Fig. 1. Commonly used machine learning (ML) algorithms and their application in various processes of laser micro/nano machining

Fig. 2. Deep learning models generally used in laser intelligent machining and classification process

Fig. 3. Optimization of laser impact drilling process parameters by fully connected neural networks and genetic algorithms

Fig. 4. Machine learning applied to laser parameter recognition^[81]. (a) Schematic of experimental setup for real-time closed-loop feedback realized by convolutional neural network (CNN); (b) CNN architecture; (c) CNN predicting the remaining number of pulses until breakthrough of a thin film; (d) predicting a beam transformation (translation and rotation) by CNN

Fig. 5. Machine learning applied to laser polarization detection^[83]. (a) Schematic of colorimetric polarization-angle detection; (b) schematic of the nanopillar; (c) prediction accuracy of polarization angle by CNN

Fig. 6. Machine learning detecting flow and results of pore generation during L-PBF^[85]. (a) Thermal images, high-speed X-ray images and corresponding multiphysics simulation of intrinsic keyhole oscillation and perturbative keyhole oscillation; (b) time-series signal of the average light emission intensity around the keyhole extracted from thermal images series;(c) mislabeled rate and accuracy of machine learning detecting pore as a function of scalogram window length; (d) architecture of the proposed network for classifying “non-porous” and “porous”

Fig. 7. Bi-stream DCNN for recognition of defects induced by improper SLM process conditions^[58]. (a) Microscope view of defects; (b) variation of classification accuracy with the number of iteration; (c) performance comparison among DCNN, HoG and Visual words; (d) bi-stream DCNN architecture

Fig. 8. MsCNN for recognition of defects induced by improper L-BPF process conditions^[53]. (a) Representative examples of the six different powder bed anomaly classes chosen by the authors; (b) MsCNN architecture; (c) performance comparison among Bow, CNN and MsCNN; (d) heat exchanger printed with L-BPF and its 3D rendering image

Fig. 9. Classification of defects caused by improper SLM process conditions using acoustic signals^[61]. (a) Optical microscopy cross-section images of defects produced with three laser energy densities; (b) scheme of the sound detection system; (c) classification results

Fig. 10. Machine learning applied to laser micro/nano machining process control^[62]. (a) Schematic of the experiment; (b) self-correction during processing by machine learning; (c) arbitrary patterns processing by machining learning

Fig. 11. Predicting laser processing results using adversarial generative networks^[71]. (a) Scheme of predicting processing results using generative adversarial networks; (b) comparison of predicted and experimental results; (c) neural network predicted results show the existence of processing limit distance

Fig. 12. Prediction of 3D surface topography for laser machining using deep learning^[73]. (a) Schematic of the laser machining setup; (b) illustration of the sample characterization method; (c) examples from the training data set

Fig. 13. Prediction of melt pool temperature in directed energy deposition using LSTM^[90]. (a) Schematic diagram of the LENS 450XL system; (b) diagram of a typical LSTM; (c) comparison of experimental and predicted melt pool temperatures