A method and system for real-time detection of welding defects based on high-frequency time-series data

Abstract

The invention discloses a method and system for real-time detection of welding defects based on high-frequency time-series data. The detection method includes: first sampling the collected high-frequency welding time-series data according to a set window length, and marking each sample Defect occurrence time period and defect type to generate a data sample set; then use the generated data sample set to train the ResNet and TCN fusion network model to obtain a trained detection model; finally obtain new real-time high-frequency welding data, according to the set window length Input it into the trained detection model for prediction, and output the welding defect category in real time. The present invention can apply ResNet to the field of timing detection by combining the network structure of the ResNet network and the TCN, aiming at the high-frequency welding timing data with a large amount of data and a long sequence length, through parallel convolution calculation, in the training process In the process, the training speed is accelerated, and in the prediction process, strong real-time prediction is achieved.

Description

technical field [0001] The invention relates to a real-time detection method and system for welding defects based on high-frequency time series data, belonging to the technical field of automatic welding. Background technique [0002] Intelligent welding is one of the most important research topics in the field of intelligent manufacturing, and sensing technology and its information processing are the key elements to realize the intelligentization and automation of the welding process. In recent years, sensing technologies with the characteristics of miniaturization, non-contact and large transmission capacity have been more applied to the real-time control of welding process and quality, such as arc sensing, visual sensing, sound sensing, spectral sensing Wait. These sensors use different information sources to obtain large-scale information related to welding quality. How to mine effective information and timely feedback and use it in the real-time detection of welding qu...

AI Technical Summary

Problems solved by technology

Based on machine vision, on the one hand, the cost is high, and a professional high-definition camera needs to be installed. On the other hand, there are many uncontrollable factors, such as light intensity, plate impurities, camera shooting angle, etc., and require a lot of training. network of

For the modeling of time series data, the most commonly used models are cyclic neural networks such as LSTM and GRU, but the cyclic neural network cannot be calculated in parallel, and the training speed is slow, especially for high-frequency welding time series data, which has a long sequence length and data It is particularly important to be able to perform parallel computing due to the large amount of

Images