Author: Omar Nasir
Edge devices exist at the end of the technology chain where they directly interact with users. Examples are mobile devices and WiFi routers in telecommunication networks. In the context of lighting control, edge devices represent luminaires and sensors that power the ubiquitous lighting of our modern world. With the advances in computing efficiency, these components have become increasingly clever; their responsibilities have evolved from switching lights on/off to providing optimised lighting that learns user behaviours, mimics natural sunlight patterns and automatically conserves energy by detecting ambient light levels.
Many of the examples quoted above implement the concept of ‘Intelligence on the edge’. It is a paradigm that further enables intelligence on edge devices, albeit locally. Edge devices can act as a cost-effective conduit for developing smarter applications that enhance user experience and provide quality lighting environments. The benefits are clear: low-cost chips, the pervasiveness of the devices and direct impact on the user. Furthermore, there may arise situations where implementing decision-making at a global level can also improve user experience.
Therefore, it is imperative that companies evaluate the optimal approach when designing AI-based services and decide if the technology is to be implemented in the cloud, at the edge or as a hybrid fusion of both.
By design, local intelligence has a limitation that each device exists in its own learning sphere. If the objective is to dim down light levels because the luminaire is next to a windowpane with abundant ambient light, it could simultaneously impact user experience if there are multiple luminaires in the same space with varying dimming levels which produce uneven lighting. Smart lighting should adapt to its surroundings, rather than force the environment to change to accommodate its limitations. Therefore, it makes sense to envision intelligence globally, where the devices communicate and optimise their behaviour based on a collective objective seeking to both minimise energy consumption and user impact. A possible outcome could be a lesser reduction in the dimming level of the luminaire, which, at the same time, maintains uniformity of lighting and reduces the adverse impact on user wellbeing.
With Helvar’s AA Generation 2.0, it is possible to achieve the same result by implementing intelligence locally.
Every device has a lifetime value. For luminaires, it is defined as the number of expended burn-hours. The LED driver is responsible for ensuring a consistent supply of electricity to power the LED, however, with time, the wear and tear of the device will increase the electrical demand in order to produce the same amount of lux. This is a special area of application where a hybrid model suffices. The number of burn hours can vary between different kinds of spaces; a room in a remote part of the building will yield lesser burn-hours compared to a busy entrance lobby. Hence a holistic comparison of the decreasing efficiency over time of all the luminaires in the building can lead to incorrect attribution of faulty devices. An AI-driven maintenance model can analyse the lifetime efficiency of every component compared to its neighbours and produce more accurate insights into deteriorating device mechanics. A predictive model allows for improved monitoring and maintenance where action can be undertaken proactively. Replacing devices that are faulty or operating at the limits will boost energy conservation and minimize the impact on user experience as well.
Real-world problems are often solved by complex models which, more often than not, require the deployment of hybrid solutions.
For example, a predictive model that controls lighting environment can have a local component for each device that learns specialized patterns of occupancy, and an inter-communicable global component which leverages information from local components of nearby devices to improve prediction accuracy, since occupancy patterns do not exist in isolation. It is pertinent to consider possible trade-offs such as cost, efficiency, complexity, device capability and cloud support systems when creating AI services IoT devices, and this will become more apparent with the gradual advances in smart building systems. With the increasing focus on wellbeing and energy efficiency, the lighting industry faces similar technical challenges and it is imperative to avoid the one-size-fits-all fallacy.
Omar Nasir works as a Data Scientist at Helvar. His primary area of research is the application of Machine Learning to IoT sensor data in the context of lighting control. His previous diverse roles include Data analyst/Optimization engineer and Full-Stack Developer in the Telecommunications Industry. He has briefly dabbled with entrepreneurship as CTO of a startup and holds dual master’s degrees from Royal Institute of Technology (KTH), Sweden and Aalto University, Finland.