The Xiaomi Particle Counter Is So Inaccurate It Should Not Control the Purifier

科技公司很喜欢说谈论自动化,网络化的产品和智能家居要让我们每天都能用到的东西变得更方便使用。但是实际情上有那么方便吗事实是这样吗?用过苹果Siri功能的人用户都明白科技公司的豪言壮语有时候会落空。

 

小米空气净化器相信自动化会让我们生活方便多了。问题是这个机器觉得有毒害的空气是安全的。

 

这个是小米的空气净化器Mi2:

 

 

这个是Mi2旁边有一个空气质量测试仪。这是它app显示的空气质量读数:

 

 

小米的理想情况保证

如果小米能实现他们的自动化系统,这是小米保证的计划:测量你家里的空气到底有多差,空气不好的时候,空气净化器机器运转,空气变好以后关闭机器。也就是说我们不用一直让扇叶整天高速运转,也能呼吸到新鲜空气。当然了,低噪音和少耗电是个超酷的事。

 

实地测试

我测了Mi1,Mi2和更高级的Mi2 Pro。我比较了小米的颗粒物测量数值和三个不同的空气颗粒物 。三台小米都不是新的,是用在普通的房子里。 Mi2相对新一些,只在之前用来做过几个礼拜的实验。

 

我在Mi2的内置颗粒物测量仪旁边放了一把椅子,椅子上挨着内置测量仪的位置放了两个镭豆和Air Visual Node。

 

 

我在一个密闭的12平米房间里燃烧了一只烟,然后把Mi2打开到高速,直到空气变得干净(整个过程大概30分钟)。这样我们就可以测试几个颗粒物测量仪在空气洁净和严重污染时的结果。我给手机设定了自动拍照,在整个测试过程中每30秒就拍一张照片记录读数。

 

我对小米的颗粒物测量仪要求太高吗?

在我公布结果以前,想先让大家有一个预期。我没对小米颗粒物测量仪的准确度抱有太高的期待。小米整个机器比我对比的空气质量测试仪要便宜,所以我们设定预期的时候也应该现实一点。

 

现实的预期是怎么样的呢?我不要求它特别精确,只要它可以实现设计之初的设想就可以:让自动开关模式正常运行。

实验结果

小米距离最低的预期都还差着一截。空气很不好的时候,小米测量的数据偏差能达到218微克。

 

 

这个偏差到底有多大呢?世卫组织24小时测量偏差限制是25微克。小米的测量误差已经超过世卫组织误差上限的8倍了。

 

下面的图片是我的手机记录的测量拍照:

 

 

小米好像是到50微克就不数了。 这样小米的读数就显示室内空气指数是橙色级别(“对敏感人群来说不健康”),但事实上室内空气指数是属于紫色的范围(“非常不健康”) 。

 

所以,小米的颗粒物测量读数是比实际情况要低的,并且有时候低非常多。这还不是它唯一的毛病,如果我们聚焦在整个空气颗粒物数量较低的情况,小米的测量值又高了。

 

 

可能9微克的差距感觉起来不是个事儿,但从另一个角度来说,小米对颗粒物的测量却是实际颗粒物的10倍。

 

 

Mi1也不精确

也许是我的这台Mi2有问题。可能快递公司运输过程中摔过机器弄坏了内置的颗粒物测量仪。

 

为了排除类似的可能性,我比较了更旧的机型Mi1和Dylos Pro(这台机器和官方PM2.5测量数据对比也得到一个很高的分数)。结果跟Mi2一样。

 

我还对Mi2 Pro进行了测试,结果也是一样。所以这个问题看起来是小米空气净化系统的普遍问题。

 

我们怎么知道那些数据都是对的呢?

等一下,我们是在假设镭豆和Node的数据都是准确的了?我们怎么知道那些数据是对的,小米是错的呢?

 

Smart Air用六天把镭豆和Node的测量数据和官方PM2.5数据进行对比。Node和镭豆跟官方数据的相关性很高,达到了r = .98,Node的测量误差平均是4.8毫克,而镭豆的测量误差平均为6.5微克。所以我相信这两个测量仪得到的数据是和颗粒物真实浓度相近的。

 

结论

小米的颗粒物测量仪准确度非常低,这么低的精确度说明这个测量仪不应该被用来控制我们家里的空气净化器。问题是小米并没有给用户选择的余地(我在下文会具体说明)。

 

