Abstract: To solve the problem of low efficiency in defect collection of ordinary railway bridge equipment and improve the efficiency of on-site inspection operations, this paper proposed a model that integrated pre-trained model RoBERTa wwm ext, bidirectional long short-term memory network, and attention mechanism (referred to as the improved BiLSTM Att model), namely the text classification model for bridge equipment defect of ordinary railway. It adopted the model and aimed to classify the defect text of the masonry bridge defect parts (bridge deck, bearings, piers and abutments, beam arches, bridge crossing hydrology, ancillary facilities). Experimental verification was conducted on the defect description text data of 15 railway group companies' ordinary railway masonry bridge equipment. The results indicate that the accuracy, recall, and F1 score of the improved BiLSTM Att model have all reached over 90%, and these indicators have significantly improved compared to the comparative model. The improving BiLSTM Att model can effectively identify defects in bridge equipment and assist in on-site bridge equipment inspection operations.