Robot Welding via Semantic Feature Clustering for Flexible Programming

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing welding robot programming methods, such as teach-in and offline programming, struggle with low precision, efficiency, and flexibility when dealing with small batch, complex-structured components, leading to increased workload and deviations.

Innovation Solution

A robot welding method and system utilizing semantic feature clustering to extract deep features from workpiece images, employing parameterized online programming to generate welding programs, and integrating an edge computing server for training and data recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If teach-in programming is used to guide the robot to required positions manually, then the robot can perform repetitive welding tasks with high consistency, but the programming accuracy and efficiency are poor

Engineering Contradiction:

Improvewelding consistencyVSAvoidpositioning accuracy

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The patent replaces manual mechanical teaching operations with automated image recognition and semantic feature clustering systems. The system uses visual sensors to capture workpiece images, processes them through neural networks to identify semantic features, and automatically generates robot motion paths, eliminating manual positioning errors while maintaining welding consistency.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent creates digital copies of workpieces through image acquisition and semantic feature extraction. Instead of manually teaching each position, the system generates virtual models and feature representations that the robot can reference, improving positioning accuracy through precise digital reproduction rather than manual measurement.

Inventive Principle:

Principle #26Copying

2Ease of manufacture

If offline programming is used to program on computer with workpiece information, then the program can be transferred to robot controller, but the program cannot adapt when workpiece position changes and requires rewriting

Engineering Contradiction:

Improveprogramming convenienceVSAvoidprogram flexibility

Core Design Contradiction:

Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent transforms static offline programming into a dynamic system that continuously adapts to workpiece variations. The semantic feature clustering model processes real-time workpiece images, identifies positional and structural variations, and dynamically adjusts welding parameters and motion paths without requiring program rewriting, enabling the system to adapt to position changes automatically.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent employs parameterized programming where welding programs use variable parameters instead of fixed values. When workpiece position or structure changes, the system modifies relevant parameters (position coordinates, welding speed, torch angle) while maintaining the core program logic, allowing flexible adaptation without complete program rewriting.

Inventive Principle:

Principle #35Parameter changes

3Productivity

If offline programming is used with digital model coordinates, then the program can be executed repeatedly, but deviations between digital model and real object cause execution failures

Engineering Contradiction:

Improveprogram reusabilityVSAvoidmodel-real object alignment

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent implements feedback loops where the image recognition system continuously compares actual workpiece features with the digital model, detects deviations, and corrects positioning errors in real-time. This feedback mechanism ensures that even when digital model and real object differ slightly, the robot can accurately locate and weld the actual workpiece features.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent performs preliminary semantic feature extraction and workpiece identification before welding execution. The system pre-processes workpiece images to extract key semantic features and generates preliminary positioning data, which serves as a foundation for accurate welding even when digital model deviations exist, ensuring program reusability while maintaining precision.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12528195B2Robot welding method and system based on semantic feature clustering

Publication Date: 2026.01.20 QINGDAO UNIV OF TECH

AI summary

A robot welding method based on semantic feature clustering; the method comprises: acquiring an image of a workpiece, and generating an image training sample set; performing semantic annotation on the image of the workpiece to generate a semantic training sample set; establishing a semantic clustering model, and training same by using the image training sample set and the semantic training sample set; by using the trained semantic clustering model, identifying an image of a workpiece to be welded, so as to obtain data of the workpiece to be welded; on the basis of the data of the workpiece to be welded and by using parameterized online programming, generating a welding program for the workpiece to be welded; and on the basis of the welding program, executing welding of the workpiece to be welded. By means of the robot welding method based on semantic feature clustering, neither teaching programming nor offline programming is required, thereby improving the precision, efficiency, universality and flexibility. Further provided is a robot welding system based on semantic feature clustering.