It's environmental stress cracking: a cleaner, oil, plasticizer or adhesive that is harmless on an unloaded part seeps into a spot held in tension and lets the chains slide apart below yield. Amorphous PC, ABS and acrylic are most at risk. Fix it with less sustained strain, a different chemical, or a resin proven in a strained test.
Environmental stress cracking (ESC) is a brittle crack that grows in a plastic part when two things act at once: a tensile stress, usually far below the yield stress on the datasheet, and contact with a liquid, paste, vapor or soft material that the plastic would ignore if it were unstressed. The Madison Group's chapter in the ASM Handbook estimates that about 25% of plastic part failures trace back to ESC, and its list of usual victims reads like a product catalog: device housings, enclosures fastened with metal screws, lenses, pipes and handheld electronics.
The mechanism looks more like creep than corrosion. Nothing reacts and no polymer chains get cut. A mild agent works its way into the surface wherever tension has opened a little extra space between chains, weakens the attraction between neighboring chains, and lets them slide apart. Microscopic voids called crazes form, join up, and become a crack that steps inward over time. Because the agent acts only locally, an unloaded coupon soaked in it can show no weight gain and no strength loss, which is how the same cleaner can look harmless on a chart and still split a housing. Scratches and surface flaws make it worse: the ASM chapter puts the stress concentration at such defects anywhere from 1 to 50 times.
That's also why ESC shows up after shipment rather than at the press. The part molds clean, passes inspection and assembly, then meets hand lotion, a disinfectant wipe or oily fingers in use. The Madison Group authors call these secondary agents: chemicals nobody planned for, so nobody tested them. Failures from primary agents, the fluids a part is designed to hold, are uncommon because those pairings normally get tested.
The biggest single factor is how the resin's chains are packed. Amorphous polymers are a random tangle with free volume a chemical can slip into. Semicrystalline polymers have tightly ordered crystal regions that block it. The ASM chapter calls amorphous resins far more susceptible and singles out polycarbonate as failing from ESC out of proportion to how widely it's used. DuPont's general design guide describes its acetal, nylon and PET families as particularly resistant to the effect.
| Resin | Structure | ESC tendency | Trouble agents from the cited cases, or notes |
|---|---|---|---|
| Polycarbonate (PC) | Amorphous | High; the most reported | Isopropyl alcohol, sunscreen lotion, cutting oil on screws, uncured adhesive (in a PC/PBT blend) |
| ABS | Amorphous | High | Phthalate plasticizer migrating out of a rubber O-ring |
| Acrylic (PMMA) | Amorphous | High | Methanol in windshield washer fluid on lenses |
| Polystyrene, SAN | Amorphous | High, like the rest of the family | Treat oils and solvents as suspect until tested |
| Polyethylene, polypropylene | Semicrystalline | Low for most agents, with known exceptions | Detergents and surfactants on polyethylene; silicone oil on low-density polyethylene |
| Nylon (PA6, PA66), acetal (POM) | Semicrystalline | Low | Resistance drops if a cold mold leaves the skin under-crystallized |
Semicrystalline isn't a free pass. Crystallinity only protects the part if the part actually crystallizes: the ASM chapter notes that a cold mold can freeze the skin partly amorphous, and the skin is exactly what the chemical touches. The same chapter lists other material levers that help: higher molecular weight (lower melt flow) grades, a narrow molecular weight distribution, glass-fiber reinforcement and rubber impact modifiers. Plain mineral fillers do little. For the everyday trade-offs among the three most common resins, see ABS vs polycarbonate vs polypropylene.

Organic esters are the family behind most ESC failures, according to the ASM chapter, and they turn up in synthetic lubricants, plasticizers, adhesives, cosmetics, lotions, soaps, fragrances, medications and vegetable oils. Its broader list adds solvents, fuels, degreasers, aerosols, paints and thinners, insect repellent, disinfectants and ice melt. Four patterns come up again and again:
Heat usually speeds ESC up, the same way it speeds creep. The exception is a volatile agent that evaporates off a warm part before it can do much.
Stress can be molded in, assembled in, or applied in use, and the sources stack. The ASM authors report that molded-in residual stress on its own is often enough to crack a part that carries no external load.
Direction matters too. Load across the flow direction pulls oriented chains apart; load along it is far better tolerated. For PMMA, the ASM chapter cites up to four times the ESC resistance when stress runs parallel to the orientation.

