AI4EIoT 2026 Abstracts

Area 1 - Core AI–IoT Integration

Full Papers

Paper Nr:	5
Title:	A Robust Computational Pipeline for the Rapid Generation of Structural Digital Twins for Seismic Vulnerability Mapping in Complex Urban Environments
Authors:	Nima Dolatabadi, Renato Zona, Lelio Campanile, Fiammetta Marulli and Vincenzo Minutolo
Abstract:	The rapid expansion of urban environments has increased building vulnerability to seismic hazards. Traditional structural evaluations lack the scalability required for city-wide assessment, particularly for plan-irregular structures prone to torsional coupling. Model an accurate digital twin and perform structural analysis, is complex and time consuming. To resolve this, we introduce an automated computational pipeline integrating multimodal Large Language Models (LLMs) with procedural mesh generation. Our framework extracts deterministic architectural metrics from remote street-level imagery and systematically translates them into solver-ready, 3D finite element wireframes via a custom Python-based generator. By autonomously producing fully connected analytical models, this IoT-driven edge-to-cloud methodology bypasses traditional labor costs, enabling the massive scalability required for modern urban infrastructure safety analytics.

Paper Nr:	6
Title:	An AI-Driven Methodology for Intellectual Asset Evaluation in Industry 4.0 and IoT-Driven Innovation Ecosystems
Authors:	Renato Zona, Antonio Perfetti, Franco Rosatelli and Lelio Campanile
Abstract:	The rapid growth of innovation in IoT-driven and Industry~4.0 environments has generated a vast landscape of intellectual assets that extends well beyond formal patent filings, encompassing proprietary know-how, internal research outputs, and early-stage technological concepts. Traditional patent evaluation methodologies, relying on keyword retrieval, citation metrics, and expert-driven analysis, are increasingly inadequate for capturing the semantic complexity of such heterogeneous assets and for supporting data-driven investment and technology transfer decisions. This paper presents Evalentia, an extended AI-driven framework for the unified evaluation of both patented and non-patented knowledge assets. Building on a previously validated methodology for IoT patent analysis, the proposed approach integrates Large Language Model (LLM)-based semantic representation using LLaMA 70B, sentence-level embeddings via Sentence-BERT, and a multi-dimensional scoring model structured around seven interpretable indicators: Feasibility, Multipurpose, Obsolescence, Geo-context, Protection, Technological Interest, and Technological Resonance. The framework further incorporates semantic similarity analysis, best-fit comparable ranking, and graph-based ecosystem mapping to support competitive positioning and strategic decision-making. A synthetic case study in the advanced manufacturing domain demonstrates the coherence and applicability of the proposed pipeline across heterogeneous knowledge assets.

Paper Nr:	7
Title:	Deception through Synthesis: Using AI-Generated Data to Enhance IoT Security
Authors:	David Weissman and Anura P. Jayasumana
Abstract:	The widespread growth of the Internet of Things (IoT) has introduced significant cybersecurity challenges, necessitating effective defense strategies. Preparing to differentiate normal from abnormal IoT network traffic can be achieved using re-alistic training datasets to improve the detection of adversarial attacks. This paper presents a new approach that uses Artificial Intelligence (AI) to generate synthetic IoT network traffic that resembles real devices. By leveraging Large Language Models (LLMs), we enhance Machine Learning (ML) and neural network-based Conditional Tabular Generative Adversarial Networks (CTGAN) algorithms, helping security researchers produce a mix of original and synthetic network flow patterns for deception-based security. Our work uses AI and ML to improve CTGAN's ability to handle both continuous and categorical features simultane-ously, which is critical for accurately modeling IoT traffic characteristics. We evaluate our approach on two datasets: a live home IoT environment from the publicly available Colorado State University IoT FlexData and CIC (Canadian Institute for Cybersecurity) IoT 2023 benchmark datasets. For each dataset, syn-thetic data is generated at 5x, 10x, and 100x replication levels, focusing on device traffic flow patterns. Fidelity is assessed using established ML metrics to ensure the synthetic data maintain statistical and behavioral similarity to the original data. One of our key contributions is customizing the open-source CTGAN implemen-tation with AI instructions to include realistic temporal dynamics and reconstruct flow identifiers compatible with the CIC FlowMeter feature extraction frame-work. This ensures the synthesized traffic can be easily integrated into existing security workflows. Our preprocessing pipeline is also enhanced with AI, sys-tematically removing data errors and missing entries, normalizing values, detect-ing outliers, and reconstructing network flows according to the original feature specifications. The main goal of this research is to show how high-quality syn-thetic IoT traffic can strengthen security through deception techniques, reduce the attack surface of known devices, and give security practitioners scalable, privacy-preserving datasets for training and testing detection systems without exposing sensitive operational networks.

Area 2 - Cross-Domain and Applied Intelligence

Short Papers

Paper Nr:	8
Title:	AI for Regulatory Intelligence in Civil Engineering: A Retrieval-Augmented Approach to Compliance and Incentive Optimization
Authors:	Lelio Campanile, Sabatino Carbone, Fiammetta Marulli and Renato Zona
Abstract:	Artificial Intelligence (AI) is rapidly transforming civil and building engineering by enabling data-driven design, predictive maintenance, and automation of complex workflows. Despite significant progress, several challenges remain, particularly in regulatory compliance, integration with legacy systems, and domain-specific reasoning. This paper reviews recent AI applications in the field, identifies open research challenges, and proposes a domain-specific Retrieval-Augmented Generation (RAG) framework built on a compact Large Language Model (LLM), aimed at supporting professionals in navigating building regulations and fiscal incentives. The proposed system dynamically integrates official regulatory sources to enhance decision-making and streamline administrative processes. Finally, AI can enhance civil engineering workflows, particularly in regulatory domains. The proposed RAG framework supports compliance and decision-making.