Solid particle erosion would result in severe damage to elbows and other fittings, even safety accidents. It is an inevitable issue in oil and gas production and transportation. Thus, accurate erosion prediction is critical in oil and gas industry research. Addressing current research and model limitations, this work introduced an AI-driven, ensemble learning-enhanced Relevance Vector Machine (RVM) model for precise erosion rate prediction in oil and gas elbows. The approach built multiple RVM base learners and adjusts their weights by Adaptive Boosting (AdaBoost) algorithm to form a strong learner for effective erosion prediction. The results demonstrate that the RVM performs better as the base learner with about 90% of the data within the allowable error range compared to the Back Propagation Neural Network (BPNN) and Support Vector Machine (SVM). Compared with BPNN and SVM model optimized by AdaBoost algorithm, the proposed method shows higher prediction accuracy. Its prediction relative errors are all below 80%, the average absolute error and root mean square error are 0.0153 and 0.0274, respectively, with R2 of 0.933. This approach serves as an effective means for predicting erosion in oil and gas pipelines, offering new perspectives for ensuring safe production and transportation under multiphase flow conditions.