Democritus University of Thrace (DUTH)
Intelligent decision-making and optimization models for complex and dynamic environments utilizing artificial intelligence
Abstract
dc:descriptionThe growing diversification and complexity of home environments enabled by the increasing use of Internet of Things (IoT) technologies and smart home appliances brings its own set of challenges for reliability, energy, and sustainability. Traditional maintenance strategies including corrective, preventive, and even predictive remain insufficient for addressing the constraints of home settings, where appliances often operate under heterogeneous usage patterns, limited sensing capabilities, and scarce labeled data. Within this context, prescriptive maintenance (PsM) emerges as a promising paradigm, offering not only the anticipation of failures but also actionable, context-aware interventions that optimize device performance and minimize energy waste. This thesis presents the development and validation of an holistic framework for prescriptive maintenance in smart homes. The first step taken in the research was to develop a simulation methodology to create synthetic anomalies, meaning allowing for realistic representation of faults on many different appliances, including refrigerators, washing machines, dryers, dishwashers, and water heaters. This tackles the problem of the lack of publicly available datasets with labelled anomalies, providing a strong foundation to build and evaluate anomaly detection models. Then, lightweight and resource-efficient architectures are developed with AE-LSTM and AE-CNN-MAML approaches that have demonstrated able to give accurate anomaly detection on resource-constrained edge devices. These methods have been shown to provide high performance, while maintaining low computational and energy overhead, thus ensuring practical deployment in embedded environments. The thesis also provides advance operational phase detection methods for both constantly-on and program-based appliances, achieving reliable phase segmentation that serves as the basis for contextualized anomaly interpretation and targeted prescriptions. The models utilize dynamic anomaly thresholding, meta-learning, and adaptive retraining methods, which produce excellent resilience to data sparsity, changing fault patterns, and data variability.To consolidate these innovations, two integrated frameworks are proposed. The INFORM framework brings together anomaly detection, operational phase modeling, and prescriptive decision-making into a unified system that transitions from awareness to action. Complementing this, the GUIDE framework introduces behavioral modeling and large language models (LLMs), translating technical prescriptions into accessible, personalized recommendations for households. Together, these frameworks validate the feasibility of real-time, adaptive, and user-centric prescriptive maintenance in residential environments. The results provide evidence that prescriptive maintenance can be realized in a practical context under conditions where data is sparse and use heterogeneous devices. Beyond technical advances, the research demonstrates broader societal and environmental benefits, including reductions in invisible energy waste, extended appliance lifespan, and contributions to sustainability goals. From an academic perspective, the research expands the pool of knowledge relating to anomaly detection, meta-learning, and decision-making for intelligent residential systems. From a practical standpoint, it offers scalable solutions for next-generation smart homes, laying the groundwork for future developments in energy-aware living and sustainable digital ecosystems.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papaioannou, Alexios
- Παπαϊωάννου, Αλέξιος
Subjects
dc:subject × 28- Καθοδηγητική Συντήρηση
- Ανίχνευση ανωμαλιών
- Έξυπνες Κατοικίες
- Διαδίκτυο των Πραγμάτων (IoT)
- Υπολογιστική στην Άκρη (Edge Computing)
- Παραγωγή Συνθετικών Ανωμαλιών
- Ανίχνευση Λειτουργικών Φάσεων
- Μετα-μάθηση
- Ενεργειακή αποδοτικότητα
- Συσκευές με Περιορισμένους Πόρους
- Τεχνητή νοημοσύνη
- Βιώσιμα Οικιακά Συστήματα
- Prescriptive Maintenance
- Anomaly detection
- Smart Homes
- ΙοΤ
- Edge computing
- Synthetic Anomaly Generation
- Operational Phase Detection
- Meta-learning
- Energy efficiency
- Resource-Constrained Devices
- Artificial intelligence
- Sustainable Residential Systems
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Computer and Information Sciences
- Natural Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60704