Machine Learning for Costing Sheet Metals

Machine learning has proven to be an effective method for quantify-ing the costs of mechanical parts early in the design process. One of the most complex aspects remains the application of the costing method in real industrial contexts, such as made-to-order manufacturing. This paper introduces a novel cost modelling method based on machine learning for the early design phase. The training dataset is generated using an automatic and analytic 3D-based software tool for process planning, time, cost and resource estimation. Subsequently, the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology is applied to preprocess the data. CRISP-DM is a data science process model that outlines the data mining lifecycle and offers flexibility for tailoring the model to specific project goals. The proposed approach has been effectively applied to develop resource prediction models for manufacturing sheet metals (i.e., cutting and bending) that can be used during early design. These resource prediction models serve as the foundation for cost calculations. The mean accuracy of the obtained cost models is lower than 10%, a value accepted by design engineers during preliminary design and feasibility studies.