Predictive technology assessment by means of a structure-based method of machine learning

Abstract

From a user perspective, the current development of the generic term Industry 4.0 increasingly moves its orientation towards flexible production. Due to increasingly variable products with small quantities and the resulting high degree of adaptability of a plant over its entire operating phase, the need for rapid production commissioning gives rise to the demand for live commissioning support and technology evaluation of induced production start-ups. Classification axioms can be formed by 1-class learning procedures for the predictive state evaluation of subsequent production start-ups based on collected machine and process data from past production start-ups. The starting point is an adaptive algorithm that performs a dynamic tolerance band formation based on different criteria, emphasizing on adaptive characteristic segmentation. This first step represents comprehensive condition monitoring. Based on this algorithm, correlation considerations can be performed on the data structure, the measured variables, and the diagnostic parameters. Moreover, the structure of production systems can and should be included in the analyzation, so that probabilistic causalities can be postulated and then be added to the underlying data sets for quantification. Using these adaptive structure-based segmentations is the first step to interpret data sets of new production systems without the need for complex pre-configuration.