A new approach to risk assessment in failure mode and effect analysis based on engineering textual data

Failure Mode and Effect Analysis (FMEA) is a valuable tool for improving the quality of products and service systems. However, traditional FMEA methods require examining various engineering textual materials and attending multiple meetings, which can be time-intensive. Additionally, accurately evaluating the severity (S), occurrence (O), and detection (D) of failure modes is essential to ensure the accuracy of FMEA results. Despite efforts by researchers to improve the efficiency of FMEA, the evaluation of these risk factors (S, O, and D) still relies too heavily on a manual and inefficient process. To address these issues, this paper proposes a machine learning-enabled FMEA approach. This new approach combines the strengths of the BERT (Bidirectional Encoder Representations from Transformers) model in transforming textual failure descriptions into word vectors, the advantages of the VSM (Vector Space Model) in determining semantic similarity between different failure modes, and the merits of TOPSIS based on objective weights in handling multicriteria risk assessment. The proposed approach was applied to an actual case study for failure mode and effect analysis in auto parts processing. Comparisons between the proposed method and conventional FMEA were conducted to demonstrate the effectiveness of the new approach.