scroll

City Heat Sinks with Machine Learning

The challenge

As cities welcome more people, and as temperatures trend upward, we increasingly find ourselves designing for indoor and outdoor environments. But working with nature is like working with children: you just never know what you’ll get. Glare and wind change rapidly, and temperature fluctuates way outside the conditioned 18-24ºC of office buildings. How can we design for peoples’ comfort in these ever-changing environments?

In this article

  • What outdoor thermal comfort means to engineers, and how to design for and measure it
  • How to use sensors, data and machine learning to create designs that change with the environment
  • An introduction to database infrastructure and modelling software: particle.io, EnergyPlus and Rhino/Grasshopper

Building physics engineers like Lauren Boysen and Alex Smith are increasingly asked to create the perfect outdoor environment in otherwise windy city alleys, sweltering rooftop gardens, and sun-filled balconies. Lauren sees this is part of a bigger trend.

“We have increasing populations, urban sprawl, bigger buildings and larger building footprints. Add to this, the amount of outdoor space in our cities is reducing.” If we’re going to live comfortably in city spaces year-round, we need to make sure those spaces are inviting in all weather conditions.

To do this, a building physics engineer’s first step is usually to consider the thermal comfort of occupants, which is dependent upon several variables, such as air temperature, air velocity, relative humidity, mean radiant temperature (MRT) and the clothing and metabolic rate of an individual. MRT is the radiant exchange between people and their surrounding surfaces, depending on how hot or cold the surfaces are, and how close or far you are to them. It is a key factor that influences an individual’s thermal comfort and is also notoriously difficult to calculate.

Stand outside in a city and observe how these variables feel – the sun on your skin, the wind rushing between skyscrapers, the radiant heat of concrete pavement that’s been exposed to the sun. You’ll notice that their effect on you changes, almost by the minute. It follows that good design would also change by the minute, something that both Lauren and Alex were keen to explore.

A thermal image of pedestrians in Melbourne shows how bodies radiate heat into their surroundings.

They call this responsiveness dynamic design. True dynamic design is an ambitious goal. But, as the use of sensing data increases, and following a machine learning course at the University of Technology, Sydney, Lauren and Alex saw that it could be actually achievable.

“We often simplify our calculations or perform static calculations because calculating MRT dynamically is challenging. Through this piece of work, we wanted to investigate ways to improve our methodology to make it flexible and scalable so that we can ultimately design better outdoor spaces. It also offers a feedback loop for our designs” Lauren explains.

The team began with a literature review on experimental methods to measure MRT. They then chose a north-facing balcony in Sydney, and set up three sensors for air temperature, light and a thermopile (infrared sensor). The sensors were all analogue and came with calibration charts. The data was fed into a wi-fi enabled microcontroller, which filtered it and published air temperature, surface temperature and light level data into an online database. So far, so good.

 

However, when they analysed the data, it became clear that the results were flawed and didn’t match the conditions they thought they had on the balcony. After trying four thermopile sensors and continuing to get data with inaccuracies outside an acceptable range, they set up an infrared thermal camera as an alternative data source.

Lauren Boysen in front of the spiral staircase leading up to the outdoor Skypark in Melbourne outside our office, which has been designed to be comfortable outdoor space.

Lauren and Alex aren’t the only ones to run into issues with quality of sensor data. The Internet of Things, which we tongue-twistingly define as connected sensors measuring real-time data for better decision making, is an exciting area for engineers who need to measure their buildings, structures and environment. But a big challenge we often find when using sensor data is that they are often not calibrated to a level where we can confidently use them for engineering applications. We’ve been working through this problem on a number of other research projects, and have built our own custom sensor array and sensor platform that has been applied to measure and report on environmental conditions along the alignment of the Melbourne Metro tunnel project.

Back to the outdoor space, Lauren and Alex realised the value of their research was less about the quality of the data and more about what they could do with it once collected. Rather than try to improve the data itself, they decided to figure out how they’d process and analyse it for dynamic design. They’d then be ready to use it for when sensor technology improves. Judging by the rate of technological change, we don’t think it will be long.

Lauren and Alex designed the workflow with the software, aiming to build a system that could be scalable, applicable to other projects, accurate and reliable.

