With the widespread application of ML in studying PSP relationships within the realm of AM, the accuracy of data becomes paramount for the performance of ML models on small datasets. In this study, the DBSCAN algorithm was utilized to identify and eliminate noise points from the porosity dataset of IN718 as-built parts, which were fabricated by QCW laser DMD. The primary objective was to enhance the dataset's quality and consequently improve the predictive performance of the model. The elimination of noise points resulted in a notable improvement in both the fitting degree and predictive accuracy of the ML model. This improvement was clearly reflected in the substantial increase in R2 values, which rose from 0.367 to 0.915 for the training set and from 0.546 to 0.959 for the test set. Furthermore, SVR was employed to generate process parameter maps, revealing the intricate effects of various process parameters, including laser peak power, scanning speed, powder feeding rate, and pulse frequency, on the porosity of the components. Experimental validation was conducted using optimized combinations of these process parameters, further corroborating the enhanced accuracy of the ML model trained on the improved quality dataset.

Predicting the porosity of as-built additive manufactured samples based on machine learning method for small datasets