Automated assembly time prediction tool using predefined mates from CAD assemblies.

摘要

Current Design for Assembly (DFA) methods and tools require extensive amounts and types of user inputs to complete the analysis. Since the methods require extensive amounts and types of inputs, certain issues arise: the analysis can become tedious, time consuming, error prone, and not repeatable. These issues eventually lead to the DFA methods being used as a redesign tool or not being implemented at all.;The research presented in this thesis addresses the current DFA limitations and issues by developing and implementing an automated assembly time prediction tool that: extracts explicitly defined connections from SolidWorks assembly models, determines the structural complexity vector of the connections, and inputs the complexity vector into trained artificial neural networks (ANNs) to predict an assembly time. The automated assembly time prediction tool does not require any user inputs other than a mated assembly model. To complete the analysis with the automated tool, the user has to open up the assembly model and click on the developed SW add-in button. Since no additional inputs are required to complete the analysis, the results are completely repeatable when given the same SolidWorks assembly model to evaluate.;The results in this thesis show that the developed tool can predict a product's assembly time with as little as 4% error or with as much as +68% error depending on the ANN training set used. Eight different ANN training sets are tested in this thesis, the results show that larger more variable ANN training sets typically predict assembly times with less percent error than smaller less variable ANN training sets. Since the tool extracts mates from assembly models, the sensitivity of the method with respect to different mating styles is also investigated. It is determined that the mating style does have an effect on the predicted assembly time, but this effect is typically within the normal variation ranges of existing DFA methods.