Optimizing access point placement in industrial IoT : a deep reinforcement learning approach, with verification using Q-learning and visualisation using signal heat map | |
| Author | Mahanta, Manash |
| Call Number | AIT Diss. no.IOT-25-01 |
| Subject(s) | Internet of things--Industrial applications Deep learning (Machine learning) |
| Note | A Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Engineering in Internet of Things Systems Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | Industry 4.0 has brought about a significant transformation in the design, manufacture, and distribution of products. Technologies such as artificial intelligence (AI), cloud connectivity, and industrial internet of things (IIoT) have become an integral part of the manufacturing process. Optimal access point (AP) placement inside an industrial layout is important to ensure excellent connectivity across the industry floor, which in turn ensures that industries can make use of these latest technologies. However, wireless fidelity (Wi-Fi) AP placement in an industrial layout is complicated because industries exhibit distinct radio propagation characteristics and channel models compared to typical indoor offices and hotspots, due to being larger in size, containing numerous metal machine tools, and the presence of large barriers for storing finished products or raw materials.This study proposes a novel method to optimise Wi-Fi access point placement in industrial environments using DRL and Q-learning, with circle packing serving as a baseline for com parison. The approach addresses the unique challenges of radio propagation in such settings. The performance metrics used are average received signal strength (RSS), average interfer ence, and signal coverage across the industrial floor. These performance metrics are used to tailor the reward function used in DRL and Q-learning.The proposed approach demonstrated promising results, including an increase in average RSS by3.55%andadecrease in average interference by 5.73% in one instance, and a 4.59% increase in signal coverage in another instance, compared to the results obtained using circle packing. DRL performed the best, followed by Q-learning and circle packing. Q-learning, though, matched the performance of DRL when the number of APs was low. Although the exploration-prioritised epsilon decay yielded better placement results, it, for the most part, required longer training times than its exploitation-prioritised counterpart, highlighting a trade-off between performance and computational efficiency. The proposed technique is effective for both small and large scaled industrial areas. However, the learning process takes significantly longer to complete in larger industrial areas.The optimised AP locations and their corresponding RSS distributions are visualised using Wi-Fi heat maps and compared with those obtained through circle packing. The improve ment in signal coverage is evident from the comparison, indicating that DRL-based place ment strategies can substantially enhance wireless coverage, which is critical for real-time industrial applications in smart factories. The technique presented in this work can be used during the initial network setup in an industry or to improve the existing network. Future work can explore adaptive DRL architectures to reduce training time, especially in larger environments. |
| Year | 2025 |
| Type | Dissertation |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Internet of Things (IoT) Systems Engineering |
| Chairperson(s) | Attaphongse Taparugssanagorn; |
| Examination Committee(s) | Poompat Saengudomlert;Teerapat Sanguankotchakorn; |
| Scholarship Donor(s) | AIT Fellowship;Osotspa Scholarship; |
| Degree | Thesis (Ph. D.) - Asian Institute of Technology, 2025 |