Smart Home: 1.5 years of Aqara Temperature and Humidity Sensors

As one of the initial steps in my smart home journey back in November 2023, I purchased a set of 12 Aqara Temperature and Humidity Sensors for £101 (approximately £8 each) from Aliexpress. Today, this same set is available for around £89, or £7.40 each.

This was the start of my Zigbee experiments, and you can read more about the initial setup in my first post talking about my initial setup and home assistant dashboarding.

Since then, my Zigbee mesh has continued to expand with a bunch of powered devices, and I have had to start my first round of battery changes on the Aqara sensors. There have also been some power cuts leading to my hub turning off, many home assistant restarts and upgrades, and I feel that I have some more to share.

To this date I have still only permanently fixed one of these sensors in any way, most of them are just resting in nice out-of-the-way places, or on top of door frames.

Mesh placement

When starting down the Zigbee path with these Aqara sensors, I only had my main Zigbee coordinator in the centre of 12 sensors spread far and wide around the house, including one outside in a bike shed. In the early days this all seemed fine however as time passed, the odd sensor, particularly the one in the garden would drop off, and not come back until I physically went and pressed the button. Anyway, this is mostly my fault for not really having a mesh of devices.

As time passed I introduced 4 Zigbee light switches, as well as a fleet of 10 switchable socket plugs with power monitoring, and another light controller, all of which were powered.

This would be great for the future, but the fact that I had to go around again and re-pair all the sensors just to make them adjust to the newly created mesh was a giant pain. And even after doing so, there are still some devices which (in my opinion) are not taking the optimum route.

Below is a rough visualization of my Zigbee mesh, with the node organized trying to reflect the layout of the house (top is top floor, bottom is bottom, left is one end, right is the other). The orange lines show current links, and green show what would really be optimum in my opinion. The green lines include connections to powered devices that are less than 3m away in the same room, rather than travelling through 4 walls into a cupboard in another room…

More specifically

• The front room climate sensor is line of sight to the desk sensor. Instead, it decides to connect to the kitchen hood near the other end of the house, through 2 walls and a cupboard.
• The master bathroom climate sensor decides to connect through 2 floors (and arguable some walls) instead of just connecting through a single wall to a plug 5m away.
• The Airing cupboard climate sensor is roughly 5m away from the Office desk 2 right plug. Instead, it decides to connect to the main hub through a floor

Okay, the last bullet point might actually make more sense, but my primary point is that this mesh should be dynamic, self adjusting and healing, and or I should be able to tell devices to adjust without needing to run around and click 12 buttons.

Perhaps this is one of the downsides of Zigbee, I may assume that it is dynamic and self-healing, but perhaps it isn’t? or perhaps it depends on implementations? Aqara seem to think that this is one of the benefits of Thread after all…

Key Benefits of Thread:
– Dynamic routing and self-healing capabilities, ensuring robust and resilient network operation.

Battery reporting & drop-offs

As mentioned above, I had the sensors sometimes just drop off the network and stop reporting to home assistant. This would generally expose itself as data no longer coming into ZHA or Zigbee2MQTT (so static last value being shown) as well as the last updated time not updating.

Most recently, for example, You can see the WC climate sensor showing a straight line at the start of this year.

To bring this device back, I changed the battery, and touch wood, for the last few days it has been behaving well. However, before changing the batter, the last reported percentage was only 83%, which is really my annoyance here.

Looking at all of these sensors (unfortunately I trimmed off a bunch of data when moving from ZHA to Zigbee2MQTT), the lowest battery percentage that these devices managed to get to is 63%, however the failures started to happen anywhere from 90% down to this value. So really, the battery percentage reporting is kind of useless?

As I have had to start replacing batteries in devices now, I have added the Battery Notes integration which will be able to track a date of last battery change, so hopefully as time passes I can start to come up with a sensible schedule, or some other rota for changing these batteries? It also tracks the battery type, and you can set a custom battery low level (not that I would know what to set it to for these sensors.

A discussion on the Zigbee2MQTT Github repo mentions power_outage_count which could be an interesting thing to look into moving forward. JonSilver says:

It’s the number of brown-outs. Alas, the battery voltage measurement on many of these sensors isn’t a great guide to a dying battery. When the device tries to power up and do something that draws a lot of current (comparatively), the voltage will drop off a cliff all of a sudden and the processor will suffer a brown-out. When it resets, it’ll note the reason for the reset, and count the brown-out. The battery voltage might still read as healthy under a light load, but actually very few of these devices ever read their battery voltage particularly accurately anyway.

So… a rapidly rising power outage count is a warning that the battery needs replacing. Shame it’s so hard to monitor rapid rises in this sort of sensor in platforms like HA.

Also, Aqara state battery life of 2 years, and perhaps some of my devices will actually get there, but some have had to be replaced after just over 1 year.

Automations, or the lack of

Dipping my toe into the water with a set of temperature and humidity sensors was easy, however I am not actually using any of them in automations yet, simply using them as a nice little thing to plot and look at.

You can see the trends in house temperature throughout each day, and I could dive into this more and plot some other mildly interesting stuff, such as average temp per floor?

The closest I have come to actually using this data is monitoring a bathroom that I am brewing some beer in this week. The instructions say that the ideal temperature is between 20 and 24 degrees Celsius, and I’m using the data from the sensor to manually tweak the radiator settings, and make sure that the door is not left open (should probably be using a door sensor for this).

I actually wanted to control the bathroom extractor fans based on humidity, but that’s probably still some months off until I figure out how I want to rewire the extractor fans with some smart switching in there!

Reflection

Overall, I’m glad that I bought them, and they certainly were a good little introduction to Zigbee. I did some research beforehand and generally Aqara as a choice seemed sensible, and a post from back in 2022 on notenoughtech.com listed them as a good all round option. However, if I were to do it again, perhaps I’d pick something with a slightly larger coin cell? Maybe the Sonoff SNZB-02?

I like the small form factor, but in a few locations it might be nice to have something with a screen so that I can actually see what the temp or humidity is without having to open up home assistant. Of course there are a multitude of ways that this can be achieved, voice assistants, other more dynamic and useful displays, but I do wonder if I should have something a bit more clock like.

At some point I will end up doing some automations to do with humidity and extraction, and perhaps custom heating profiles for my smart electric radiators, but for now, it’s all just pretty graphs…