A Novel Hybrid Bees Regression Convolutional Neural Network (BA-RCNN) Applied to Porosity Prediction in Selective Laser Melting Parts

Convolutional Neural Network (CNN) is a Deep Learning (DL) technique used for image analysis. CNN can be used in manufacturing, for predicting the percentage of porosity in the finished Selective Laser Melting (SLM) parts. This paper presents a new approach based on Regression Convolutional Neural Network (RCNN) for assessing the porosity which was better than the existing image binarization method. The algorithms were applied to artificial porosity images that were similar to the real images with a 0.9976 similarity index. The RCNN yielded a prediction accuracy of 75.50% compared to 68.60% for image binarization. After the RCNN parameters were optimized using the Bees Algorithm (BA), the application of the novel Bees Regression Convolutional Neural Network (BA-RCNN) improved the porosity prediction accuracy further to 85.33%. When three noise levels were used to examine its sensitivity to noise, the novel hybrid BA-RCNN was found to be less sensitive to noise.