这个问题可以解释为什么小米开机之后,室内污染 还是超标

SmartAir在北京一个普通的公寓里做了12晚的测试。测试结果让我很惊讶。我原本以为它可以做的不错,因为说到底空气净化器就是扇叶和滤网,虽然有一些品牌努力试图说服我们空气净化器是一种高科技的东西。可是小米的测试当中, 有86%的时间,室内污染超标 。

 

 

小米空气测试仪严重低估污染等级正好解释了为什么开机之后,室内污染还是不安全。我在测试飞利浦的自动模式是也发现了一样的问题。因此我认为空气净化器中自带的自动模式技术还不够好,我在自己家里也不会使用自动模式。

 

不仅仅是精确度的问题

Mi2开高档的时候,是个不错的空气净化器。我们的公开数据实验表明Mi2高档非常有效(看看上面测试图的前三个小时数据)。但问题是Mi2强制用户使用自动模式,不管你对机器做什么,三个小时后以后它自己就会跳转到自动模式。听着怪怪的对吧?为了确认这不是误会,我们问了小米客服三次

 

也就是说除非你晚上每隔三个小时起来一次,把净化器开到高速运转,否则你就只能乖乖用受随机自带的颗粒物测量仪控制开关的自动模式。我希望小米尽快改正这个简单的设计缺陷,但他们作出更正之前,我肯定不会在我家使用小米。

 

更多数据和方法

Mi1测试方法

我在Smart Air办公室测了Mi2,在我家测了Mi1,所以房间和方法上会略有不同。我在办公室点了一根烟使室内的颗粒物指数上升。我家里没有烟,所以我就点了一张纸。

 

办公室的空间是12平米,我的房间比办公室大一些,大概有15平米。

 

小米不精确是因为颗粒物测量仪藏在机器里吗?

我想过是不是小米的颗粒物测量仪不准确是因为它被放在机器里面,测到的空气跟机器外面的空气不一样。一个验证办法就是把测量仪从机器里拿出来再做测试。这倒不难,因为测试仪可以拿出来。我这么做了,可是得到的数据还是一样:污染严重的时候机器显示的数据还是过低,污染情况变好以后机器显示的数据还是过高。所以,我认为小米的颗粒物测量仪问题不在于它被藏在机器里。

 

小米测量仪得出的具体数字是怎么样的?

小米颗粒物测量仪的另一个问题是它不显示自己给出的空气质量数据单位。他们用的单位是毫克?还是中国空气质量指数,或者美国空气质量指数,还是别的什么?我不明白为什么他们不标注一下。

 

这不只是一个书呆子的顾虑,而是真的会影响测试结果,因为毫克和空气质量指数的关系并非线性关系。

如果你拉着小米的人追根究底,他们会跟你说他们的空气测量仪用的单位是毫克,所以我在数据对比时都是用毫克作为单位进行比较的。

 

小米的测量数据能到10毫克以下吗?

小米报告的最低数据是9毫克,Node是0.2毫克,镭豆是1毫克。所以我在想,小米是否会显示低于9的数字?还是测量仪设计之初就定好了不显示低于9的数据?

 

为了探个底,我把DIY1.1直接冲着小米颗粒物测量仪运转。我之前用这个办法来测试Dylos的时候,Dylos数据低到了0。不过对小米的测试中,小米显示的数据一直在10左右。所以我觉得不是小米在测算幽灵颗粒物,就是它已经被设置好了不显示低于9的数据。

 

原始数据

我把原始数据的Excel放在这里,你可以点击下载

 

Node的数据 

为什么Node的显示数据就很低?

文章里的图表显示Node的数字比两个镭豆都要低。我在实验中使用的那台Node已经用了一年了(虽然镭豆也不是全新的),比较老的颗粒物测量仪会有的一个问题是灰尘堆积在机器内部会影响空气流动。做AQIcn.org网站的小伙子之前测试过一个旧的Dylos,发现污染物颗粒浓度大的时候机器就是会低估实际情况。他用压缩空气清理了一下机器内部,发现得到的数据又上升了。

 

我怀疑我们的Node也是一样的情况。这应该是颗粒物测量仪被暴露在颗粒物密度高的空气里时会遇到的问题,就像在我们的香烟试验里那种情况。

Thomas Talhelm

芝加哥大学行为科学系的助理教授; Smart Air创始书呆子。

在 “The Xiaomi Particle Counter Is So Inaccurate It Should Not Control the Purifier” 上有 16 条评论

  1. I tested in my home two Xiaomi Air purifiers (2nd generation), and the stand-alone Xiaomi PM2.5 meter.

    The stand-alone meter is using laser technology, just like any other decent PM2.5 meter on the market. Yet, Xiaomi refuses to answer what technology lies behind the integrated sensor within the Air purifier (1st or 2nd generation). It is interesting to note, that on its 3rd generation of the air purifiers, Xiaomi ditched the old sensor technology, and switched to laser technology too (same technology as in the stand alone meter).

