Hybrid Machine Learning Chain for Manufacturing Time Estimation Across Multiple Product Families

In Engineer-To-Order industries, precise estimating of manufacturing time is crucial for optimizing the operations lead time, managing costs and overall efficiency. This study presents a new approach to enhance the reliability of manufacturing time estimations by utilizing a hybrid machine learning chain composed of supervised, unsupervised, and memory-based learning components. This work makes two main contributions: (i) identifying critical features that impact time estimation across product families, followed by an automated Classification system using Neural Networks to categorize product families based on these features and (ii) a structured retrieval process that leverages KD-Tree to improve the efficiency of retrieving relevant historical cases. This new approach was applied on a real-world case in a French metallurgy industry, where it achieved enhanced Mean Absolute Error in time estimates compared to previous methods, underscoring its effectiveness.