Data-driven causal knowledge graph construction for root cause analysis in quality problem solving

Abstract

Root cause analysis (RCA) plays an essential role in quality problem solving (QPS). Due to the difficulty of obtaining causal knowledge of quality problems, companies often rely on expert experience and conventional RCA tools when conducting RCA. Rich QPS data have remained mostly untapped but provide the potential for causal knowledge mining, while the semistructured nature of these data poses enormous challenges to this task. Thus, we propose a data-driven framework to mine large-scale causalities between quality problems and production factors from QPS data and exploit a causal knowledge graph for quality problems (QPCKG) to express these causalities. We first classify QPS data to identify the data containing causality. The causal linguistic patterns are then employed to extract cause slots and effect slots from these data. Subsequently, we apply the BiLSTM-CRF to extract the core content of problems. A vertex fusion method is last proposed to integrate discrete causalities into QPCKG. The approach is validated in a real-world application at a leading automotive company. Three potential applications of the QPCKG are demonstrated for quality diagnosis and prediction. The QPCKG reveals a grand picture of the core interaction mechanism of product quality and production factors and provides decision-making support for RCA.