The aim of this work is to define, implement, test, and validate an AI methodology using existing machine learning (ML) algorithms to predict sand erosion in 90° elbows for a broad range of multiphase operating conditions. Based on information obtained from the experimental UT wall thickness loss data collected for different flow regimes (gas-sand, liquid-sand, dispersed-bubble, churn, annular, and low liquid loading multiphase flows), the methodology has been developed to predict the maximum erosion magnitudes in standard metallic elbows. In order to expand the range of application of the method to situations where data is not available, the erosion database has been expanded by including state-of-the-art validated CFD simulations and 2-dimensional CFD-based mechanistic model predictions. The ML algorithms, including elastic net (EN), random forest (RF), support vector machine (SVM), extreme gradient boosting (XGB), and k-nearest neighbors (KNN) classification. The models are optimized using cross-validation and their performance is evaluated by different metrics. More than 650 case studies from previous literature as well as ongoing research have been used to train and test the ML models. The RF and XGB results show the overall best performance for a variety of flow conditions and pipe sizes. The resulting technique helps in saving time and resources to predict erosion in elbows and develop operational limits both within and beyond the current experimental domain while utilizing the most common production input parameters used by oil and gas production operators and other industrial applications.