Artificial Neural Networks for Weld Joint Strength Prediction in Various Welding Techniques

• Published in: 2024 5th International Conference on Recent Trends in Computer Science and Technology (ICRTCST) pp. 188 - 192
• Main Authors: Panigrahi, Bhawani Sankar, H R, Manjunath, Chipade, Amar, Tilak Babu, S. B G, G., Pavithra, Hussain, Beporam Iftekhar
• Format: Conference Proceeding
• Language: English
• Published: IEEE 09.04.2024
• Subjects: Artificial neural networks; Artificial Neural Networks (ANN); Feature Extraction; GMAW; Laser Welding; Machine Learning; Predictive Modeling; Predictive models; Quality Control; Resistance; Shape; Training; Ultrasonic; Voltage; Weld Joint Strength Prediction; Weld Quality Assurance; Welding; Welding Process Optimization; Welding Techniques
• DOI: 10.1109/ICRTCST61793.2024.10578356

It is very important to be able to guess how strong weld joints will be because it affects the structural stability and performance of the parts that have been welded. In recent years, artificial neural networks (ANNs) have become useful for predicting complicated relationships in many fields, such as engineering and materials science. Artificial Neural Networks are the main focus of this research to guess how strong weld parts will be when they are made using various welding methods. The research entails gathering empirical data from a variety of welding processes, including arc welding, resistance welding, and laser welding, on a wide range of materials and joint configurations. Input variables such as welding current, voltage, speed, material characteristics, and joint shape are included in the files. The result of the testing processes is the strength of the weld bond. A neural network is utilised to create prediction models that accurately depict the complex patterns and non-linear connections seen in welding data. The network structure is carefully improved by considering factors such as Number of layers, neurons per layer, and activation functions. The acquired data is fed into the network throughout the training phase, and the weights are modified repeatedly to lower the frequency of incorrect predictions. To test and confirm the models that were created, different datasets that were not used during the training phase are employed. The accuracy, precision, and generality of artificial neural network (ANN) models in various welding processes are tested. Through comparison studies, it is demonstrated that Artificial Neural Networks (ANNs) outperform standard empirical models in predicting the strength of weld joints. The results show that artificial neural networks can accurately and reliably guess how strong weld parts will be for a wide range of welding methods. The new models are better than the old empirical models. This shows that artificial neural networks (ANNs) could be useful for improving and making sure the quality of welding processes. The results of this research help us understand better the complex relationships that determine how strong weld parts are. They also lay the groundwork for using AI in welding technology.