# US 2026/0087343 A1 · Machine-learning for assembling mechanical parts

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0087343 A1
- Document ID: US20260087343A1
- Status: Published application
- Kind: A1
- Application number: 19/334,657
- Filed: 2025-09-19
- Published: 2026-03-26
- Earliest claimed: 2024-09-20
- Assignees: Dassault Systemes
- Applicants: Dassault Systemes
- Inventors: Tong ZHAO; Asma REJEB SFAR; Sâad AMMARI
- Web page: https://patentgrep.com/patent/US20260087343A1

## CPC

- G06F 30/17: Mechanical parametric or variational design
- G06N 3/08: Learning methods
- G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G06N 3/045: Combinations of networks
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A computer-implemented method of machine-learning for assembling mechanical parts, based on a mating score and a mating axis. The method includes providing a dataset of pairs of B-Reps, each pair comprising at least one B-Rep representing an assembly of mechanical parts, being labelled with mating compatibility data and, when the B-Reps of the pair are compatible according to the mating compatibility data, mating axis compatibility data. The method also comprises training a neural network based on the dataset, configured for taking as input a pair of B-Reps. The neural network also outputs a mating score of a pair of single embeddings, each single embedding representing a B-Rep, the mating score representing a score of mating compatibility between the mechanical parts represented by the pair, and if the B-Reps are compatible according to the mating score, data defining a mating axis.

## Claims

### Claim 1

A computer-implemented method of machine-learning for assembling mechanical parts, based on a mating score and a mating axis, the method comprising:
obtaining a dataset of pairs of Boundary Representations (B-Reps) representing mechanical parts, each pair including at least one B-Rep representing an assembly of mechanical parts, each pair being labelled with mating compatibility data and, when the B-Reps of the pair are compatible according to the mating compatibility data, mating axis compatibility data, the mating compatibility data representing an extent of mating compatibility between the mechanical parts represented by the pair, the mating axis compatibility data representing an extent of compatibility between the mechanical parts represented by the pair along a mating axis defined by a pair of B-Rep entities of the B-Reps of the pair; and
training a neural network based on the dataset, the neural network being configured to
take as input a pair of B-Reps each representing a mechanical part or an assembly of mechanical parts,
output a mating score of a pair of single embeddings, each single embedding representing a B-Rep of the pair, the mating score representing a score of mating compatibility between the mechanical parts represented by the pair, and
when the B-Reps of the pair are compatible according to the mating score, output data defining a mating axis.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein the neural network includes:
an embedding network configured to be applied to a respective pair of B-Reps and to output a single embedding for each B-Rep of the pair,
a mating compatibility score network configured to take as input a concatenation of the pair of single embeddings and to output the mating score, and
an axis network configured to be applied to a respective pair of B-Reps and to output data defining a mating axis of the pair.

### Claim 3 (depends on 2)

The computer-implemented method of claim 2, wherein the embedding network further includes a Siamese encoder configured to take as input the pair of B-Reps, the output of the Siamese encoder being passed as input to the mating compatibility score network and the axis network.

### Claim 4 (depends on 3)

The computer-implemented method of claim 3, wherein the neural network includes a pooling module applied on the output of the Siamese encoder.

### Claim 5 (depends on 2)

The computer-implemented method of claim 2, wherein the embedding network, the mating compatibility score network, and the axis network are trained simultaneously using a loss, the loss penalizing, for each pair of B-Reps of the dataset:
a disparity between the mating compatibility data of the pair of B-Reps and the mating score outputted by the neural network for the pair of B-Reps, and/or
a disparity between the mating axis compatibility data of the pair of B-Reps and data defining the mating axis outputted by the neural network for the pair.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, further comprising computing the embeddings of the B-Reps of the dataset by applying the neural network.

### Claim 7

A computer-implemented method of applying a neural network learnable according to a computer-implemented machine-learning, the method comprising:
machine-learning the neural network by:
obtaining a dataset of pairs of Boundary Representations (B-Reps) representing mechanical parts, each pair including at least one B-Rep representing an assembly of mechanical parts, each pair being labelled with mating compatibility data and, when the B-Reps of the pair are compatible according to the mating compatibility data, mating axis compatibility data, the mating compatibility data representing an extent of mating compatibility between the mechanical parts represented by the pair, the mating axis compatibility data representing an extent of compatibility between the mechanical parts represented by the pair along a mating axis defined by a pair of B-Rep entities of the B-Reps of the pair, and
training a neural network based on the dataset, the neural network being configured to
take as input a pair of B-Reps each representing a mechanical part or an assembly of mechanical parts,
output a mating score of a pair of single embeddings, each single embedding representing a B-Rep of the pair, the mating score representing a score of mating compatibility between the mechanical parts represented by the pair, and
when the B-Reps of the pair are compatible according to the mating score, output data defining a mating axis;
obtaining a set of B-Reps, each B-Rep being associated with a single embedding obtained by applying the neural network to the B-Rep; and
determining an assembly of parts based on the set, by applying iteratively the neural network to a pair comprising a B-Rep of the set and an assembly resulting from a previous iteration.

