The Industrial Internet of Things (IIoT) connects vast networks of sensors, machines, and control systems to enable real-time data collection, analysis, and decision-making in industrial settings. This connectivity drives improved operational efficiency, reduced downtime, and enhanced productivity in sectors such as manufacturing, mining, and energy. Traditional architectures, however, rely heavily on cloud computing, which struggles with latency requirements and bandwidth constraints. Distributed edge-to-cloud computing offers a promising alternative by bringing computational resources closer to data sources, thereby enabling faster responses while reducing network demands.

Remote industrial environments present unique difficulties. Network connectivity in underground mines, offshore platforms, and isolated facilities is often intermittent or unavailable. Cloud-centric systems fail under such conditions, leaving critical monitoring gaps that threaten safety and operational continuity. Edge computing addresses these challenges by processing data locally, but questions remain about how existing IoT platforms support key IIoT requirements that edge-to-cloud architectures must satisfy and how systems can maintain reliable operation during connectivity disruptions.

This thesis investigates distributed edge-to-cloud computing for IIoT applications, with particular attention to remote and connectivity-constrained environments. It presents three contributions. First, a comprehensive survey identifies key IIoT requirements from the distributed edge-to-cloud computing perspective and provides a comparative analysis of three prominent IoT platforms. It reveals how each platform addresses the identified key IIoT requirements. Based on this analysis, open challenges and research opportunities are identified for advancing edge-to-cloud systems in industrial contexts.

Second, an experimental study implements and evaluates a fully edge-based data analytics solution for remote industrial environments using an open source platform. The solution combines real-time monitoring, alarm management, and sensor health analysis without dependence on cloud connectivity. Experiments conducted in an underground mining setting with LiDAR perception systems demonstrate low-latency performance and reliable operation under realistic conditions. Performance evaluation in terms of throughput, latency, and failure rate provides practical insights for future deployments in similar settings.

Third, a distributed edge-to-cloud AI architecture is proposed for maintaining prediction capabilities during intermittent connectivity. The architecture deploys machine learning models at both edge and cloud tiers, with a store-and-forward messaging layer that ensures zero data loss during network outages. A case study on methane hazard prediction in underground coal mining validates the approach using over nine million sensor readings from an operational mine. Experiments across three network scenarios demonstrate continuous edge prediction, stable inference latency, and complete data preservation regardless of connectivity conditions.

Together, these contributions advance both understanding and practical application of distributed edge-to-cloud computing in industrial settings. The findings demonstrate that edge-based solutions can operate independently when cloud connectivity fails, that open source platforms provide viable foundations for industrial deployments, and that distributed edge-to-cloud AI architectures can deliver resilient monitoring for safety-critical applications.