Clustering algorithms for deductive FMEA and failure identification in complex production systems

The increasing complexity of manufacturing systems and machineries compels advanced maintenance and reliability engineering techniques to enhance the system OEE. The lack of unified knowledge on the design and the behavior of the whole manufacturing line affects the implementation of accurate and tailored maintenance tasks. While inductive top-down approaches (i.e., FMEA and FMECA) to understand the failure behavior of a system are well-known by industry their implementation in practices is inhibited by the lack of robust failure probability data, which are hard to be quantified. In this paper vintage clustering algorithms are applied to quantify the failure probability values of components and functional groups in real-world high complexity manufacturing lines. The failure probability of components and functional groups are thereby deducted from spare parts demand records. The illustrated procedure has bee validated through a real-world application of complex manufacturing lines in the tobacco industry.