Applying machine learning to predict production capacity for engineer-to-order products: Learning from wind turbine industry

To shorten lead times, engineer-to-order (ETO) companies develop production capacity plans before finalising product designs. However, the production capacity planning of ETO products tends to be highly unpredictable due to factors such as changes in customer requirements, leading to discrepancies between actual demand and planned capacity. Despite the enormous body of literature on capacity planning, there is a lack of research in the context of ETO. Especially, the literature on the use of data-driven methods, such as machine learning (ML) for production capacity prediction, is sparse. Recognising this potential, this study focuses on early production capacity prediction for ETO products using ML and aims to improve the accuracy of production capacity planning. In this paper, design science research is employed in a real company to develop a ML implementation framework. We find that the stacking model outperforms other three models, demonstrating the feasibility of using ML methods to predict production capacity early in dynamic environments. The developed artefact demonstrates a method for employing ML to predict production capacity for ETO products within a real-world problem domain. Furthermore, the challenges encountered during the ML implementation are discussed based on the proposed artefact, and corresponding suggestions are provided for practitioners. • We follow a design science research method to develop a ML solution for predicting production capacity in ETO environments. • An early cost forecasting framework for engineer to order products using machine learning is proposed. • ML improves the accuracy of production capacity prediction despite the limited information resulting from design uncertainty. • We discuss the challenges related to ML implementation, focusing on the importance of business understanding.