Aiming at the problem that the welding quality of steel box girder is difficult to judge, this paper proposes a strategy of combining the intelligent BP neural network algorithm with the fuzzy set theory. First, the BP neural network is modeled according to actual needs, and then the BP neural network's predicted output value is combined with the fuzzy set theory. Combine theories. Finally, by comparing with the original BP neural network algorithm, the superiority and robustness of the fuzzy BP neural network algorithm proposed in this paper are determined.