Research On Thermal Damage Prediction Of Carbon Fiber-reinforced Polymer Laser Cutting Based On Acoustic Emission Sensor

Carbon fibre-reinforced polymer(CFRP)is one of the main application materials for the manufacture of high-strength and lightweight components of important aircraft such as satellites and airplanes.Milling,drilling and other conventional processing methods are inefficient and prone to tool wear.Nanosecond ultraviolet laser processing technology has the advantages of high processing efficiency and excellent processing quality,which is one of the important ways for CFRP processing.By monitoring the state of CFRP nanosecond laser processing and then evaluating the thermal damage,it is helpful to improve the design and manufacturing process.proposes a deep prediction model based on VGG-16 by constructing a large-scale data set,combining signal processing techniques,using signal time-frequency analysis means,and fusing convolutional neural network techniques around the acoustic emission signals released during CFRP nanosecond ultraviolet cutting process,including the following aspects:Building a CFRP nanosecond UV laser cutting acoustic emission signal acquisition platform,determining important parameters such as laser frequency,cutting speed,processing times,cutting angle,and performing signal collection to build a large-scale CFRP laser cutting acoustic emission signal dataset(1800 groups).using wavelet denoise method,combined with pre-processing methods such as band-pass filtering,de-averaging,and de-trending,the signal-to-noise ratio is greatly increased from 15.18 to 35.36.a signal endpoint detection algorithm based on short-time Fourier transform and spectral center of mass is proposed to accurately separate the signals under different process parameters and realize effective and error-free segmentation of signal data.Using the pre-processed signal as the analysis set,the statistical features such as acoustic signal energy counting and root mean square are used to explore the time-domain intensity change pattern,and the time-frequency domain transformation of the signal as a whole is performed by means of wavelet analysis to explore the influence behavior of the main frequency bands on the forming of the heat-affected zone.The results show that in the time domain,the RMS amplitude level of the time domain signal and the fluctuation amplitude of the energy count are more consistent with the width of the heat-affected zone.In the time-frequency domain,the frequency component levels in the range of 100-300 k Hz are consistent with the width of the heat-affected zone.A heat-affected zone prediction model based on the convolutional neural network VGG-16 is proposed,which takes the time spectrum after normalization and maximum pooling as input,and the width of the heat-affected zone as output,and the learning rate,dropout rate and weight decay rate are adjusted to optimize the model loss.The results show that the model can reduce the prediction error of the width of the heat-affected zone to 20?m.The Pearson's correlation coefficient is 0.95,and the R~2 correction determination coefficient is 0.91 which has good accuracy and generalization ability.