Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article


Doctor of Philosophy


Electrical and Computer Engineering


Dr. A. Shami

2nd Supervisor

Dr. A. Refaey


Manhattan College



This work explores reduced complexity solutions for the increased efficiency and safety of the industrial Internet of Things (IIoT) sensing domain. Resource virtualization, security, and predictive modeling are the main subjects of these studies.

The first solution is a joint throughput and time-resource allocation scheme for virtualization of IEEE 802.15.4-based wireless sensor networks. Virtualization is realized through the utilization of the guaranteed time slot mechanism for scheduling on the medium access control (MAC) layer. The solution abstracts resources into logical units that are allocated to segregated applications with different service requirements. The problem is formulated in a linear optimization framework and solved with a heuristic fair resource allocation (FRA) algorithm. The proposed scheduling approach provides fast and efficient resource management for low-power networks.

The second solution performs a reduced complexity symmetric group rekeying in low-power wireless networks. The novel pseudorandom key chaining (PRKC) scheme uses pseudorandom sequences generated at lower layers of the communications stack to enable synchronous refresh of encryption keys in network nodes during broadcasting. The suitability of generated keys for cryptographic applications is validated using the National Institute of Standards and Technology Special Publication 800-22 (NIST SP 800-22) statistical suit. When implemented on a Raspberry Pi board, the PRKC algorithm runs faster and requires smaller CPU effort to refresh keys than the reference schemes.

Our final contribution is a concept drift-aware solution for adaptive modeling of multivariate time series in nonstationary environments of the IIoT sensing domain. In the proposed three-layered three-state (TriLS) system, the gateway and the cloud collaborate to accurately model industrial process trends towards intelligent factory automation. In this scheme, all computationally demanding functionality of a model building and concept drift detection is shifted to the cloud side. The gateway uses a trained model for prediction in real-time and fine-tunes it to changing data. When tested on synthetic and real datasets, the TriLS system demonstrates a better prediction quality in nonstationary conditions than the conventional approaches. It also requires a reduced computational effort on the gateway side and smaller communications overhead for adjusting the model to drifting concept.

Summary for Lay Audience

Industrial Internet of Things (IIoT) is an ecosystem of interconnected objects that collect and exchange data about their physical surroundings in an industrial setting. This thesis focuses on networks of communications-enabled low-power sensing nodes that are at the core of the IIoT sensing domain. Our work contributes to three important areas of the IIoT sensing domain: network sharing (virtualization), security, and adaptive learning. We explore solutions that improve the efficiency and prolong the lifetime of low-power wireless networks. At the same time, we focus on reducing the complexity of the proposed schemes to accommodate the practical limitations of sensing nodes (memory, energy, and computational power).

The first part of this thesis provides a resource management solution that enables efficient sharing of the same network by multiple applications. The problem is formulated as optimization and is solved with a heuristic algorithm. This approach virtualizes network resources according to different applications’ requirements while satisfying communications latency constraints.

The second part presents a lightweight scheme for synchronous updating of group encryption keys in a network to protect the confidentiality of broadcast messages. This solution utilizes pseudorandom sequences generated during communications to enable the evolution of keys in time. The suitability of generated keys for cryptographic applications is extensively tested. In addition, when implemented on an IoT device, the proposed rekeying algorithm runs faster and requires smaller computational effort than the reference algorithms.

The final part of this thesis describes a solution for adaptive modeling of industrial process trends in changing IIoT environments. The proposed system facilitates efficient collaboration between the gateway and the cloud towards timely adaptation of a predictive model. All computationally expensive operations (such as model training and change detection) are performed in the cloud. The gateway uses ready models for prediction and fine-tunes them locally. Thoroughly tested on synthetic and real datasets, the proposed system improves prediction quality and reduces the computational load when compared to reference schemes. It also reduces the communications overhead between the gateway and the cloud for adaptation.

This work provides novel theoretical contributions and practical solutions for more efficient, secure, and intelligent IIoT sensor networks.

Creative Commons License

Creative Commons Attribution-Share Alike 4.0 License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.