Robot welding method and system based on semantic feature clustering

A robot welding system based on semantic feature clustering, comprising:

• a welding robot configured to perform machining on workpieces to be welded;
• an image acquisition module configured to capture images of the workpieces to be welded;
• an edge computing server configured to train a semantic clustering model and utilize the trained semantic clustering model to recognize images of the workpieces to be welded, obtaining data about the workpieces to be welded;
• wherein training the semantic clustering model includes semantically annotating workpiece images to generate a semantic training sample set,
• establishing the semantic clustering model and using the image training sample set and the semantic training sample set for training;
• a programmable logic control module configured to generate welding programs for the workpieces to be welded based on the data about the workpieces to be welded using parameterized online programming;
• and
• a robot control module configured to retrieve the welding programs generated by the programmable logic control module and control the welding robot to perform welding based on the welding programs;
• wherein establishing the semantic clustering model comprises:
• using a Resnet50 model with the image training sample set as input to extract shallow features of the workpiece images;
• using a Resnet50 model with both the image training sample set and the semantic training sample set as input to extract semantic features of the workpieces;
• and
• fusing the shallow features and semantic features;
• wherein the fusion of shallow features and semantic features comprises:
• assigning weights to the semantic features based on
• α
• =
• m
• ×
• A
• m
• ×
• A
• +
• m
• ×
• B
• ;
• assigning weights to the shallow features based on β=1−α;
• and
• performing weighted overlay fusion to obtain the fused feature T=αV+βU;
• wherein, m represents number of samples, A represents the clustering results of shallow features, B represents clustering results of semantic features, V represents semantic feature expressions, and U represents shallow feature expressions;
• wherein the programmable logic control module is configured
• to analyze process characteristics and structural features of the workpieces to be welded to generate welding program modules,
• to extract common features of the workpieces to be welded, replace welding data associated with common features with variables to generate common feature programs, and
• to use data about the workpieces to be welded to assign values to variables corresponding to welding data for common features, thereby generating welding programs.