Ph. D. Project
Title:
Machine Learning based IEEE 802.15.4-TSCH Scheduling for Low Power Industrial Wireless Networks
Dates:
2020/10/01 - 2023/09/30
Supervisor(s):
Description:
Manufacturing is moving from being digital to be intelligent thanks to the rapid evolution of the Internet of Things. Industry 4.0 or
the Industrial Internet of Things (IIoT) allows to interconnect all the actuators, sensors and controllers in a smart factory. It is
expected to achieve tens of trillions of dollars of the global GDP in the next 20 years.
In industry, the requirements in terms of energy efficiency and quality of service (delay, reliability, determinism and robustness) are
strict and of paramount importance. The IEEE 802.15.4-2006 standard for Low Power Lossy Networks (LLN) meets partially these
requirements. The 2015 IEEE 802.15.4e amendment defines a number of channel access modes (MAC protocols), including Time
Slotted Channel Hopping (TSCH), which received the attention of both academia and industry researchers. The specification of the
access schedule is left to higher protocol layers. Several distributed and centralized scheduling algorithms for TSCH have been
proposed over the past ten years [HGT + 2017].
A scheduling algorithm assigns a set of cells to each active link in the network. A cell is a combination of a timeslot (transmission
time) and a channel offset (frequency). The scheduler has to be carefully designed to meet the strict requirements of a low power
industrial application. This is a challenging task due to the underlying environment, the inherent uncertainty and high complexity as
well as wireless communication unreliability.
Meanwhile, machine learning (ML) returned to the front stage as a suitable solution to address challenging issues from various
domains. Reinforcement learning (RL) [PL+2015] , for instance, has already been adopted to implement a scheduler that adapts to
traffic and topology changes. RL is well suited to a distributed implementation since it requires reasonable computation and memory
resources. However, it suffers from relatively long convergence time that may affect the timeliness of an industrial application.
The aim of this work is to explore some ML techniques to design a link scheduler that satisfies the requirements of an IIoT application
while being energy efficient. A suitable Cloud-Fog-Edge organization can be adopted in order to combine both centralized and
distributed learning algorithms. A deep learning algorithm can be executed at the Cloud while a RL is performed at the edge nodes.
The Fog may host a distributed deep learning [HF+2018] in order to eliminate latency issues since it brings processing power close to
the data source while being at the same time close to the end user.
References
[HGT+2017] Hermeto, R. T., Gallais, A., & Theoleyre, F. (2017). Scheduling for IEEE802. 15.4-TSCH and slow channel hopping MAC in
low power industrial wireless networks: A survey. Computer Communications, 114, 84-105.
[PL+2015] K.-H. Phung, B. Lemmens, M. Goossens, A. Nowe, L. Tran, K. Steenhaut, Schedule based multi-channel communication in
wireless sensor networks: a complete design and performance evaluation, Elsevier Ad Hoc Netw. 26 (2015) 88-102.
[HF+2018] Huang, L., Feng, X., Feng, A., Huang, Y., & Qian, L. P. (2018). Distributed Deep Learning-based Offloading for Mobile Edge
Computing Networks. Mobile Networks and Applications, 1-8.
the Industrial Internet of Things (IIoT) allows to interconnect all the actuators, sensors and controllers in a smart factory. It is
expected to achieve tens of trillions of dollars of the global GDP in the next 20 years.
In industry, the requirements in terms of energy efficiency and quality of service (delay, reliability, determinism and robustness) are
strict and of paramount importance. The IEEE 802.15.4-2006 standard for Low Power Lossy Networks (LLN) meets partially these
requirements. The 2015 IEEE 802.15.4e amendment defines a number of channel access modes (MAC protocols), including Time
Slotted Channel Hopping (TSCH), which received the attention of both academia and industry researchers. The specification of the
access schedule is left to higher protocol layers. Several distributed and centralized scheduling algorithms for TSCH have been
proposed over the past ten years [HGT + 2017].
A scheduling algorithm assigns a set of cells to each active link in the network. A cell is a combination of a timeslot (transmission
time) and a channel offset (frequency). The scheduler has to be carefully designed to meet the strict requirements of a low power
industrial application. This is a challenging task due to the underlying environment, the inherent uncertainty and high complexity as
well as wireless communication unreliability.
Meanwhile, machine learning (ML) returned to the front stage as a suitable solution to address challenging issues from various
domains. Reinforcement learning (RL) [PL+2015] , for instance, has already been adopted to implement a scheduler that adapts to
traffic and topology changes. RL is well suited to a distributed implementation since it requires reasonable computation and memory
resources. However, it suffers from relatively long convergence time that may affect the timeliness of an industrial application.
The aim of this work is to explore some ML techniques to design a link scheduler that satisfies the requirements of an IIoT application
while being energy efficient. A suitable Cloud-Fog-Edge organization can be adopted in order to combine both centralized and
distributed learning algorithms. A deep learning algorithm can be executed at the Cloud while a RL is performed at the edge nodes.
The Fog may host a distributed deep learning [HF+2018] in order to eliminate latency issues since it brings processing power close to
the data source while being at the same time close to the end user.
References
[HGT+2017] Hermeto, R. T., Gallais, A., & Theoleyre, F. (2017). Scheduling for IEEE802. 15.4-TSCH and slow channel hopping MAC in
low power industrial wireless networks: A survey. Computer Communications, 114, 84-105.
[PL+2015] K.-H. Phung, B. Lemmens, M. Goossens, A. Nowe, L. Tran, K. Steenhaut, Schedule based multi-channel communication in
wireless sensor networks: a complete design and performance evaluation, Elsevier Ad Hoc Netw. 26 (2015) 88-102.
[HF+2018] Huang, L., Feng, X., Feng, A., Huang, Y., & Qian, L. P. (2018). Distributed Deep Learning-based Offloading for Mobile Edge
Computing Networks. Mobile Networks and Applications, 1-8.
Keywords:
Machine Learning, Communication Schedule, TSCH, IEEE 802.15.4e, Industry 4.0, Energy Efficiency
Department(s):
| Modeling and Control of Industrial Systems |
