Evaluating the performance of desk sensors for closed-loop daylight harvesting in an office building

Increased daylight entry into office buildings generally does not necessarily save electric lighting energy. Electric lights may remain switched on, even in spaces that are well-lit by daylight or are unoccupied. Energy savings can be obtained when sensors are installed to turn off or dim electric lights. Daylight harvesting can save electric lighting energy by dimming lights in response to increased daylight admittance. The main two modes of daylight harvesting control are open-loop and closed-loop (Figure 1).

Figure 1 Diagram that shows two daylight harvesting control systems used to save energy: open-loop control (above) and closed-loop control (bottom)

Open-loop control detects changes in daylight only by using a sensor that is placed on or nearby the window. This system sends input signals directly to the electric light without feedback from the illuminated desk surface. Hence, open-loop control may not guarantee that workstations are always illuminated to the intended design setpoint (i.e., 300 lx or 500 lx). This can be overcome by placing sensors near workstations (e.g., on the ceiling or walls), so that they measure light from both electric and daylit sources. Feedback signals from the workstation can then be gathered by the sensor and sent to the light controller, creating a closed-loop system. Where the sensor is located is important to ensure that the lighting system performs optimally to maximize savings on electric lighting energy.

Desks sensors work best when placed at the locale designers aim to provide sufficient illumination, enabling them to gather information about the light directly received at the desk surface. However, desk sensors risk being blocked by other clerical items (e.g., cups, papers, personal objects, etc.) also stored on the same surface. These occupant behaviors are normal, and enforcing workstation policies to prevent sensor blocking does not overcome these issues and may even interfere with occupant office work performance. Nevertheless, sensor blocking hampers the system’s ability to save energy, and may even cause more energy compared to a conventional lighting system that is unable to harvest daylight. This may be why manufacturers and design guidelines specify the ceiling or walls as locations for closed-loop daylight harvesting sensors. However, research studies mostly place sensors on desk surfaces.

What did we learn from using desk sensors? We measured desk sensor performance for closed-loop daylight harvesting in an office in Singapore. Nearly half a million data points were collected from 39 workstations for 1-month, measuring sensor lighting (illuminance) and the electric ceiling light’s power output. Each workstation had a corner-mounted desk sensor that operated dedicated electric ceiling light. We found the following when desk sensors were used for daylight harvesting:

• 68% of the time, sensors performed as intended (i.e., were unblocked)
• 24% more electric energy was used compared to ideal operating conditions
• Sensor blocking was exacerbated by the lack of storage space at the workstation

Figure 2 Comparison between blocked and unblocked sensors

?Our study showed that desk sensors do not work well during actual building operation. Blocked sensors not only drew more electric energy from the system (Figure 2) to maintain the design setpoint required at the workstation, but they also sent incorrect feedback signals to the light controller. Because the desk sensor is covered by an office item, it is unable to measure the actual light received at the surface. This caused workstations to be over-illuminated.

How did we solve this problem? To detect blocked or unblocked sensors, we built a machine-learning model using the relationship between the electric ceiling light’s power output and the light (illuminance) reading from an unblocked desk sensor (Figure 3). This relationship is used to calibrate the lighting system. Although this calibration is mostly used to ensure sufficient light is provided to workstations, it usually has no further application for daylight harvesting. In our work, we used the system’s calibration to create a sensor blocking algorithm.

Figure 3 Workflow used to classify blocked and unblocked desk sensors, build and train a machine-learning model, and apply our sensor blocking algorithm to a dataset collected from an office that used closed-loop daylight harvesting.

?Blocked sensors could be easily detected when the electric ceiling light’s power output was disproportionately higher than the light measured at the desk sensor. This allowed the sensor blocking algorithm to accurately (99%) detect sensors that were either blocked or unblocked. Accuracy was verified by training and testing a machine learning (support vector machine) model on nearly half a million data points, measuring illuminance and power output from every desk sensor and electric ceiling light from the 39 workstations. By integrating our sensor blocking algorithm into the lighting system’s control interface, performance metrics can be conveyed to the facility management. This will show how many and how often sensors are being blocked and what implications this will have on its energy performance.

Our recommendations: We do not recommend the use of desk sensors for closed-loop daylight harvesting. This recommendation also applies to scaled models and computer simulations since touted energy savings would not reflect in real conditions. Sensors should be located near the desk, on surfaces (e.g., workstation partition walls) that have a low risk of being blocked yet still produce measurements that are representative of the light found at the desk surface. Alternatively, open-loop control could be considered when the installation of sensors for closed-loop control cannot avoid issues that may hamper the system’s performance.

To optimize energy performance for daylight harvesting applications, our sensor blocking algorithm can be applied to identify a wide range of sensor performance issues, irrespective of both sensor location and problems only caused by blocking. The application of our algorithm is not restricted to sensors mounted on desk surfaces (e.g., ceiling sensors), and can diagnose other important issues that may impede sensor performance (e.g. recalibration or maintenance), maximizing the amount of electric light energy that is saved. We will be happy to work with companies interested in implementing this approach in their own products.

Acknowledgments. This research is funded by the Republic of Singapore’s National Research Foundation through a grant to the Berkeley Education Alliance for Research in Singapore (BEARS) for the Singapore-Berkeley Building Efficiency and Sustainability in the Tropics (SinBerBEST) Program. BEARS has been established by the University of California, Berkeley as a centre for intellectual excellence in research and education in Singapore.

?References. Kent, Huynh, Schiavon, Selkowitz, 2022. Using support vector machine to detect desk illuminance sensor blockage for closed-loop daylight harvesting. Energy and Buildings. https://doi.org/10.1016/j.enbuild.2022.112443 (free pdf download)