In the welding process

electrical signals are closely related to the welding quality. Ultra-narrow gap welding is a new welding method with high efficient and low heat input.However

its arc control and metal transfer process are complex

the traditional signal feature extraction and analysis method often can not fully express and make full use of the time series information. Combining with the characteristics of ultra-narrow gap welding process

the complete time series of electrical signals are used to construct the convolutional neural network

while deeply excavating the timing information of the same attribute signal

the time correlation information between welding current and arc voltage signals acquired synchronously is fully consi-dered to predict welding quality. The test result shows that the proposed deep network model based on the complete time series signal can predict the welding quality accurately

and the accuracy rate reaches 95%

in the case of adopting RTX2080 GPU as the computing accelerator

the time required for the model prediction is only 0.178 ms

which lays a foundation for the real-time prediction of ultra-narrow gap welding quality.