Single fiber in a box under tension
A unit cube of matrix is stretched in x. One fiber sits entirely inside the box — it does not reach the faces and is not pulled at a tip. Load enters through the matrix; the cohesive interface then loads the fiber. Right-face ux is taken to 0.5, the MatFib ramp that drives slip past the tail along almost the whole fiber. The force plot is the extra left-face reaction relative to matrix-only.
Rx = 0
EfAf uf″ = bw τ(uf − εx) · uf′(tips) = 0 · left face fixed, right ux → 0.5 (full debond)
Not a pull-out test
Pull-out drives the fiber tip while the matrix is held. Here the opposite happens, matching
MatFib’s FinalProgram_full_1F_NonlinearInterface: the matrix cube is loaded,
and the fiber is an interior inclusion. Endpoints are
(0.21, ½, ½) → (0.81, ½, ½). Face x = 0 is fully fixed; face x = 1 has prescribed
ux → 0.5 with uy, uz free.
Discretization (MATLAB)
- Unit cube, structured H8 matrix
- Fiber as 3-D truss bars, 20 segments
- Newton–Cotes interface along the fiber
- Em = 1, ν = 0, Ef = 104, d = 0.01
What you are watching
- Matrix only — cube stiffness with no fiber
- Linear bond — constant kt = 100, Cox shear-lag
- Cohesive — same 1-D fiber, nonlinear τ(s)
- At ux = 0.5 the fiber is unloaded: ΔRx → 0, rings grey
v2: τ = ft s/a0 (toe), then linear softening to wc = 2 Gf/ft
Why the tips go first
Far-field matrix motion is um = ε x with ε = ux/L. The fiber is orders of magnitude stiffer, so its strain is smaller and more uniform. Relative slip s = uf − um therefore peaks at the ends (Cox shear-lag). Traction-free tips force N = 0 there; axial force builds toward the mid-fiber. Once |s| passes the peak of the bond law, softening fronts march inward from both tips — the same progressive debonding the MATLAB NR solve records on the 3-D mesh.
This page solves that 1-D embedded-fiber problem live (bars + nodal Newton–Cotes bond, displacement-controlled NR), which is the reduced model of the 3-D H8 + bar assembly. Matrix compliance around the fiber is omitted, so the cohesive curve is a slightly stiffer cousin of the MATLAB RVE reaction. The MATLAB run at ux = 0.5 reports 34/40 Gauss points in the tail and Rx equal to the matrix-only value; only a vanishing core at mid-fiber still sits below the tail threshold (slip is kinematically zero there). After that, ΔRx is gone and the cube behaves as matrix-only.