Session: 02-10-01: Sustainable Design

Paper Number: 145347

145347 - A Conceptual Costing Software Tool Based on Machine Learning for Turbomachines. 

Nowadays, the cost of a product is a design driver as important as performance, environmental sustainability, and quality. Determining the production cost of a product during the preliminary design phases (e.g., conceptual design) is essential for a company’s competitiveness. Cost-estimating methods based on parametric approaches (top-down) are the most suitable in these phases of product development. Parametric cost estimation methods work well when relationships between design variables (namely, cost drivers) and the cost are easily identifiable. The recent innovations introduced by Industry 4.0 (e.g., data mining and the Internet of Things) provide new tools and opportunities to overcome the issues of Cost Estimation Relationship methods. Machine learning (ML) applications have drawn interest because they make industrial processes more efficient and make complex parameter causations easier to understand. ML is typically more effective in manufacturing cost estimation than conventional statistical and mathematical models. 

In the scientific literature, several approaches aim to create cost models where cost formulas are made using regression analysis, neural networks, and decision trees. The methods are implemented using data analysis tools that are not entirely suitable for product designers or cost engineers. The potential benefits of ML methods are not wholly exploited due to the limited effectiveness of the software tools. 

The paper presents a software tool with a dual purpose. An administration module allows cost engineers to develop parametric cost models based on machine learning algorithms. The cost modelling process is based on the CRISP-DM method. It supports cost engineers in all phases, such as collecting and preparing data, training and evaluating the models and benchmarking. The machine learning engine is based on the ML.NET library (by Microsoft). The cost models are stored in a database to make them available from a user module conceived for design engineers. This module allows designers to quickly and accurately determine the cost of a part by providing the cost drivers required by the cost model as input. Through feature importance algorithms, the cost model indicates the most critical cost drivers from a cost point of view. 

The tool was tested in collaboration with a company that designs turbomachinery. During the experimentation, cost models were developed for the main parts of the rotor of an axial compressor (e.g. discs, spacers, shafts, blades). The test allowed the authors to evaluate quantitative indicators (e.g., time required for a cost estimate) and qualitative indicators (e.g., interface usability). The tool is a prototype to be further developed to manage an entire bill of materials, integrating risk management concepts.

Presenting Author: Mikhailo Sartini Università Politecnica delle Marche (UNIVPM)

Presenting Author Biography: Student attending the last year of PhD at the Università Politecnica delle Marche. Main areas Cost Engineering, 3D Printing, LCA, and the complete sphere of Design for.

Authors:

Mikhailo Sartini Università Politecnica delle Marche (UNIVPM)
Luca Manuguerra Università Politecnica delle Marche (UNIVPM)
Giacomo Menchi Università Politecnica delle Marche (UNIVPM)
Marco Mandolini Università Politecnica delle Marche (UNIVPM)
Giulio Marcello Lo Presti Baker Hughes
Francesco Pescatori Baker Hughes