Surface quality predictive model for ultraprecision fly-cutting based on an integrated hybrid CNN-LSTM

The emergence of predictive quality has significantly changed manufacturing processes and facilitated real-time product quality assessments. Ultraprecision fly-cutting (UPFC) demands a near-perfect, mirror-like surface without further machining. On top of that, even a minor deviation in the cutting process may substantially affect product quality. Traditional methods are frequently constrained by limitations in accuracy and often make them unsuitable for real-time applications, driving the adoption of intelligent approaches. The current work tackles this challenge by developing a surface quality predictive model for UPFC using a lightweight and robust deep learning (DL) model trained on data collected from tooltip vibrations. In this work, an integrated hybrid model using a convolutional neural network (CNN) and long short-term memory (LSTM), namely 1DCLM, was developed. For a fair comparison, the performance of 1DCLM model was compared with the one-dimensional CNN (1DCNN) model. Six datasets were prepared where each single vibration signal was split into several sub-signals, generating ample samples even from a limited number of cutting experiments to train and test DL models. Experimental results show that the proposed approach achieves state-of-the-art results, with the integrated hybrid 1DCLM model attaining 97.5% accuracy in contrast to 93.8% achieved by 1DCNN. Lastly, the proposed model’s exceptionally low inference time substantiates our claim of making it ideal for real-time application.