Combined environments and evidence of resistance

A spacecraft surface faces more than one input

Radiation, vacuum, thermal cycling and reactive species can act together in some service contexts. One exposure can change the surface or defects encountered by another. Independent single-environment tests provide useful pieces but may miss coupling.

A surface may pass separate radiation and thermal-cycle tests yet fail when the changed state left by one exposure encounters the other. A combined claim therefore needs more than two labels on a material. Identify the state that carries the history: for example, accessible cracks, reactive sites, remaining fibres or a surface film. Then ask how another input acts on that state.

Three different questions follow. A physical coupling is a causal route by which the response to one input depends on another. A statistical interaction is a departure from a chosen response model, such as adding two strength changes in MPa. A sequence effect means that reversing the order changes the final state. These can be related, but neither a nonzero interaction nor an order effect uniquely identifies a microscopic mechanism.

The preceding lessons supply radiation damage, outgassing and chemical attack as individual processes. Here a researched oxidation example establishes a coupling route; an invented four-condition record teaches its quantitative evidence; and a separate invented state model makes order dependence calculable. Neither invented calculation is a fitted prediction for the researched composite.

A crack changes access; reaction changes the load path

In studied carbon-fibre/silicon-carbide composites, oxygen could enter along matrix and coating cracks and attack carbon fibres and their pyrocarbon interphase. Sustained-load air exposures showed increasing strain that was largely absent in matched argon exposures. Reported cross-sections located reacted fibres near cracks and edges. The proposed interpretation links fibre-section loss, debonding and redistribution of load to surviving fibres. The spatial evidence and atmosphere comparison support that interpretation together; the strain curve alone would not isolate oxidation from mechanical creep. Possible oxidation during cooling after fracture was considered, rather than simply treating every observed reaction feature as in-service damage.

This is a two-way route: loading can alter crack access, while attack changes the structure that carries load. Crack closure and an oxide that seals access can oppose it. An open pathway is therefore an evolving state, not a permanent property specified by the original material name. This research does not supply a universal composite lifetime or prove that every hotter or more highly loaded state gives the same ranking.

A useful local force balance is simpler than the real composite. Let a fixed force P be carried uniformly by an effective load-bearing area A_eff. The mean stress on that area is σ_eff = P/A_eff. If reaction removes 20% of A_eff while P remains fixed, σ_eff rises by 1/0.80 = 1.25. This is a 25% stress increase on the remaining effective area; it is not a 25% rise in nominal stress calculated from the unchanged original coupon area.

The balance explains a possible positive feedback: a smaller load path carries greater stress, which may promote further mechanical damage and access. It does not calculate crack-tip stress, fibre strength, matrix load sharing or time to failure. With displacement held fixed instead of force, the force can fall as stiffness falls. A loading-control condition is consequently part of the mechanism argument, not a detail to omit.

Keep all four exposure combinations and the specimen identities

The table is an original invented record of residual tensile strength S in MPa. Factor A switches a sustained tensile load during exposure off or on. Factor B switches an inert atmosphere to the specified oxidising atmosphere. In every cell, three independently prepared coupons remain intact through the same scheduled exposure endpoint and are then unloaded and tested destructively under the same residual-strength conditions. Each number is a different coupon, not another reading of one specimen. These are not data extracted from the composite research.

For this illustrative comparison, specimens begin in a matched state and are randomly allocated to the four conditions, with run order balanced across the conditions. Actual specimen temperature–time histories, exposure duration, specimen geometry, atmosphere pressure and intended flow conditions are matched; A and B are the intended changes. In a real study, loading, atmosphere or radiation can themselves change specimen temperature, so matching a heater setting would not establish this premise.

The unexposed-to-A-and-B cell is still given the common thermal and handling history in inert gas. Otherwise a difference attributed to oxidation could include a different thermal history. Record early failures separately if they occur: the mean residual strength of surviving coupons alone can favour a condition that already destroyed its weakest specimens. The invented table explicitly has no such losses.

