The arc welding process is widely used in industry but its automatic control is limited by the difficulty in measuring the weld bead geometry and closing the control loop on the arc, which has adverse environmental conditions. To address this problem, this work proposes a system to capture the welding variables and send stimuli to the Gas Metal Arc Welding (GMAW) conventional process with a constant voltage power source, which allows weld bead geometry estimation with an open-loop control. Dynamic models of depth and width estimators of the weld bead are implemented based on the fusion of thermographic data, welding current and welding voltage in a multilayer perceptron neural network. The estimators were trained and validated off-line with data from a novel algorithm developed to extract the features of the infrared image, a laser profilometer was implemented to measure the bead dimensions and an image processing algorithm that measures depth by making a longitudinal cut in the weld bead. These estimators are optimized for embedded devices and real-time processing and were implemented on a Field-Programmable Gate Array (FPGA) device. Experiments to collect data, train and validate the estimators are presented and discussed. The results show that the proposed method is useful in industrial and research environments.

This study aims at evaluating the efficiency of sensor fusion, based on neural networks, to estimate the microstructural characteristics of both the weld bead and base material in GMAW processes. The weld beads of AWS ER70S-6 wire were deposited on SAE 1020 steel plates varying welding voltage, welding speed, and wire-feed speed. The thermal behavior of the material during the process execution was analyzed using thermographic information gathered by an infrared camera. The microstructure was characterized by optical (confocal) microscopy, scanning electron microscopy, and X-ray Diffraction tests. Finally, models for estimating the weld bead microstructure were developed by fusing all the information through a neural network modeling approach. A R value of 0.99472 was observed for modelling all zones of microstructure in the same ANN using Bayesian Regularization with 17 and 15 neurons in the first and second hidden layers, respectively, with 4 training runs (which was the lowest R value among all tested configurations). The results obtained prove that RNAs can be used to assist the project of welded joints as they make it possible to estimate the extension of HAZ.

The Gas Metal Arc Welding (GMAW) process is widely utilized in industrial production, requiring careful selection of appropriate procedures to ensure the highest quality. A key area of study closely related to GMAW quality is the control of process stability. This research presents a methodology for analyzing welding data to identify instability, thus enabling the development of a stability indicator. Our approach focuses on sensory fusion by integrating multiple sources of information, including sound signals, images, and current signals captured during the welding process. This work explores various configurations of variables to analyze the three primary transfer modes. Additionally, a comprehensive statistical analysis of the results obtained is conducted. Image processing techniques, sound analysis, and artificial intelligence methodologies are employed to enhance the analysis process.

The dataset was collected from experiments using the gas metal arc welding (GMAW) process. The experiments were planned with Central Composite Design to obtain a greater variety of data. This variability helps to develop a predictive model more generalistic with machine learning techniques. It was collected welding arc images and weld bead geometry images. Welding arc images were processed with a deep learning technique to detect drop detachment and short circuit transfer mode. These detections were useful to calc drop detachment frequency, short circuit frequency, and molten volume in every moment of GMAW process time. It was obtained the weld bead geometry parameters by process time too. All these data, joining input parameters were correlated, resulting in the datasets shown in this article.