CBN cutting tool’s surface roughness and tool wear prediction using JOA-optimized CNN-LSTM

• Published in: Scientific Reports
• Main Authors: Subash Khetre, Arunkumar Bongale, Satish Kumar
• Format: Article
• Language: English
• Published: Nature Portfolio 2025-11-01
• Subjects: Surface roughness; Flank wear; CNN; LSTM; JOA; Inconel 718
• Online Access: https://doi.org/10.1038/s41598-025-29658-z

Abstract Nickel-based superalloys like Inconel 718 exhibit large machining challenges attributable to their poor thermal conductivity and pronounced work hardening behaviors. They normally contribute to quick tool wearing away and poor surface quality. In order to get around this problem, a hybrid deep learning system that fused a CNN and an LSTM was devised and successfully optimized through the JOA. The model was incorporated in MATLAB/Simulink, making it possible to predict surface roughness and flank wear in real-time while hard turning using a CBN tool under MQL conditions. The experimental data used for model training and validation were derived from 27 full-factorial machining trials that covered a range of cutting speeds, feed rates, and depths of cut. The data was preprocessed through normalization and outlier removal using the IQR and Z-score methods. The CNN–LSTM model that was optimized by JOA displayed remarkable prediction power with R = 0.9991, RMSE = 0.0095, and MAPE = 2.21%, thus being far superior to the conventional models, like SVM, ANN, and ANFIS, etc. The findings indicate the model’s ability to precisely understand complex nonlinear interactions between the machining parameters and the corresponding responses, hence, the model’s strong generalization across different cutting conditions. The inclusion of the MATLAB/Simulink environment extends the model’s real-time deployment potential, digital-twin compatibility, and scalability, providing a low-cost and sensor-free solution that is perfect for smart and sustainable manufacturing. This study offers a scientifically interpretable and industrially deployable method for predictive modeling in the machining of hard-to-cut materials.