Machine Learning-Based Prediction of Missing Components for Assembly – a Case Study at an Engineer-to-Order Manufacturer

• Main Authors: Peter Burggraf, Johannes Wagner, Benjamin Heinbach, Fabian Steinberg
• Format: Article
• Language: English
• Published: IEEE 2021-01-01
• Series: IEEE Access
• Subjects: Production control; assembly; prediction methods; lead time reduction; machine learning; supervised learning
• Online Access: https://ieeexplore.ieee.org/document/9416418/

For manufacturing companies, especially for machine and plant manufacturers, the assembly of products in time has an essential impact on meeting delivery dates. Often missing individual components lead to a delayed assembly start, hereinafter referred to as <italic>assembly start delayers</italic>. Identifying the assembly start delayers early in the production process can help to set countermeasures to meet the required delivery dates. In order to achieve this, we set up 24 prediction models on four different levels of detail utilizing different machine learning-algorithms &#x2013; six prediction models on every level of detail &#x2013; and applying a case-based research approach in order to identify the model with the highest model quality. The modeling approach on the four levels of detail is different. The models on the coarsest level of detail predict assembly start delayers utilizing a classification approach. The models on the three finer levels of detail predict assembly start delayers via a regression of different lead times and subsequent postprocessing operations to identify the assembly start delayers. After training of the 24 prediction models based on a real data set of a machine and plant manufacturer and evaluating their model quality, the classification model utilizing a Gradient Boosting classifier showed best results. Thus, performing a binary classification to identify assembly start delayers was the best modelling approach. With the achieved results, our study is a first approach to predict assembly start delayers and gives insights in the performance of different modeling approaches in the area of a production planning and control.