A complete two-factor comparison lets us ask how the A contrast changes when B changes. It does not force the treatments to isolate just one microscopic feature: loading can change several aspects of the structure, and atmosphere can change more than one reaction. Mechanism evidence must still distinguish those routes.

Define the reference model before calling a result synergistic

Write the four cell means as S̄_00, S̄_10, S̄_01 and S̄_11, with the A index first. The load contrast in inert gas is S̄_10 − S̄_00 = −15 MPa. In oxidising gas it is S̄_11 − S̄_01 = −60 MPa. Their difference is D = S̄_11 − S̄_01 − S̄_10 + S̄_00 = −45 MPa. The load-associated decrease is 45 MPa greater in the oxidising condition on this additive strength scale.

An additive no-interaction model predicts S̄_11,add = S̄_10 + S̄_01 − S̄_00 = 255 MPa. It predicts a combined loss of 45 MPa from the 300 MPa reference. The supplied combined mean is 210 MPa, a loss of 90 MPa. D is the observed-minus-additive-predicted difference. Here D is explicitly a difference of two conditional contrasts; it is not a coded regression coefficient with a different normalization.

An alternative reference assumes independent fractional retention, not independent MPa increments. It predicts S̄_11,rel = S̄_10 S̄_01/S̄_00 = 256.5 MPa. The dimensionless ratio I_rel = S̄_11 S̄_00/(S̄_10 S̄_01) is about 0.8187 rather than 1. The combined strength is below both chosen predictions in this record. The word “synergy” needs the response, scale, reference and sign convention: a negative strength interaction corresponds here to an extra loss.

Scale choice can matter even without noise. If a different record kept the first three means but had S̄_11 = 256.5 MPa, fractional retention would multiply exactly, yet D would be +1.5 MPa rather than zero. Neither scale is automatically a physical law. Choose a model that answers the intended engineering question and test its assumptions; changing scales after seeing a result can change the apparent claim.

For the supplied independent replicates, each cell has sample standard deviation s = 4 MPa. The estimated standard uncertainty of each mean from between-coupon scatter is s/√3 = 2.3094 MPa. For D, independence gives u_rep(D) = √(s²/3 + s²/3 + s²/3 + s²/3) = 4.6188 MPa. The plotted bars are one such standard uncertainty of a mean, not a 95% confidence interval. Three coupons per cell give a fragile scatter estimate; the arithmetic does not include unquantified between-batch effects or measurement corrections.

If errors of cell means are correlated, the variance is cᵀΣc, with coefficient vector c = (1, −1, −1, 1) in cell order (00, 10, 01, 11), and Σ their covariance matrix. The positive and negative coefficients make covariance signs matter. An equal common additive measurement offset cancels in D because the coefficients sum to zero; a cell-dependent temperature or calibration error need not cancel. No p-value, coverage probability or population resistance guarantee follows from the displayed scatter bars.

An exposure-control problem can imitate a material interaction

Changing an exposure source can change several inputs at once. In an atomic-oxygen/vacuum-ultraviolet study, changing source conditions altered both oxygen arrival and radiation. The researchers developed more independent control of the two. Their selected FEP-relative-to-Kapton erosion comparison showed no detected VUV dependence over the tested range, while unresolved facility-to-flight differences remained. Oxygen fluence inferred from a reference polymer was an effective damage-equivalent quantity, not a direct absolute atom count. These findings make the control and response definition part of the inference; they do not establish general absence of radiation–oxygen coupling.

For the load–oxidant record, a credible mechanism argument would combine the response contrast with observations of accessible cracks and reaction location, and with evidence that the intended gas and specimen temperature were actually achieved. If the combined condition simply ran hotter, the strength interaction could partly describe that temperature change. Holding the measured local temperature fixed would address a different question: load and oxidant effects at matched temperature. Both are legitimate questions if the distinction is explicit.

The tested response also matters. Residual strength after exposure, strain accumulated during exposure, mass change and reflectance are different endpoints. A material can retain one while losing another. Converting an erosion comparison to a strength guarantee, or a mean coupon strength to a component survival probability, requires additional evidence. Acceleration by increasing a flux or temperature can change transport, sealing or recovery and thus the coupling itself.

