An unsupervised clustering method for assessing the degradation state of cutting tools used in the packaging industry

The objective of the present work is to develop a method for the identification of the degradation state of cutting tools (knives) used in the packaging industry. The main difficulties to be addressed are that i) only measurements of a physical quantity indirectly related to the knives degradation are available and ii) only the beginning and the end of operation of the knives are known, whereas no information is available on the component degradation state during its operation life. A method to identify the component degradation state is here proposed. First the general setting for extracting health indicators to measure the amount of knife degradation from a set of signals measured during operation is discussed. Then, an optimal subset of health indicators is selected based on monotonicity and trendability indexes. Finally, the optimal subset of health indicators is fed to a Fuzzy C-Means (FCM) clustering algorithm, which allows assessing the knife degradation state. The application of the proposed method to real condition monitoring knife data is shown to lead to satisfactory results. S2) Selection among the set of extracted features of an optimal subset of Health Indicators (HIs) for the identification of the component degradation; S3) Unsupervised clustering of HI data in order to identify the component degradation state. The main novelty of our work lies in the strategy to select the optimal subset of HIs based on monotonicity and trendability indexes and in its application to knives used in the packaging industry. The paper is organized as follows: in Section 2, the general setting for HI development is proposed; in Section 3, Fuzzy C-Means (FCM) clustering for identifying the knife degradation state is discussed. The application of the methodology to Tetra Pak® A3/Flex filling data is described in Section 4, whereas in Section 5 conclusions are drawn.