Abstract: With the increase of the service time of the lifting wire rope and the influence of complex working conditions such as tensile, bending and torsional loads, it is easy to cause friction damage to the rope, resulting in wire breakage and wear of the rope, reducing the effective bearing area and bearing strength of the lifting wire rope cross-section, and even inducing rope breaking accidents. Therefore, the quantitative identification of friction damage of wire rope is very important for predicting the residual load strength of wire rope and improving the load safety of wire rope. This paper took 6×19+IWS wire rope commonly used in lifting system as the research object. Based on the self-made wire rope friction and wear test bench, the friction and wear experiments of wire rope under different slip amplitudes and cycles were carried out. A quantitative characterization method was proposed for the cross-sectional area loss of wire rope. The machine learning database of cross-sectional area loss and deep learning database of cross-sectional area loss of wire rope were constructed by using discrete wavelet denoising, eigenvalue correlation analysis and principal component analysis to process the damaged signal. A quantitative recognition model of friction damage of steel wire rope was obtained by designing and training several algorithms. The results showed that the optimal denoising of the damaged signal was the 3-layer decomposition of the basic wavelet db4 and the signal-to-noise ratio could reach 25.2. There was a strong linear correlation between some characteristic values of damage signal, the linear correlation coefficient between mean value and root mean square characteristic value was as high as 0.98, the linear correlation coefficient between peak factor and margin factor was as high as 0.89, and the linear correlation between waveform factor and pulse factor was as high as 0.9. The linear correlation coefficient between wavelet energy and peak-to-peak value was higher than that between wavelet energy and other eigenvalues and was 0.8. After principal component analysis, the feature value of the reduced signal was reduced from 15 to 7 and the cumulative contribution rate was 99.19%, which reduced the complexity of signal processing and retained the damage feature information. The damage signal wavelet time-frequency graph could effectively reflect the damage degree of wire rope cross-sectional area. The larger the damage of wire rope cross-sectional area, the darker the color of damage signal wavelet time-frequency graph. The performance of different quantitative recognition models was different, in order from high to low: IncepRegCNN neural network > PSO-BP neural network > BP neural network, in which IncepRegCNN neural network RMSE and MAE were reduced by more than 50%, and R² was increased by more than 15%. The maximum absolute error of IncepRegCNN neural network for quantitative identification of the cross-sectional area loss of steel wire rope friction damage was 0.4%, which had a high recognition accuracy.