A multi-modal based laser weld seam defect detection system

Abstract

The application discloses a kind of laser weld defect detection systems based on multi-modal, comprising: sensor module, for collecting multi-modal data in welding process and pre-processing;Feature extraction module, for inputting improved MoSE model, performing preliminary feature extraction and mapping;Feature fusion module, for inputting gated attention neural network, performing weight calculation and interactive attention alignment operation;Defect feature modeling module, for inputting improved capsule network, correlation is calculated based on routing attention mechanism and weight is assigned;Classification and positioning module, for performing activation value calculation, classification determination and saliency distribution mapping;Result output module, for combining the classification result and positioning result of weld defect to generate real-time detection result and feedback to welding control end.The application realizes intelligent classification and positioning of weld defect by improving MoSE model and improved capsule network, with the advantages of sufficient feature fusion, high detection precision.

Images