PANDORA D6.1: Instantiation of PANDORA pipelines in industrial settings

This deliverable (D6.1 - Instantiation of PANDORA pipelines in industrial settings) reports the work conducted in Tasks 6.1, 6.2, and 6.3 during the first reporting period (M09-M18) of the PANDORA project within WP6 (Real-life experimentation to demonstrate scientific progress towards adopting trustworthy and energy-efficient AI systems). WP6 translates PANDORA'sscientific advances into operational reality by instantiating AI-driven data pipelines across five heterogeneous industrial sectors. The deliverable addresses three interconnected objectives: Task 6.1 establishes the regulatory and organizational foundation by aligning business and validation scenarios across all five industrial pilots (FRA, SAG, HALCOR, PCL, and Ericsson) with PANDORA experimental protocols while ensuring compliance with GDPR and sector-specific data protection regulations. This task defines data acquisition mechanisms, execution plans, andstakeholder engagement strategies that enable systematic experimentation. Task 6.2 creates detailed architectural blueprints that instantiate PANDORA data pipelines for each pilot use case. Through an innovative integrated approach, Tasks 6.1 and 6.2 have been unified into a comprehensive template structure that consolidates regulatory compliance specifications with technical pipeline architectures. Five complete pipeline blueprints have beendeveloped, each defining data flows from acquisition through preprocessing, mining, discovering, modeling, and filtering, tailored to specific business sectors. These blueprints integrate core PANDORA components from Phase 1 (SDG, DACL, DEL, DRFE) and Phase 2 (FRLM, ADOA, CRV, DICE, CIM) with AIaaS, CaaS, and IM services to form modular, interoperable pipelines. Task 6.3, led by NTUA, advanced ahead of schedule to establish verification and validation frameworks that systematically link KPIs to PANDORA strategies, components, and use case scenarios. The KPI mapping process ensures that both Phase 1 (trustworthy dataset generation) and Phase 2 (autonomous, energy-efficient AIoT operations) can be evaluated against definedbenchmarks using appropriate metrics (Dunn's, Silhouette, Davies-Bouldin indexes for data quality, FL benchmarks like LEAF for federated learning, inference acceleration metrics for computing efficiency). The integrated template approach represents a key innovation, enabling consistent documentation across heterogeneous industrial settings while maintaining the flexibility required for sector-specific adaptations. This modular framework supports the preparatory work for Task 6.4 (operational deployment and experimentation) and Task 6.5 (impact assessment), establishing the foundation for demonstrating PANDORA's applicability across multiple industrial sectors and validating its contribution to trustworthy, energy-efficient AI deployment in real-world AIoT systems.