally do not add up to extra assurance, and removing one of them without closing the gap it was covering does not give the same assurance either.

A colleague cites the factor of two in this lesson as the standard radiation design margin. What does the course say about it?

The factor of two is used in this lesson for illustration only; nothing here recommends that two is sufficient for a given mission or technology, and the right value depends on which uncertainties the margin is meant to cover and on what evidence already covers them. Lesson 3.3 notes that NASA guidance uses a factor of 2 on the predicted local dose as a common policy where the environment and geometry carry ordinary uncertainty, and that the factor is a policy, not a statistic. The convention also has to be stated: the NASA guideline defines the margin as the failure level divided by the specification level and ECSS as the sensitivity divided by the received level (lesson 3.3), and the minimum the project requires should be kept separate from the ratio a test demonstrated.

A test report gives a confidence interval on the mean failure dose of the sample, and the project reads it as the fraction of the flight lot that will stay within limits. What is the mix-up?

A confidence interval describes how well a parameter has been estimated from the data; a tolerance bound describes what fraction of a population falls inside a limit, with a stated confidence; an environment percentile describes the modeled variability of the exposure. The three are not interchangeable, and a number from one cannot be read as a number from another or combined with it without a model that says how and the assumptions that model makes. A statement about the lot needs a tolerance argument, which chapter 8 takes up.

Further reading