Supervision with hierarchical classifiers for a joining process

The work presented in this paper deals with a diagnostic problem related to stability of the join in joining pressing. A fuzzy classification model based on membership functions of objects (learning samples) in the feature space and rule-based integration of explizit expert knowledge is applied. 28 features are determinated from the measured force behavior during pressing. Considering three particular tasks based on different criterions like amount of force necessary for pressing back, the amount of shaft oversize and results of visual estimation of surface quality three idendependent class models are designed and integrated into a global model by a rule-base.