Machine learning-driven backpropagation neural network for robust prediction of surface roughness in Ti6Al4V abrasive water jet machining with experimental validation

In this work, machine learning driven back propagation neural network framework has been developed to predict the surface roughness during abrasive water jet machining (AWJM) of Ti6Al4V. The experiments were performed as per the Taguchi based L 27 orthogonal array, which generated the required dataset for developing the predictive model. A backpropagation neural network (BpNN) supported by a purpose-built graphical user interface (GUI) was used to formulate the data and allow multi-stage validation. To amplify the learning database, 50 synthetic datasets were produced and evaluated using K–fold Cross–Validation out of which, 70% of the data was used for training, 20% was used for testing, and 10% for validation. The 1–10–5 architecture achieved a high correlation coefficient ( R ) of 0.9728, with an average accuracy ranging between 90–95%. The model’s predictability was further verified through confirmatory experiments, yielding deviations of 0.12–7.09%. The findings confirmed that the proposed BpNN model effectively captured the non-linear relationships amongst the different process parameters used in this study. SEM analysis corroborated the predicted trends, showing smoother morphology at lower erosive energies and pronounced micro-ploughing, ridging, and brittle micro-chipping at higher parameter intensities. Sensitivity analysis revealed that water pressure and traverse speed were highly influential, contributing to about 36.7% and 29.7%, respectively, on surface roughness variability. Overall, the proposed BpNN-GUI technique provided a reliable foundation for digital-twin-based optimization of AWJM for difficult-to-cut titanium alloys.