Cloud Cover: How machine-learning diagnostics can keep your air conditioning system healthy and save you energy.

If you’re responsible for your site’s air conditioning plant, you’ll know that if it has a problem, you have a major headache on your hands. Whether in an office, school, university, shopping centre or hospital, HVAC (Heating Ventilation & Air Conditioning) equipment is mission critical. If it goes down, your business suffers.

The coronavirus pandemic, along with rising levels of air pollution,  has put IAQ (Indoor Air Quality) on the map, and HVAC is so important to ensuring not only comfort, but also health and safety, in occupied buildings. Maintaining air exchange rates and ensuring air supplied into a building is filtered is now an expectation on building owners and people are becoming increasingly cognisant of the criticality of a properly maintained and operated HVAC system.

So, HVAC is a critical component of a building’s infrastructure. Unfortunately, it is also one of the most expensive to run. Typically, the electricity required to run a HVAC system represents 40% of a building’s total energy cost, a significant percentage.

However, HVAC systems are also complex. A typical system usually consists of a chiller, producing cold water that is then used to cool air that is brought in to a building via an Air Handling Unit (AHU). The AHU filters and conditions the external air, then distributes it to the building spaces via ducts, diffusers and indoor fan coil units.

The many variables involved in a HVAC system, like water temperatures, system pressures, valve positions, room set points, ambient temperatures and pump settings, can wildly vary the amount of energy consumed. Effective management and maintenance of such systems is key to ensuring they run as effectively and efficiently as possible.

Energy Managers / Facility Managers have a lot on their plates, this is no secret! But they have a helping hand in today’s age of connectivity and digitisation…software.

Software has long been mooted as the silver bullet for overworked energy / facilities personnel. The proliferation of building management systems (BMS) across the built environment means these people can supposedly spend most of their time indoors, sat in front of dashboards tapping keyboards rather than on cold rooftops tapping pressure gauges, safe in the knowledge that they have full visibility of their HVAC systems.

But

Most building management systems tell you what’s going on, without telling you what’s really going on. They will tell you your water temperatures, system pressures, valve positions, room set points, ambient temperatures and pump settings, without telling you what they mean. Data is nothing without intelligence. To measure is to know, but to analyse is to understand.

Airedale is the largest manufacturer of HVAC equipment in the UK. We also have one of the largest service teams…and they’re very busy. Mechanical equipment, operating outdoors and operating non-stop, will suffer breakdowns. It’s unavoidable.

The symptoms of a full breakdown can be devastating to your site and your maintenance budget. But as with people, the smaller symptoms often start before the main ones hit; a runny nose, a slight fever, a cough in people. Elevated energy use, pressure fluctuations and performance drops in HVAC. These tell-tale signs are often undetectable to a human looking at BMS dashboards or monitoring dials and meters.

One of the big issues with chillers especially is refrigerant leaks. Refrigerant is costly, both in terms of price per kg and often cost to the environment. R410A, a commonly used refrigerant in HVAC systems, has a global warming potential over 2,000 times that of CO2. Gas has a way of escaping sealed mechanical systems; braised joints can degrade over time, valves and connections can become loose, coils can degrade. If your chiller loses refrigerant, its performance will start to suffer and it will become less efficient as refrigerant levels drop. Unfortunately, this initial drop in performance can be un-noticeable to the human eye, until levels reach approximately 80%. At this point performance will be so badly affected that building staff will almost certainly notice and the chiller may well completely break down. At this point, the damage has been done and you are looking at an expensive repair bill, not to mention an increased energy bill from the time the machine started leaking.

So what is the answer?

The solution is to go one layer deeper with your software systems, to analyse instead of monitor. To predict instead of report. It is to couple software with HVAC knowledge and leverage data science in the form of machine learning, to detect small drops in performance, compare data with field experience and identify problems before they manifest into outages / expensive failures.

If this all sounds expensive, it doesn’t have to be. The advancement of the internet of things (IoT) in recent years has seen sensor technology become more affordable, connectivity options increasing and analytics becoming more accessible.  ‘Software as a Service’ (SaaS) business models have evolved to deliver flexible options to clients, with up front costs reduced in favour of subscription fees that allow budgets to be matched with pay-back periods.

How it works

A gateway is added to the equipment allowing it to collect data from the unit controller and transmit this over a secure internet connection (provided by a built in modem or SIM card) to a remote cloud server. 

Connected units will be analysed for performance utilising a variety of algorithms and machine learning techniques. If a drop in performance against operating conditions is detected, this will act as an early warning system for the customer/maintenance team to investigate further. Early intervention can prevent prolonged periods of higher energy consumption and eventual breakdown and all the costs associated with that.

The system will judge the unit’s performance on a variety of factors – for a refrigerant based cooling system like a chiller, superheat, sub-cool, suction/head pressures and water flow/airflow are all analysed for deviations against “normalised” behaviour, however the overall output of the unit such as cooling duty/capacity, efficiency, power consumption against its operating conditions are all considered instantaneously and over time as well. The system will use all this data to recognise “failure patterns” and warn the user of a potential failure before it happens. For example, a gradual increase in power consumption could be normal if the unit is working against an increased load over time, for example as a building fills up on a warm day. However, paired with other indicators such as a gradual change in normal system pressures, this could indicate a breakdown is imminent.

Information is transmitted to the user via a secure dashboard, accessible via the internet on desktop or mobile. Alerts can be configured to be sent to multiple email addresses and live data as well as historical can be analysed and reported.

Refrigerant Leak Detection

As mentioned above, catching refrigerant leaks earlier by detecting small losses in performance or changes in the power profile, delivers huge benefits in terms of environmental impact, operational impact and overall cost savings to the end user. The ability for connected units to be able to learn from and compare against each other utilising intelligent unit modelling means that the performance analysis techniques continually improve and get stronger over time. In our tests we have been able to detect refrigerant leaks as small as 5% of the total refrigerant volume.

The above dashboard displays the analysis result of one chiller where refrigerant was temporarily removed from one of the circuits. The machine learning algorithm compared the telemetry received from the chiller against a pre-learned model and correctly detected numerous anomalies on the chiller’s behaviour. Once the test was completed, refrigerant was then reintroduced in the chiller. The algorithm responded to this and no longer detected anomalous data.

Energy / Facilities Managers could be forgiven for thinking “not another software system”, but as HVAC grows in importance and machine learning continues to be pervasive, it is natural that large capital and operational investments like chillers and air handlers require extra layers of protection that a standard BMS cannot provide.

Leveraging technological advancements with evolving business models like SaaS mean investment decisions are more straight forward, with investment spread and payback periods easy to quantify.

For more information look up Cloud Diagnostics on www.airedale.com