[6B1] Natural language-based information in intelligent condition monitoring and control
E Juuso
University of Oulu, Finland
Materials in natural language can provide useful information about variable interactions. Generative artificial intelligence (AI) tools are introduced for finding connections within this material. However, everything remains in natural language. A consistent representation is needed to use them with data based on measurements in quantitative analysis. Generalised norms provide informative features from measurements and signals. Intelligent indicators in the range [–2, 2] are obtained by using generalised norms in the non-linear scaling. These indicators are also suitable for the temporal analysis. For information which does not contain numerical values, the corresponding scaled values can be obtained by using interactions with labels such as {very low, low, normal, high, very high} or {fast decrease, decrease, constant, increase, fast increase}. The linguistic terms can be interpreted as fuzzy numbers, which can be made sharper or wider with powering modifiers ‘extremely’, ‘very’, ‘more or less’ and ‘roughly’ and then processed with the conjunction, disjunction and negation. All these can be represented with mathematical expressions within the range [–2, 2]. The methodology is used in the analysis of the event data: the information represented with these numbers produces a real-valued relation base to be used to analyse interactions. Known relationships can also be handled with the fuzzy calculus and the extension principle to keep the uncertainty in the calculation. Hybrid models can utilise both crisp and uncertain information in condition monitoring and control.
Keywords: natural language, non-linear scaling, intelligent indices, fuzzy set systems, event data, condition monitoring.
Keywords: natural language, non-linear scaling, intelligent indices, fuzzy set systems, event data, condition monitoring.
