Predictive Modeling of Assembly Time Using Machine Learning

Abstract

Assembly time estimation is an essential part of manufacturing directly affecting the control of cost, productivity, as well as delivery performance. The traditional estimating techniques with the help of expert heuristics or rigid parametric models lack effectiveness in understanding the inter-complexity of new-age products. The current research suggests the use of machine learning (ML) algorithms—namely, linear regression, decision tree, and random forest models—for estimating assembly time with respect to important design parameters such as part count, joining practices, tolerancing, and indices of complexity. The simulated as well as actual-world datasets obtained from Kaggle.com were utilized for training as well as cross-validation. The outcome suggests that ML algorithms, with special attention given to random forest regressors, deliver significantly better predictive capabilities compared to conventional estimating techniques. Feature importance evaluation identifies part count along with the complexity of the design as significant determinants of assembly time. The new approach presents an extensible, accurate, and flexible remedy enabling better DFMA (Design for Manufacturing and Assembly) practice, as well as process optimization in the field of manufacturing. The given data file obtained from Kaggle.com.