Take a hypothetical polycarbonate housing with a steel dowel pressed into a molded bore. The drawing calls the pin 4.06 ± 0.01 mm and the bore 4.00 +0.04/−0.00 mm. Steel barely compresses, so the pin forces the bore open to its own size. The bore's circumference grows by π times the diametral interference, so the hoop strain at the bore surface is simply interference divided by bore diameter.
Now set those against the measured critical strains. Against isopropyl alcohol at 1.21%, the nominal part survives with a thin margin and every part from the tight end of the band is over the line. That is the classic field pattern: most units are fine, a scattering crack at the pin after the first wipe-down, and a good sample on the bench never reproduces it. Against the glycol ether at 0.29%, essentially the whole band is over; only the very loosest fits, below about 0.012 mm of interference, stay under it. Molded-in stress at the bore adds on top of these numbers, so the real margin is thinner than the arithmetic shows.
The telling number is the width of the band. The two tolerances together span 0.06 mm, which is 1.5% strain on a 4 mm bore, wider than the whole window below the isopropyl alcohol limit. Tightening tolerances won't rescue that joint. Change the joint instead: crush ribs so a small feature takes the strain, a slip fit with a separate retention feature, a heat-staked or ultrasonically inserted pin, a semicrystalline resin, or a cleaner the part tolerates. The press-fit interference calculator handles holding force once the strain question is settled. One caution: published critical strains come from one grade, one temperature and one test method, so treat them as a ranking, then test your own grade.
A chemical-resistance chart is where you start, not where you finish. The ASM chapter warns that many charts come from stress-free immersion, which catches swelling and degradation but misses mild ESC agents entirely. Real ESC tests hold the plastic under strain or load while the chemical is on it:
The bent-strip arithmetic is plain geometry: the outer surface strain is about the strip thickness divided by twice the jig radius. For 3 mm strips, a 300 mm radius gives 0.5%, 150 mm gives 1.0% and 100 mm gives 1.5%. Run a ladder of radii, apply each chemical the way the product will meet it, and record the lowest strain where crazes appear. Use strips molded from the exact grade and color, ideally cut from parts off the production tool.
To confirm that a returned part failed by ESC, look at the fracture. The ASM chapter describes multiple crack origins that merge into one fracture, a typical sign of chemical attack at many surface flaws, and recommends FTIR or GC-MS on the fracture surface to identify the agent.
"Must be chemical resistant" gives a molder nothing to test against. Spell it out:
ESC needs tension. A feature that is purely compressed won't stress-crack, but the ASM chapter points out that bending, shear and even compression usually create a tensile zone somewhere nearby, so check before relying on it. A volatile agent on a hot part may evaporate before it does damage. Primary agents that were tested for the application rarely cause trouble. And the ASM authors note that a region strained past yield can relax and reorient enough to resist ESC better, which is an accident to recognize in a failure analysis, not a design method.
On an unstressed part, usually yes. On a part held in tension, it depends on the strain. One published study measured polycarbonate's critical strain in isopropyl alcohol at 1.21%, so a press fit, preloaded latch or high-stress gate area near that level can craze after repeated wipes. Test the actual grade at your design strain before approving an alcohol wipe in the manual.
Often not. Many charts come from soaking unstressed coupons and measuring weight, size or strength change. That catches swelling and chemical attack but misses mild stress-cracking agents, which only act on strained material. Use the chart to rule out bad pairings, then confirm the shortlist with a strained-specimen test such as ASTM D543 Practice B or ISO 22088-3.
Because strain and exposure vary part to part. Parts at the tight end of a press-fit tolerance carry more hoop strain, molded-in stress varies with cavity and process drift, and only some units meet the cleaner, lotion or oily fastener. A crack pattern clustered at one feature, appearing after use rather than at assembly, points strongly to ESC.
It might help, but prove it. The ASM chapter credits glass-fiber reinforcement and rubber impact modification with better ESC resistance, while plain mineral fillers do little. Blends still contain amorphous resin, so they can still crack with the right agent and strain. Rank the candidates side by side in the same bent-strip test, with the same chemicals and strain levels.
ESC cracks usually start at several points on a stressed, chemically exposed surface and merge, look brittle in a resin that is normally tough, and appear after a period in use. An impact crack starts at the point of the hit. A failure lab can confirm ESC from the fracture features and identify the agent on the surface with FTIR or GC-MS.
Send the part file and the quantity you expect to run, and we'll review it before you commit to tooling. NDA first if you want one.
Send a part for reviewBook a 30-minute call