Fig. 1 Prediction strategy for the relative density of materials manufactured by L-PBF, based on a combination of dimensionless parameters and ML method

Fig. 2 Feature selection based on four feature engineering methods (the final selected features are marked with black five-pointed stars), including a correlation analysis: a Spearman correlation analysis; three importance analysis: b Random Forest Model, c Linear Model, and d Lasso Model

Fig. 3 Classification accuracy of the four ML models on the sub-test set is quantitatively evaluated using four distinct accuracy metrics, where the score closer to 1 indicates higher classification performance

Fig. 4 Classification confusion matrices obtained from the application of four methods in the CoCrFeNiMn alloy system. a Model proposed in this work; b Phan's method; c single-system ML model based on process parameters; d cross-system ML model based on process and material parameters

Fig. 5 A quantitative comparison of the prediction performance of four different methods on the test alloys. a Classification scores of the four methods are shown across four metrics: Accuracy, Precision, Recall, and F1; b a comparison of the ROC curves of the four methods; c comparison of the AUC values for the ROC curves of the four methods

Fig. 6 RD classification effect of four methods on CoCrFeNiMn alloys in the $q_{{{\text{fraction}}}}^{*} - v^{*}$ space. According to Phan et al. [32], the two dotted lines at $q_{{{\text{fraction}}}}^{*} = 0$ and $q_{{{\text{fraction}}}}^{*} = 1$ separate the space into three regions, RD ≥ 99%, keyhole porosity and lack of fusion porosity. a Actual experimental values for the CoCrFeNiMn alloy; b–e represent the proposed model in this work, the Phan's method, the single-system ML model based on process parameters and the cross-system ML model based on both process and material parameters, respectively

Fig. 7 Implementation of the proposed model for classifying the RD of Ni41.8Co19Cr10Fe10Al1.5Ti2Mo2Hf0.1B0.1. a Confusion matrix of predicted RD versus measured RD; b ROC curve obtained from the classification results of the proposed model

Fig. 8 Comparison of defects of two samples #1 and #2. a SEM micrograph of #1; b SEM micrograph of #2. c 3D reconstruction results of defects in the CT results of #1; d 3D reconstruction results of defects in the CT results of #2; e statistical histogram of defect volumes for #1; f statistical histogram of defect volumes for #2