A decision support system for sequencing production in the manufacturing industry

In the context of labour shortages and an aging population, it is important to support knowledge transfer from experienced workers. Sometimes, this can be done through optimization models, but it is not always possible to get and/or have explicit rules and constraints that could impact a decision when making complex decisions, such as in job sequencing. In such situations, a gap can occur between what the decision models suggest and what experienced planners will actually do. This is because workers may take into account a range of information and knowledge that has been gained from previous decisions that cannot be included in the decision models. The objective of this paper is to address this point by providing a decision support tool based on learning, without producing an optimization model, that is capable of replicating the production sequencing decisions of an experienced planner in a dynamic and complex production context.A methodology is proposed based on a Seq2Seq-LSTM model using the Teacher Forcing method with the Beam Search algorithm. Results are refined by a statistical and constrained method. The N scenarios with the best global scores are proposed as production sequencing recommendations. Promising results obtained in an industrial case study are presented. The best-performing learning model using production and demand data achieves a prediction rate of 52.43% for the prediction of a sequence of three products when one option is considered, and almost 70% when 10 options are considered.