Cognitive Failure Cluster as an Approach to Enhance the Precision of a Self-optimizing Process Model

An important indicator for the acoustic quality of rear axle drives is the contact pattern of the gear sets. Due to the complex interactions in the production process numerous factors have influence to the result to the contact pattern. In general, their effect on product variations is not fully comprehended and interdependencies are unidentified. This impedes the design and control of the production process based on a holistic analytical model for new variants fulfilling the acoustic requirements. Prior projects have shown that the self-optimization approach provides convincing outcomes but need a long processing time for the training of the artificial neural networks and the necessary iterations until a satisfying precision for the predicted process parameters is achieved. Also it can occur that the algorithm is not converging and therefore no satisfactory result is turned out at all. In this paper an approach is presented combining the flexibility of self-optimizing systems Cognitive Tolerance Matching (CTM) with the higher precision of delimited solution finders called the Cognitive Failure Cluster (CFC). The improvements provided by the clustering of the optimization program are evaluated regarding the training time and the precision of the result for a production lot of bevel gear sets.