Prediction Of Hyperbaric GMA Welding Appearance Based On Neural Networks

The development of offshore oil resources has become one of the new worldwide research hotspots. More and more oil platforms, submarine pipelines, underwater structures appear in the ocean. Due to their long-term corrosion in seawater, it is very difficult to repair underwater structure. So it is critical to apply the research to deep-sea underwater welding technology. High pressure dry gas metal arc welding (GMAW) is a realistic method of underwater welding. As underwater welding technology, it has been one of underwater welding technology for storage and transportation of marine oil and gas exploitation.Weld appearance is an important indicator of the welding quality, analysing high-pressure weld forminghigh helps to study the law of high-pressure GMAW. The prediction of weld appearance can guide high-pressure welding technology and save the cost.of hyperbaric welding experiments. For this reason, this paper established a welding appearance prediction model based on BP neural network. The model can realize welding seam information prediction, thus guide the underwater welding technology and improve the quality of welding.Using the orthogonal design method to design the experiment scheme, in which the welding process parameters were taken as influence factors, weld size were taken as output variables. Each factor had four levels. High pressure welding experiments were taken in the hyperbaric welding system for acquiring weld sample datas, which were taken as training samples and forecast samples of the neural network.Tanking Analysis of welding datas, the influent law of GMAW welding appearance in high-pressure environment can be summarized as follows. For example, with increasing the pressure, the weld penetration increases and the weld wide decreases.The regression equation of the weld size and welding technology parameters was established by using regression analysis method, and empirical formula was obtained, too. By testing the regression equation with the R2 test, F test and confidence interval estimation, it showed that the regression equation of welding size was highly significant, and it can be used to predict welding appearance.The predictive model of high-pressure GMAW welding appearance was established and the model was based on the BP neural network. In the model, environmental pressure, welding current, protective gas composition, length of wire extension were considered as variables, and weld wide, weld penetration and reinforcement were taken as output. In VC++ environment, the visual man-machine interface of prediction system based on neural network was developed. Applying prediction system to forecast welding appearance, the results showed that the established forecast system was in high reliability.Finally, comparing the predictive ability of regression equation and BP neural network model by using experimental datas, the result showed that BP neural network model was better than regression equation. So the model based on neutral network is more suitable for high-pressure GMAW welding appearance prediction.