A data-driven approach for constructing the component-failure mode matrix for FMEA

• Published in: Journal of intelligent manufacturing 2020, Vol.31 (1), p.249-265
• Main Authors: Xu, Zhaoguang, Dang, Yanzhong, Munro, Peter, Wang, Yuhang
• Format: Article
• Language: English
• Subjects: Algorithms; Automotive parts; Business and Management; Cluster analysis; Clustering; Control; Data mining; Failure analysis; Failure modes; Feature extraction; Knowledge acquisition; Knowledge representation; Machines; Manufacturing; Mapping; Mechatronics; Processes; Production; Robotics; System reliability; Texts; Vector quantization
• ISSN: 0956-5515; 1572-8145
• DOI: 10.1007/s10845-019-01466-z

Failure mode and effects analysis (FMEA) is one of the typical structured, systematic and proactive approaches for product or system failure analysis. A critical step in FMEA is identifying potential failure modes for product sub-systems, components, and processes, for which component-failure mode (CF) knowledge is necessarily needed as an important source of knowledge. However, this knowledge is usually acquired manually based on historical documents such as bills of material and failure analysis reports, which is a labor-intensive and time-consuming task, incurring inefficiency and plenty of mistakes. Nevertheless, few existing studies have developed an effective and intelligent approach to acquiring accurate CF knowledge automatically. To fill the gap, this paper proposes a method to construct the CF matrix automatically by mining unstructured and short quality problem texts and mapping as well as representing them as CF knowledge. Starting with mining the frequent itemsets of failure modes through Apriori algorithm, the method uses the semantic dictionary WordNet to find synonyms in the set of failure modes, based on which the standard set of failure modes is finally built. Subsequently, upon the previous work and components set, we design the component-failure mode matrix mining (CFMM) algorithm and apply it to establish the CF matrix from unstructured quality problem texts. Lastly, we examine the quality data of the seat module of an automobile company as a case study in order to validate the proposed method. The result shows that the failure mode extraction method with standardized features can extract failure modes more effectively than the FP-growth and K-means clustering methods. Meanwhile, the devised CFMM algorithm can extract more combinations of CF than the FP-growth method and build a richer CF matrix. Although different industries have distinct domain characteristics, our proposed method can be applicable not only to manufacturing but also to other fields needing FMEA to enhance product and system reliability.