The Architect

Most slicers flatten geometry. We constrain a field and keep it printable.

Non-planar slicing is not a style preset. It is a constrained optimization pipeline: volumetric mesh, field solve, isosurfaces, toolpaths, G-code. AxisBlend is FDM slicing software for 5-axis, non-planar, and non planar toolpaths.

Brutal Honesty

We publish the model.
Not adjectives.

L_{c} = {x \in Ω | h (x) = c}

Layers are extracted as level sets of a scalar field.

\min_{h} (w_{smooth} \sum_{(i, j) \in ℱ}^{} V_{i j} {∥ \nabla h_{i} - \nabla h_{j} ∥}^{2} + w_{flat} \sum_{k = 1}^{n} {(h_{k} - z_{k})}^{2})

Objective: smooth gradients while staying close to target height.

h_{min} \leq h_{i} \leq h_{max}, \frac{\partial h}{\partial z} \geq 0.001

| \frac{\partial h}{\partial x} |, | \frac{\partial h}{\partial y} | \leq \tan (θ_{max}) \frac{\partial h}{\partial z}

Feasible set: bounded field values, monotonic ascent, and linearized slope limits.

t_{physical} \approx \frac{Δh}{∥ \nabla h ∥}

Caveat: h_min/h_max bound field values, not direct physical layer thickness.

| \frac{\partial h}{\partial x} |, | \frac{\partial h}{\partial y} | \leq \tan (θ_{max}) \frac{\partial h}{\partial z}, L∞ linearization

Caveat: slope cone is linearized (square-pyramid approximation), so diagonal directions are permissive.

Pipeline Exposed

5 stages. No black box.

1. TetMesh

AABB subdivision and Kuhn decomposition build the volumetric domain.

2. Field Optimization

A constrained quadratic objective solves for h over tetmesh vertices.

3. Isosurface Extraction

For each isovalue, crossings are interpolated and chained into contours.

4. Toolpath

Contours are converted into ordered travel and extrude motion sequences.

5. G-code

Marlin-compatible output emits XYZ moves with varying Z along paths.

Compute Reality

Hardware-aware solve path.
Same constraints.

Same optimization model can run as interior-point CPU or ADMM on CPU/GPU.
Runtime chooses path by problem size and hardware capability.
Small jobs usually stay CPU where launch overhead is lower.
If GPU is unavailable or unstable, execution falls back to CPU.
This browser build currently executes the CPU optimization path.

No synthetic benchmark claims here. We only publish speed numbers tied to a model, config, and hardware profile.

Self-Aware Limits

Current scope.
Explicitly.

What we do not pretend to solve yet:

Single G-code dialect today: Marlin.
No collision detection for non-planar moves yet.
Quality depends on tetmesh resolution and conditioning.

Built above the Arctic Circle.