Demandflex-Lstm: A Long Short-Term Memory Model for Forecasting Adaptive Material Requirements in Lean Manufacturing

Abstract

In lean manufacturing systems, not estimating materials based on demand can lead to inadequate production flow,
increased inventory, and even waste in processes. In this paper, we present a new model named DemandFlex-LSTM.
The model is based on long short-term memory (LSTM) neural networks, capable of dynamically adjusting to the
ever-changing material demands of all lean manufacturing systems. Demand forecasting is significantly improved in
DemandFlex-LSTM compared to classical techniques since it captures dependencies faster and learns from historical
production and consumption data. The model dynamically adapts to demand changes, providing the ability to
proactively update material purchasing and inventory management. Comparing DemandFlex-LSTM against ARIMA,
it was proven that our model performed considerably better in accuracy, responsiveness, and overall forecasting
capability. Implementing the proposed solution, delivery and supply costs can be mitigated, enabling lean operations
to follow the just-in-time (JIT) concept. This also allows manufacturers to be faster and more flexible in their supply
chains. Leveraging this sophisticated model paired with modern demand and production systems will empower
operators with precise data to respond to complex variations.