Make the state carried between exposures explicit

Now use a separate invented model to isolate order. Let p be a dimensionless accessible-pathway state, bounded between 0 and 1, and q the remaining fraction of a reactive inventory. q is neither a measured tensile-strength fraction nor a fraction of intact load-bearing area. An opening operation O creates persistent access. It could idealise irreversible microcracking or debonding; it is not an elastic crack aperture assumed to remain open after unloading.

While O is active, dp/dt = α(1 − p); otherwise dp/dt = 0. While a reactive exposure R is active, dq/dt = −βpq; otherwise dq/dt = 0. α and β have units h⁻¹ when time t is in hours. Reaction consumes the inventory at a rate set by its amount and the current access. The model has one-way state coupling: q depends on p, but consumption does not change p. It intentionally omits spatial diffusion, crack closure, healing, reaction products and mechanical redistribution. It illustrates history, not the full composite feedback.

Start all histories at p₀ = 0.100 and q₀ = 1. Take α = ln(2) h⁻¹ and β = 2.00 h⁻¹. Every history has one hour of O and one hour of R within a common two-hour calendar: O then R; R then O; or simultaneous O and R during the first hour followed by one hour of quiet inert holding. During that quiet interval both derivatives are zero. The same active inputs and final clock time therefore do not conceal different total active exposure durations.

During an opening interval beginning at state p_i, solve d(1 − p)/dt = −α(1 − p): p(t) = 1 − (1 − p_i)e^(−αt). One active hour takes p from 0.100 to 0.550 in all three histories. For any reactive interval, integration gives ln(q_f/q_i) = −β∫p(t)dt. It is the access-weighted reactive time, not just the hour count or final p, that controls q.

For O then R, access stays at 0.550 throughout R, so ∫_R p dt = 0.550 h and q_f = e^(−1.100) = 0.33287. For R then O, access is only 0.100 during R, so ∫_R p dt = 0.100 h and q_f = e^(−0.200) = 0.81873. Subsequent opening changes p but cannot restore the consumed inventory.

For simultaneous O and R, integrate the evolving access over the first hour: ∫₀¹p(t)dt = 1 h − (1 − p₀)[1 − e^(−α·1 h)]/α = 0.350787 h. Thus q_f = e^(−2.00 × 0.350787) = 0.49580. Quiet holding changes neither state. The final p is 0.550 in all three cases, but q retains the history. In this model, opening first consumes the most inventory and reacting first the least; simultaneous exposure lies between them.

The ordering is conditional. If α = 0, access never changes and the three q results coincide. If pathways close after O, or reaction seals or creates pathways, these equations must change; the ordering need not survive. An interaction estimated from simultaneous exposures also does not determine the result of a reversed sequential exposure without such state information.

A bounded claim identifies what was combined

Report sequence, simultaneous conditions, duration and response criteria. Compare relevant controls and examine whether the combined mechanism changes. State untested combinations explicitly rather than promising universal environmental resistance.

For the invented four-cell record, suppose the declared descriptive criterion is a mean residual tensile strength of at least 240 MPa at the scheduled endpoint. Each single-active cell exceeds it, but the combined cell does not: 285 and 270 versus 210 MPa. This is a statement about those supplied means, not a validated population acceptance rule. All individual values, early failures, uncertainty and the actual sampling level belong in a real resistance assessment.

A defensible conclusion would name the specimen state, the load and atmosphere histories, measured temperature, duration, endpoint and response scale. It would report the four means and the −45 MPa conditional-contrast difference, and describe any crack/reaction evidence separately. It would not claim a tested coating is “environment-proof”, assign the inventory model a strength threshold, or treat an untested radiation spectrum or reversed sequence as covered.

The transferable habit is to identify the evolving state, specify the comparison and ask what would overturn the proposed coupling. A spatial reaction pattern without a matched response comparison leaves one gap; a nonadditive response without evidence of state change leaves another. Combined-environment resistance is an evidence claim about a stated history and response, not a material label.

Worked examples

Try the reasoning

1. Can exposure order affect response? Using the supplied persistent-access model, explain why identical final p does not imply identical q.

