Linked data (LD) is a rich format increasingly exploited in knowledge discovery from data (KDD). To that end, LD is typically structured as graph, but can also fit the multi-relational data mining (MRDM) paradigm, e.g. as multiple types and object properties may be used in the dataset. Formal concept analysis (FCA) has been successfully used as theoretical framework for KDD in a variety of applications , primely in clustering and association rule mining (ARM) tasks. As FCA applicability to LD is limited by its single data table input format, relational concept analysis (RCA) was introduced as a MRDM extension that successfully deals with links in the data, including cyclic ones. While RCA has been mainly adapted for conceptual clustering in the past, we present here an RCA-based ARM method. It exploits the iterative nature of pattern generation to cut cyclic references with a minimal loss of information. The utility of the rules discovered by our method has been validated by an application as a decision support in the aluminum die casting industry.