The Industrial Internet of Things (IIoT) is a key infrastructure for monitoring, control, and data-driven decision-making in industrial environments. Its effective design and deployment require balancing latency, energy consumption, reliability, information freshness, security, privacy, and cost. Although research on edge-cloud architectures, digital twins, federated learning, reinforcement learning, and security has expanded considerably, many studies focus on a single architectural layer, a single technology, or a single performance metric. This study aims to systematically review optimization approaches in IIoT architectures and applications, synthesize the available evidence, identify research gaps, and propose a conceptual framework.

A systematic search was conducted in Scopus, Web of Science, IEEE Xplore, ScienceDirect, MDPI, and Google Scholar. After applying the inclusion and exclusion criteria and screening in accordance with PRISMA 2020, 45 studies were selected for review. Data were extracted on architecture type, problem and optimization method, evaluation metrics, security level, proximity to real-world implementation, and reported limitations. The studies were then quality-assessed and analyzed using structured narrative synthesis and framework-based analysis.

The findings identify four architectural patterns (centralized, distributed, software-defined networking-based, and cross-layer) and five families of optimization methods (metaheuristics, multi-criteria decision-making, machine learning and neuro-fuzzy methods, multi-agent reinforcement learning, and federated learning). The results show that jointly integrating security, privacy, information freshness, energy, latency, cost, and practical deployability within a single multi-objective decision remains a significant gap in the literature. Based on the extracted evidence, a conceptual framework is proposed comprising five layers (perception, digital twin, network, processing and analytics, and application and decision-making), together with a cross-cutting security and privacy layer.

By organizing fragmented evidence into an end-to-end structure, this study provides a basis for designing, comparing, and evaluating IIoT solutions. The proposed framework requires empirical validation in real industrial settings