A new ontology-based approach to automatic information extraction from speech for production disturbance management

The goal of our research was to design a methodology for extracting systematized knowledge from free speech. The sources of knowledge in our analysis were records of production meetings, focused on production disturbance (PD). The main obstacle is to properly identify the specific meaning of words, in a specific, usually narrow, industry. Machine learning based on data from production records has been increasingly used to build such models. In the case of manufacturing plants with diverse production programs, acquiring the right number and structure of data is not possible; hence, proper identification of such terms is for classical NLP tools, even supported by large language models, not possible. We have attempted to use AI and NLP tools from recorded production meeting recordings to create and continuously update PD's knowledge as a supplement to data from documentation. This is an approach not previously known in the field of production management. The solution we developed consists of an expert-defined specific ontology, based on the pre-processed speeches. At this stage, a lexicon (vocabulary) is also created, supporting the transformation of the speeches into interpretable texts. The model ontology formulated this way is then used to analyze consecutively provided meeting records and thus update the operational ontology. In our research, we used the materials provided to us in the form of records of production meetings from a medium-sized pressure foundry. The obtained results confirm that the adopted knowledge model and the algorithms might be successfully utilized to solve real-world manufacturing problems.