[223] Integration of natural language in intelligent condition monitoring

A consistent representation is needed to understand the meanings of measurement values and use them together with knowledge-based information. The non-linear scaling approach is used for any numeric values, including measurements, features, indices and indicators. The scaled values in the range [–2, 2] are interpreted in natural language labels, for example very low, low, normal, high and very high. The expert knowledge is represented in the same range. Parameters of the scaling functions are obtained from the numeric values and modified to ensure the monotonic increase. Intelligent condition and stress indices are calculated from consecutive samples of the waveform signals by using generalised norms and the non-linear scaling approach. Uncertainty, fluctuations and confidence in results are estimated by a difference of norms of high positive and negative order, respectively. Temporal analysis is based on the scaled values: trend indices are calculated by comparing the averages in the long and short time windows, a weighted sum of the trend index and its derivative detects the trend episodes and severity of the trend is estimated by also including the variable level in the sum. Risk indices are obtained from stress contributions. All indices and their changes within time are in the range [–2, 2] and represented in natural language.
Keywords: knowledge-based systems, non-linear scaling, intelligent indices, fuzzy set systems, event data, condition monitoring.