In a production environment increasingly focused on efficiency and waste reduction, proactive inventory management remains a crucial challenge for companies. This thesis proposes an artificial intelligence-based approach to anticipate such events, using a recurrent neural network model trained on real data collected through the KanbanBOX management software. Following a careful phase of data extraction, cleaning, and structuring, a dataset was built to be consistent with the operational logic of the Kanban system. The developed model is able to estimate the short-term risk of stockout, providing concrete support for inventory management decisions. The solution was integrated into the application context using TensorFlow Serving, enabling real-time predictions and supporting the operational decision process. The proposed approach has proven effective in enhancing the reliability and responsiveness of the production system, paving the way for future developments aimed at a more comprehensive modeling of temporal and strategic dynamics.
In un contesto produttivo sempre più orientato all’efficienza e alla riduzione degli sprechi, la gestione proattiva delle scorte costituisce una sfida cruciale per le aziende. Questa tesi propone un approccio basato sull’intelligenza artificiale per anticipare tali eventi, attraverso un modello neurale ricorrente addestrato su dati reali provenienti dal software gestionale KanbanBOX. Dopo un’accurata fase di estrazione, pulizia e strutturazione dei dati, è stato costruito un dataset coerente con le logiche operative del sistema Kanban. Il modello sviluppato è in grado di stimare la probabilità di esaurimento delle giacenze nel breve termine, fornendo un supporto concreto alle decisioni sulla gestione delle scorte. La soluzione è stata integrata nel contesto applicativo tramite TensorFlow Serving, consentendo di generare previsioni in tempo reale e supportare il processo decisionale. L’approccio proposto si è dimostrato efficace nel rafforzare l’affidabilità e la reattività del sistema produttivo, aprendo la strada a futuri sviluppi orientati a una modellazione più completa delle dinamiche temporali e decisionali.
Modello neurale per prevedere l’esaurimento delle giacenze in un sistema Kanban
CAMPAGNARO, MASSIMO
2024/2025
Abstract
In a production environment increasingly focused on efficiency and waste reduction, proactive inventory management remains a crucial challenge for companies. This thesis proposes an artificial intelligence-based approach to anticipate such events, using a recurrent neural network model trained on real data collected through the KanbanBOX management software. Following a careful phase of data extraction, cleaning, and structuring, a dataset was built to be consistent with the operational logic of the Kanban system. The developed model is able to estimate the short-term risk of stockout, providing concrete support for inventory management decisions. The solution was integrated into the application context using TensorFlow Serving, enabling real-time predictions and supporting the operational decision process. The proposed approach has proven effective in enhancing the reliability and responsiveness of the production system, paving the way for future developments aimed at a more comprehensive modeling of temporal and strategic dynamics.| File | Dimensione | Formato | |
|---|---|---|---|
|
Campagnaro_Massimo.pdf
accesso aperto
Dimensione
652.68 kB
Formato
Adobe PDF
|
652.68 kB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/89339