AI-Based knowledge extraction for automatic design proposals using design-related patterns

Abstract

Engineering competence and the digitization of all processes along the product development process are highly decisive for today’s success of industrial companies. The design process is very individual and strongly based on design engineers’ experience. Part of this knowledge and the result of the design approach are fixated in the existing variations of the product generations, but are difficult to extract and to formalize. Conclusions about design-related patterns between products of different generations or variants can be drawn from the model tree representing the design engineer’s thinking process for each individual CAD model. However, the model tree has hardly been used so far. The aim of this paper is to examine whether there exist any common design patterns between CAD models of certain component classes by the exemplary use case in the area of mechanical engineering. To identify patterns and to extract knowledge out of complex data sets, Machine Learning (ML), especially Deep Learning, has proven an immense capability. Finally, based on the learned patterns, meaningful next design steps are to be proposed in the form of an assistance system. The results show that there exist common design patterns for various classes of components. It is illustrated on an exemplary component class that those patterns can be used to train an assistance system based on Recurrent Neural Networks (RNNs). The corresponding design patterns were extracted from data of an industrial application partner. By transferring these design patterns to the development of new product generations or variants, on the one hand the design process itself and thus the time to market can be shortened. On the other hand, the knowledge from previous product generations contained in those patterns can be preserved. For further research the design patterns of CAD models extracted by ML algorithms is a contribution to faster knowledge extrapolation.