LAUSR

As an indispensable technology of intelligent education, intelligent tutorial algorithms for solving mathematical or physical problems have attracted much attention in recent years. Nevertheless, since solving mechanics problems requires complex force analysis and motion analysis, current researches are mainly focus on solving geometry proof problems and direct circuit problems. There are some inherent challenges on developing such algorithms, including the low intelligence, mobility and interpretability of the comprehension algorithm. Therefore, this article develops a novel algorithm for solving mechanics problems. First, we propose a comprehension model for mechanics problems and convert problem understanding into relation extraction. Furthermore, a novel neural model combining pretrained model BERT and graph attention network (GAT) is proposed to extract the direct conditions of input mechanics problems. Second, a hidden information mining method is proposed for supplementing the conditions of the input problem. Third, a predicate logic based algorithm is proposed for force analysis. Finally, a solving algorithm is presented for choosing equations to acquire the solutions. Solving experiments and sensitivity analysis are provided to demonstrate the effectiveness of the proposed algorithm.