Modalities
Epoxies and Elastomers
Epoxies form durable adhesives, coatings and composite matrices. Elastomers provide the flexibility needed in seals, tyres and vibration isolation. Their formulations connect ingredient choices to mechanical performance.
Formulation and Properties
Within a controlled formulation range, useful properties often change consistently with ingredient loading. Reinforcing filler can increase an elastomer’s stiffness and hardness; a compatible plasticiser can soften it. Such monotonic trends, where a response keeps moving in the same direction, can make optimisation much more efficient.
The trend must be established for the particular material and property. Filler dispersion, ingredient interactions and cure conditions can change it. Epoxy resin and hardener proportions govern network formation, so adding more hardener does not simply produce a stiffer material. Strength, toughness and adhesion need their own measurements.
Hardness measures resistance to indentation; elastic modulus describes stiffness under a specified deformation. They are related but not interchangeable. Comparisons need consistent test conditions, especially temperature and strain, as well as controlled mixing and cure history.
AI Agent Potential
An agent can use established trends to choose informative experiments. If increasing one ingredient raises stiffness, a measurement below the target can rule out lower loadings along that formulation path. Testing between a failing and a passing composition progressively narrows the search.
The objective is to satisfy the whole specification. A stiffer elastomer may become too hard; an epoxy modification may compromise processing or toughness. Agents can track these constraints, model interactions between ingredients and reserve experiments for confirmation. Monotonicity is a useful assumption to test, not a substitute for evidence.
Example Task Demo
Find the lowest filler loading that gives a fictional elastomer enough stiffness without making it too hard. Only filler loading changes; the base polymer, other ingredients, mixing, cure and test conditions stay fixed. Here, phr means parts by mass per hundred parts of base rubber.
Assume prior screening supports increasing modulus and hardness over this range. Start with two endpoint measurements, then let the agent test the midpoint of the remaining interval. Compare the two stiffness targets: one is feasible, while the other conflicts with the hardness limit. All values are invented.
Points: measurements · Shading: search interval · Dashed lines: limits
Click again to narrow the search, one test at a time.
The measured modulus is 5.17 MPa and hardness is 58.5 Shore A. 21 phr misses the stiffness target, so 22 phr is the lowest feasible loading on the 1 phr grid, under the assumed trends.
| Filler (phr) | Modulus (MPa) | Hardness (Shore A) | Stiffness target |
|---|---|---|---|
| 0Endpoint | 2.00 | 42.0 | Below |
| 40Endpoint | 9.20 | 72.0 | Met |
| 20Test 1 | 4.80 | 57.0 | Below |
| 30Test 2 | 6.80 | 64.5 | Met |
| 25Test 3 | 5.75 | 60.8 | Met |
| 22Test 4 | 5.17 | 58.5 | Met |
| 21Test 5 | 4.98 | 57.8 | Below |
Static view: the completed 5 MPa search finds 22 phr using seven measurements. Enable JavaScript to step through the search or change the target.
In practice, the candidate needs repeat batches and checks of other required properties. A measurement that contradicts the assumed trend should prompt further investigation before the agent discards more formulations. Searching several ingredients also requires accounting for their interactions.
How the illustration works
The candidate loadings are the 41 integer values from 0 to 40 phr. For loading x, the invented small-strain elastic modulus is E = 2 + 0.10x + 0.002x² MPa, and hardness is H = 42 + 0.75x on the Shore A scale. These are independently assigned responses, not a conversion between hardness and modulus or a fit to experimental data.
Both responses increase throughout this interval. The search receives only the measurements it requests and the assumed directions of change. It narrows the stiffness boundary by testing the integer midpoint, then checks the hardness of the lowest loading that meets the stiffness target. If that loading is too hard, all higher loadings also fail the hardness limit.
The model has no measurement noise or batch variation. Each displayed search uses seven formulations, including the two endpoints, out of 41 possible loadings. The saving is conditional on the assumed trends and a resolution of 1 phr; endpoint measurements alone would not establish monotonicity in a real material.
This is purely an example task and is not part of GDPx.