CNN and Spatial GRU for accurate non-destructive testing of welding and casting process defects

In modern industries, the rapid application of welding and casting technologies intensifies the need for accurate and reliable defect detection methods. For automatic and robust detection of welding and casting defects, this research paper proposes a two-dimensional Convolutional Neural Network and a Spatial Gated Recurrent Unit. The proposed model is trained with 128 × 128 resized images, 32 batch size, a 0.001 learning rate, 100 epochs, and 128 Spatial Gated Recurrent Unit hidden units using a cross-entropy loss function with softmax classification. The spatial feature extraction is performed by a two-dimensional Convolutional Neural Network, and the long-term dependencies are captured by implementing the Spatial Gated Recurrent Unit. The proposed model is validated using two benchmark datasets, such as the Welding Defect – Object Detection dataset and the Casting product image data for quality inspection dataset. The findings demonstrate the superiority of the proposed model with the testing accuracy of 97.9%, precision of 97.2%, and a Root Mean Square Error of 1.1 on the Welding Defect – Object Detection dataset. Based on the Casting product image data for quality inspection dataset, the proposed model achieves the testing accuracy of 97.9%, precision of 97.2%, and Root Mean Square Error of 1.4.