Mining Process Factor Causality Links with Multi-relational Associations

To make knowledge-supported decisions, industrial actors often need to examine available data for suggestive patterns. As industrial data are typically unlabeled and involve multiple object types, unsupervised multi-relational (MR) data mining methods are particularly suitable for the task. Current MR association miners merely produce singleton-conclusions rules hence might miss multi-way dependencies. Our novel MR miner builds upon a relational extension of concept analysis to extract general associations. While successfully dealing with circularity in data, it avoids producing cyclic rules by limiting the description depth of relational concepts. Our rules' relevance was validated by an application to aluminum die casting.