製品情報のテキストマイニングによる設計開発・保守のナレッジマネジメント

（製品不具合のテキストマイニングによる故障事象分類）

抄録

In recent years, Japanese manufacturing industry has been facing a skill succession problem due to the retirement of skilled workers and a shortage of young workers, despite the increasing complexity of technology. To solve this problem, there is a need to extract and utilize tacit knowledge and skills, but a method for utilizing disconnected information has been expected but not yet established. For instance, in the field of product failure, a method for appropriate integration of such fragmented information is expected to obtain systematic information about failure propagation. In this study, we obtain groups of failure cases with similar failure mechanisms from past automobile failure records (FPRs) by the following steps: 1. create a network among words such as part names or failure types (failure BOM), based on a single FPR; 2. create a network among FPR documents based on the common nodes of the failure BOM; 3. apply community detection algorithms to the network among FPR documents. As a result, comparing with the manually assigned labels about symptom and cause, the resulting community boundaries matches the label boundaries. Under the assumption that the failure BOM represents failure propagation, the proposed method can be evaluated as an FPR classification based on the likelihood of failure propagation. The proposed method is effective in indicating propagation-prone failure and in getting partial failure BOM based on failure propagation likelihood, which are expected to aid human analysis of faults.