**5. Conclusions**

In this paper, we propose an edge-fog-cloud architecture for autonomous industrial monitoring of thermal anomalies. At the edge layer, robot surveyors carrying IoT nodes scan the environment's temperature in real-time for thermal anomalies in the absence of communication networks to report alerts. The robots use our proposed deep network running at 65 frames per second with a root mean squared error of 0.19, 0.02 less than the deeper VGG-19 network, and three times faster. If anomalies are detected, a mobile fog node in the form of a companion drone is dispatched looking for connectivity to deliver an early alert in 6.7 s on average. When a run is complete, the robot uploads thermal and visual videos and location information to a cloud back-end server. We use a proposed thermal to visual registration algorithm to maximize mutual derivative information and spatio-temporally align and localize thermal anomalies. End-users receive detailed reports, including the aligned video with thermal anomalies localized in 16.03 min, far sooner than the human survey time. We tested our proposed architecture in both the lab and onsite and concluded it efficiently monitors a sizeable industrial area despite its challenging characteristics. Our system's limitations are survey path segments with low lighting conditions affecting self-driving and the lack of connectivity in the environment, which we have addressed by shifting autonomous driving to rely on line-following in these segments and adding a companion drone to deliver alerts in a short time. Further research is needed to address these limitations better.

**Author Contributions:** Funding acquisition, M.M.; Investigation, M.G., T.B., M.Y., M.A., and A.S.E.-B.; methodology, M.G., T.B., M.Y., M.A., and A.S.E.-B.; project administration, M.G.; software, M.G., A.S.E.-B., T.B., and M.Y.; supervision, M.G., M.A., and A.S.E.-B.; Validation, M.G., T.B., M.Y., M.A., and A.S.E.-B.; writing—original draft, T.B., M.Y., and M.G.; writing—review and editing, M.G., M.A., A.S.E.-B., and M.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Emirates Global Aluminium, UAE.

**Acknowledgments:** This work is supported by Emirates Global Aluminum. Moreover, the authors thank Abdalla Rashed and Razib Sarker for roles in the implementation and testing.

**Conflicts of Interest:** The authors declare no conflict of interest.