   Show the answer and reasoning

   Yes. All three final p values are 0.550, but q depends on ∫ during R of p dt, which is 0.550000, 0.100000 or 0.350787 h for the different histories.

   The first exposure changes the starting state for the next. Final access alone does not record how long a reactive species encountered that access. With the supplied β, q is about 0.33287, 0.81873 or 0.49580.

2. What should a resistance claim specify? Apply the answer to the invented four-cell strength comparison and name a claim it cannot support.

   Show the answer and reasoning

   Environment, duration, criterion and specimen state, together with sequence, actual specimen temperature, loading control, response definition, replication and uncertainty. It cannot support an untested radiation or component-life guarantee.

   Without them, the claim has no clear test. The supported numerical statement is a difference of conditional mean-strength contrasts for the supplied conditions; a mechanism attribution needs additional state evidence.

3. A new invented record has means S̄_00 = 300, S̄_10 = 285, S̄_01 = 270 and S̄_11 = 256.5 MPa. Calculate D and I_rel. Is zero interaction independent of response scale?

   Show the answer and reasoning

   D = 256.5 − 270 − 285 + 300 = +1.5 MPa. I_rel = 256.5 × 300/(285 × 270) = 1. Zero interaction depends on the reference scale.

   The fractional retentions multiply exactly, while the additive MPa model predicts 255 MPa. This is a deterministic model comparison, not evidence that +1.5 MPa is statistically resolved in a new experiment.

4. Set α = 0 while retaining p₀ = 0.100, β = 2.00 h⁻¹ and one active hour of R in each two-hour schedule. What final q does every schedule predict?

   Show the answer and reasoning

   Every schedule gives q_f = exp(−2.00 × 0.100 × 1.00) = exp(−0.200) ≈ 0.81873.

   No opening occurs, so access stays 0.100 whether R occurs before, after or during O. The apparent sequence effect vanishes when the state-changing operation has no effect.

5. A UV-source setting is increased, but an oxygen witness also shows a change in effective arrival. A specimen then erodes faster. What single-factor conclusion is unsupported, and what evidence would resolve the immediate ambiguity?

   Show the answer and reasoning

   The increase cannot be assigned to UV alone. Measure the actual radiation and relevant oxygen conditions, and compare conditions that separate the intended changes while checking specimen temperature and the witness interpretation.

   A control setting is not an isolated physical input. A damage-equivalent witness can compare exposures without being an absolute atom counter, and its response can depend on conditions. Resolving source cross-talk still does not uniquely establish the microscopic route.

6. Only one of three coupons survives combined exposure; its residual strength is 280 MPa. All three single-exposure coupons survive and average 270 MPa. Does the larger surviving-coupon mean show greater combined resistance?

   Show the answer and reasoning

   No. Two early failures are part of the combined response, so comparing survivor-only strength means hides a different survival outcome.

   Report all specimen outcomes and their failure times or censoring status, as well as the surviving residual strength. A scheduled strength endpoint and a survival endpoint answer different questions. The lesson’s main invented table avoids this selection issue by explicitly having all coupons intact.

7. A displacement-controlled specimen loses 20% of its load-bearing stiffness. May the worked fixed-force factor 1.25 be used to infer its stress increase?

   Show the answer and reasoning

   No. The force can change under displacement control, and stiffness loss is not automatically the same as a uniform effective-area loss.

   Under a simple elastic bar model with fixed displacement, fixed length and modulus, uniform area loss lowers the force in proportion to area and leaves the mean elastic stress unchanged. A real cracked composite requires its own load-sharing model.

What to carry forward

• A combined mechanism links an evolving state to a later response; controls and spatial or chemical evidence test that link.
• Interaction is defined relative to a response scale and reference model. A difference of mean contrasts is not a unique microscopic explanation.
• An exposure setting, the specimen’s actual input and an inferred witness quantity are distinct evidence.
• Equal active exposure times and identical final access can leave different inventories when the sequence changes.
• Residual response, early survival and component resistance are different claims. Bound each by the tested state, history, endpoint and uncertainty.