### Claim 8 (depends on 7)

The computer-implemented method of claim 7, wherein the determining is further based on data defining a mating axis outputted by the neural network.

### Claim 9 (depends on 7)

The computer-implemented method of claim 7, further comprising applying an optimization to the assembly, the optimization fixing degrees of freedom between the parts of the assembly.

### Claim 10 (depends on 7)

The computer-implemented method of claim 7, wherein iterations stop upon reaching a predetermined number of parts and all possible compatible parts of the set have been explored.

### Claim 11 (depends on 7)

The computer-implemented method of claim 7, wherein iterations start by a user selecting a B-Rep of the dataset or by automatic selection of a B-Rep of the dataset based on a predetermined criterion.

### Claim 12 (depends on 7)

The computer-implemented method of claim 7, further comprising computing the embeddings of the B-Reps of the set by applying the neural network.

### Claim 13

A device comprising:
a processor; and
a non-transitory computer readable data storage medium having recorded thereon:
a first computer program having instructions for machine-learning in assembling mechanical parts, based on a mating score and a mating axis, that when executed by the processor cause the processor to be configured to:
obtain a dataset of pairs of Boundary Representations (B-Reps) representing mechanical parts, each pair including at least one B-Rep representing an assembly of mechanical parts, each pair being labelled with mating compatibility data and, when the B-Reps of the pair are compatible according to the mating compatibility data, mating axis compatibility data, the mating compatibility data representing an extent of mating compatibility between the mechanical parts represented by the pair, the mating axis compatibility data representing an extent of compatibility between the mechanical parts represented by the pair along a mating axis defined by a pair of B-Rep entities of the B-Reps of the pair, and
train a neural network based on the dataset, the neural network being configured to
take as input a pair of B-Reps each representing a mechanical part or an assembly of mechanical parts,
output a mating score of a pair of single embeddings, each single embedding representing a B-Rep of the pair, the mating score representing a score of mating compatibility between the mechanical parts represented by the pair, and
when the B-Reps of the pair are compatible according to the mating score, output data defining a mating axis; and/or
a second computer program having instructions for applying a neural network learnable according to the machine-learning that when executed by the processor causes the processor to be configured to:
obtain a set of B-Reps, each B-Rep being associated with a single embedding obtained by applying the neural network to the B-Rep, and
determine an assembly of parts based on the set, by applying iteratively the neural network to a pair comprising a B-Rep of the set and an assembly resulting from a previous iteration; and/or
a neural network learnable according to the machine-learning.

### Claim 14 (depends on 13)

The device of claim 13, wherein the neural network includes:
an embedding network configured to be applied to a respective pair of B-Reps and to output a single embedding for each B-Rep of the pair,
a mating compatibility score network configured to take as input a concatenation of the pair of single embeddings and to output the mating score, and
an axis network configured to be applied to a respective pair of B-Reps and to output data defining a mating axis of the pair

### Claim 15 (depends on 14)

The device of claim 14, wherein the embedding network includes a Siamese encoder configured to take as input the pair of B-Reps, the output of the Siamese encoder being passed as input to the mating compatibility score network and the axis network.

### Claim 16 (depends on 15)

The device of claim 15, wherein the neural network includes a pooling module applied on the output of the Siamese encoder.

### Claim 17 (depends on 1)

A non-transitory computer readable medium having stored thereon a computer program for machine-learning for assembling mechanical parts, based on the mating score and the mating axis that when executed by a computer causes the computer to implement the method according to claim 1.

### Claim 18 (depends on 7)

A non-transitory computer readable medium having stored thereon a computer program for applying the neural network learnable according to the computer-implemented machine-learning that when executed by a computer causes the computer to implement the method according to claim 7.

### Claim 19 (depends on 7)

The computer-implemented method of claim 7, wherein iterations stop upon reaching a predetermined number of parts of the set have been explored.

### Claim 20 (depends on 7)

The computer-implemented method of claim 7, wherein iterations stop upon reaching all possible compatible parts of the set have been explored.
