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

Abstract The goal of our research was to formulate an original set of Artificial Intelligence tools to extract knowledge from sources not intended for this purpose. The sources of knowledge in our analysis were records of production meetings, focused on production disturbance (PD) management. We used NLP tools and a procedure for transforming speeches into a knowledge model, both supported by domain ontologies. We found that using commonly available ontologies in such a case is not always possible. 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 transformation of the processed speeches into interpretable texts. The model ontology formulated this way is then used to analyze consecutively provided meetings’ 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. Thanks to the developed algorithms, it was possible to identify events related to the individual production cells, their nature, and the basic relationships between the events, the products, and the devices. The obtained results confirm that the adopted knowledge model and the algorithms might be successfully utilized to solve real-world manufacturing problems.