Project 02 · AI · Furniture · Parametric Design

AI that recognizes furniture panels and calculates resizing rules

Model analysis → AI-based feature recognition → Formulating parameter relationships → Generating and verifying manufacturing data

TensorFlowHierarchical CADNet basedGraph inputLinear regressionBOM/Drawing

When the dimensions of the furniture are changed, the panels adjust automatically according to predefined rules.

An animation showing how, when the width of a single panel is adjusted, the adjacent panels automatically update their size and position based on their defined joint relationships. — Research render A piece of furniture where the width was adjusted during the study; the adjoining panels shift in sequence, following their joint relationships.

15 typesPanel types recognized by AI
13Design parameters defining individual pieces of furniture
1 caseJoint invention patent application

1
Analysis of panels and joining surfaces
For a 3D furniture model, we calculate the dimensions and volume of each panel, identify the contact surfaces where panels join, and represent these as a graph.
Conceptual diagram: Explains panel types and their contact relationships.
2
AI panel recognition
Recognizes 15 types of panels—including side panels, bottom panels, top panels, back panels, channels, baseboards (kicks), shelves, and doors—using graph inputs.
3
Formulating relationships between parameters
Define the rules with your team regarding adjustments to width, depth, and height—including specifications for toe kicks, channels, filler panels for corner cabinets, and door clearances—and then use linear regression to formulate the changes in panel position and size.
This screen allows you to review the generated machining code in an NC viewer. Examine the machined surfaces and hole positions on the panel. Translated reference · View Korean original
4
Generation and Verification of Manufacturing Data
Place connecting pieces and hardware, then generate the BOM, drawings, and edge banding data. Use the NC viewer to verify the machining faces and hole locations in the generated machining code.

Assigned role

I was directly responsible for AI research within the team project.

Analysis of panels and joints in a furniture model
AI capable of identifying 15 types of panels
Formulating size-change rules via linear regression
Moving panels within the browser using only mathematical formulas

The furniture rules were defined in collaboration with the team, and the generation of manufacturing data was also developed as a team effort.

Key Judgment

By using linear regression to express the changes in panel position and size through coefficient equations, we have enabled customers to modify design parameters in real-time via the web and place orders.

Panel 15 type recognized by AI

Furniture design parameters based on panel recognition

Furniture design parameters are configured based on the panel types and joint relationships identified by AI. When the customer changes the value on the web, the position and size of the panel are updated numerically, leading to the ordering of the desired specifications.

Base dimensions

Width
Depth
Height

Plinth (toe kick/base)

Plinth height
Plinth inset/offset (front-to-back position)

Channel

Channel depth
Channel height

Corner cabinet finish

Corner cabinet filler width · Front finish material

Interior Configuration

Number of shelves
Use of vertical bands
Vertical band width

Door clearance

Top clearance for doors
Bottom clearance for doors

The adjustable parameters vary depending on the type of furniture and the specific panels included.

01Actions taken at each step

1
Analysis of panels and joining surfaces
- Calculate panel dimensions, volume, and thickness
- Extracting face area, centroid, and adjacency relationships
- Hierarchical graph connecting panels and their joining surfaces
2
AI panel recognition
- Recognize 15 types of Hierarchical CADNet-based models and MLP
- A tool for exporting geometry, labels, and graphs from labeled data.
- Cross-validation splitting by geometry family units
- Experiment on removing feature sets and checking for axis swaps
3
Formulating relationships between parameters
- Generating samples by perturbing reference values based on specific rules
- Position and size of each panel determined via linear regression
- Real-time dimension adjustment via mathematical formulas and integration with the ordering process, all accessible via the web.
The furniture rules were defined in collaboration with the team.
4
Generation and Verification of Manufacturing Data
- Positioning of connecting pieces and hardware placement
- BOM, drawing, edge banding calculation
- Verifying machined surfaces and hole positions in the generated machining code

02Problem and solution

ProblemEven if panels are similar in size and shape, each plays a different role within a piece of furniture.
ResolveThe geometry and joint relationships of the panels are analyzed using graphs, and AI is employed to categorize them into 15 types, such as side panels, bottom panels, and top panels.
ProblemCustomers want to adjust the dimensions and configuration of furniture in real-time via the web and place orders directly according to their preferred specifications.
ResolveWe used machine learning (linear regression) to derive a formula for calculating the position and size of panels. This allows for lightweight, real-time parameter adjustments directly within the browser, facilitating the ordering of furniture.
ProblemRegression equations may inadvertently alter values that should remain unchanged.
ResolveFor outputs that do not change, the coefficient was fixed at 0.

03Verification

Verified items

Cross-validation by geometry family
Experiment comparing results by excluding specific feature sets
Stress evaluation of input with transformed coordinate axes
Panel-based metrics
Regression testing of panel positions after modifying parameters (e.g., for a lower corner cabinet)

AI that recognizes furniture panels and calculates resizing rules

When the dimensions of the furniture are changed, the panels adjust automatically according to predefined rules.

Analysis of panels and joining surfaces

AI panel recognition

Formulating relationships between parameters

Generation and Verification of Manufacturing Data