Text Extraction and Structuring of Standard Maintenance Documents for Metallurgical Continuous Casting Equipments

The large need of standard maintenance documents (SMD) for metallurgical continuous casting equipments brings a lot of repetitive, time-consuming and laborious work. Therefore through artificial intelligence and large language models, realizing the automatic generation of SMD has important significance and value. The basis of the work is the text extraction and structuring of a large number of existing SMD, to obtain a valuable data base that can be utilized. SMD are mostly in the form of tables with a relatively fixed structure. This paper proposes the extraction and deduplicating of complex tables, the denoising, coreference resolution and step sequence pictures recognition based on rule matching and the structured storage of the extracted contents through the key-value-pair tree to retain the original structural information of the tables. The proposed method is verified through the experiment of extracting and structuring 93 SMD provided by MCC Baosteel.