Epsilon Archive for Student Projects

Abstract

The Swedish dairy sector, notwithstanding its robust economic standing, is confronted by enduring inefficiencies, notably excessive food waste, antiquated quality control methodologies, and mounting regulatory demands emanating from evolving EU sustainability directives. This thesis addresses these multifaceted challenges through the conceptualization and implementation of an advanced digital framework that synergistically integrates Artificial Intelligence (AI), Internet of Things (IoT), and Big Data Analytics to augment quality assurance, minimize waste, and advance sustainability imperatives within dairy processing operations.
The principal objective of this research was the development of a smart, scalable decision-support platform, capable of real-time process monitoring, predictive analytics, and comprehensive sustainability tracking across the principal stages of dairy production. The proposed system architecture encompasses a suite of bespoke machine learning models, each tailored to discrete processing tasks, including raw milk classification, demand forecasting, pasteurization optimization, and shelf-life prediction. The methodological approach leverages state-of-the-art techniques such as Temporal Fusion Transformers (TFT), Long Short-Term Memory (LSTM) networks, LightGBM, and Reinforcement Learning (RL) agents, selected for their demonstrated efficacy, interpretability, and energy efficiency. The platform was operationalized via Power BI dashboards and a SQL-integrated cloud infrastructure, thereby facilitating real-time data visualization and actionable decision support.
Empirical evaluation of the deployed models yielded compelling results: the milk quality classifier attained a perfect accuracy rate (100%), the shelf-life predictor achieved an accuracy of 98.3%, and the RL-driven pasteurization controller surpassed conventional PID systems in both safety assurance and energy optimization. Furthermore, each model’s deployment was systematically aligned with sustainability outcomes as delineated by the United Nations Sustainable Development Goals (SDGs), specifically targeting reductions in food waste, enhancements in energy efficiency, and improvements in food safety protocols.
This thesis thus makes twofold contributions: firstly, by delivering a functional, intelligent dairy management application; and secondly, by providing a replicable digital transformation blueprint for the broader food manufacturing sector. Recognized limitations of the current work include reliance on simulated sensor data and the absence of blockchain-enabled traceability mechanisms. Future research directions are identified, encompassing live pilot deployments, the integration of generative AI-powered diagnostics, and comprehensive alignment with the forthcoming EU Digital Product Passport framework. Collectively, the findings substantiate the transformative potential of AI-driven frameworks in rendering dairy processing systems more resilient, transparent, and sustainable.