Wanting to test tools and analytical methods that we already use, and to contribute and build on others’ work, they used open-source software, EnergyPlus, which is a dynamic simulation tool for energy modelling. The work was undertaken in the Grasshopper platform using the EnergyPlus plugin. Grasshopper is a program that allows engineers to model building performance within the Rhino platform. It allows for parametric design, meaning that engineers can design the building as a system of related variables. When they want to change something, like the angle of a window, which could affect another ten variables, they can just plug that into the model and it adjusts across the system without the engineer having to calculate over and over again.

The problem with using different programs and plugins for specialised engineering applications is that they don’t always work perfectly together. In this case, there were limitations related to the level of control of boundary conditions, even with the EnergyPlus plugin. This directly affects the accuracy of the predicted external surface temperatures, like the floor of a balcony. A workaround is to do this directly within EnergyPlus, outside of Grasshopper. You can get away with it, if you’re happy with some degree of variability in the results, but often this isn’t acceptable. So, Lauren and Alex had a trade-off to manage. They could either design a process that was efficient or accurate. They tested both.

A thermal image, taken in one of Melbourne's famous laneways shows how different surfaces radiate heat differently.

Method 1: Efficient.

Modelling completely within Grasshopper using the EnergyPlus plugin.

Maximum scalability, minimum control

They set the balcony model up as a series of mini-blocks to form a ‘grid’, designated each as concrete (from a fairly limited list of options), and as ‘ground zones’ to reflect solid blocks that have no internal zones. They set the boundary conditions of the sides and undersides based on the ground temperatures in the weather file they had, which they used as a proxy for an adjacent ambient air temperature. However, because they couldn’t alter the temperatures within Grasshopper directly, they couldn’t control the various boundary conditions enough to have confidence in the results. 

Method 2: Accurate.

Modelling in the EnergyPlus native interface, then importing to Grasshopper.

Minimum scalability, maximum control

In this method, Lauren and Alex edited the inputs in more detail in EnergyPlus first, and then imported them back into the Grasshopper platform. This gave them more confidence in the results, but the method was in no way robust enough to handle different types of projects, particularly when assessing precincts and larger scales.

We now know for certain that these easily accessible tools are not the right fit for achieving dynamic design, yet. However, we have a good understanding of where the problem areas are, and will keep track of both sensor quality and analytical ideas and developments over the next year or so, when we’ll try again. We'll be ready to apply machine learning to the data once we can get the quality right.

“We learnt a lot about sensors, calibration and filtering data during this first stage of the project” says Lauren. “This will be a key focus for the next stage, enabling accurate measurements, including exploring opportunities to perform post-occupancy surveys.”

It’s the combination of sensing the environmental conditions, through post-occupancy surveys, and designing the system to adjust dynamically that will be the game changer in improving the comfort of people on the street. Getting this feedback loop right will mean those of us living and working in the city can enjoy its urban spaces throughout the year.

Findings

  • Efficient design for our comfort in ever-changing outdoor city spaces through dynamic design requires an advance in technology, but we can prepare now.
  • The two biggest barriers to dynamic design now are the quality of the sensor data, and the interoperability of the processing and design software. Both are expected to develop quickly.
  • As with most sensor projects, data collection and analysis is easy, but setup is hard. In this project, most of the time was spent calibrating and filtering data. See here for our open source calculations.

This story was written by Eleanor Whitworth and Lauren Boysen, as part of the Research Review series. The series is produced by the Arup Australasia Research team; Alex Sinickas, Bree Trevena and Jeff McAllister with contributions from Sheda and Noel Smyth.

Lead Arup Researcher

Lauren Boysen
Lauren is an building physics engineer in our Melbourne office.

Ask Lauren about:

  • How to design for comfortable spaces in cities and buildings.
  • Experimental methods for measuring MRT, and technical aspects for determining and designing for thermal comfort.
  • Approaching solutions for dynamic design and closing the design loop.

LEAD Partner RESEARCHER

Research TEAM

Samantha
Peart
Samantha leads our building sustainability team in Melbourne.
Alexander
Hespe
Alex is a sustainable design consultant in our Sydney office.

Have a problem or a project?

We work with industry partners, governments, universities, startups and community organisations. We do this through research partnerships, and as consultants and facilitators for foresight, research, storytelling and technical writing workshops.

research with us