    From my experience, the stand-alone PM2.5 meter from Xiaomi seems to be accurate (when comparing to government PM2.5 meters in my area). Regarding the integrated meters inside the two Xiaomi air purifiers (2nd generation) that I’ve tested, one was constantly ~40 ppm higher than the stand alone meter, and the other seemed to be more in correlation with the stand alone.

    In any case, anyone who uses the Xiaomi air purifiers (2nd generation), simply purchase the stand alone meter if you want accurate control.

    1. Hi, I recently bought a Mi Air 2 purifier and was interested in buying the separate pm2.5 monitor, but so far I’ve found virtually no info on how much control and automation you can program the purifier via the data from the monitor. Can you tell me what is achievable with the two devices working together? I’ve already got a Laser Egg 2, but that has no way of controlling the speed of the Xiaomi purifier.

  2. I love your data and hard work. I curious though: 1. How does the Xiaomi standalone particle counter stack up?
    2. How does the mi2 performance rate if left at medium, hi, or even turbo?

    1. Hi Max! I tested the Mi1, Mi2, and Mi2 Pro. Unfortunately I only have a record of the Mi1 and Mi2 data. The Mi2 Pro was at a friend’s apartment, where I had access for a limited period of time. The pattern of the data on the Pro versus the Dylos and the Laser Egg was the same as the Mi1 and Mi2–when air was bad, the Mi2 Pro was vastly undercounting. When air was very good, it was overcounting.

      Does Xiaomi claim that the Mi2 Pro has a better particle counter? I’ve searched their website and haven’t found any claim about the particle counter. If you find a specific claim, let me know. Also, if you have a particle counter and the Mi2 Pro, let’s set up another test!

  3. Hi!

    You can easily find on the internet, that versions “Pro” and “2s” have new laser particle sensor. I’d hope it is far more accurate than the old sensor in the version 2. Would be great if you could test it somehow…

    1. I own Xiaomi Purifier 2S with PM 2.5 display, it’s showing really low values most of the time (like 007) and it’s quirky, reading from low to high in short intervals while nothing changed indoors. It’s showing 001 last couple of days… seems like malfunction after 3 months.

      1. My story is similar to as MATIJA’s. Received Xiaomi Air Purifier 2s few days ago. It shows 001 almost all the time. However, when smoke is around or spraying something near it – it goes to 70-200-600. Then in few minutes goes back to 001. Looks like working, but don’t understand how my indoor air could be so clean, just don’t believe it.

        Any solutions or comments appreciated.

  4. I can confirm that the air purifier 2 never shows less than 9. As for the issue with the AP2 going back to auto after 3 hours it should be possible to circumvent this using the app’s automation options. E.g. set it to change modes at certain times. Not an ideal solution for sure, but should work. I just got the Xiaomi laser pm2.5 sensor and it generally shows far lower values than the AP2 built-in sensor. The sensor is linked to the AP2, but unfortunately the automation options that deal with pm2.5 concentration don’t seem to work very well. When I set a rule that the AP2 should turn on when PM2.5 levels exceeded 15, the rule wasn’t triggered when it reached eg 17. However when I set the limit to 35 and i sprayed some shoe impregnation spray near it or burned some paper it worked without issue. It seems it has problems with the smaller numbers or something.

    1. Right on–our tests found that the Xiaomi particle counter had problems both in low AQI range and the high AQI range. My guess is similar to yours–for some reason, they seem to have programmed it not to go below a specific number.

  5. Xiaomi also offers a standard alone monitor. Will this essentially be the same tech as found within the air purifiers? If so I’m worried about wasting my money.

    I’m debating whether to buy the xiaomi air monitor or the laser egg. The reason I’d prefer the xiaomi is because I already have a fair amount of stuff in my house from them! It’ll all be the same smart system, rather than using yet another APP for the laser egg.

    Does anyone have any ideas?

    1. Good question Chris. We’re yet to actually test the standalone Xiaomi Air monitor (although it’s something we want to do soon!). Ultimately the sensors in these devices are all the same (they cost around $5-$10 per sensor) but what’s different between each monitor is the calibration and testing of the devices. I know this is something Laser Egg have worked on a lot (see the article we wrote about how humidity affects the Laser Egg), but I’d be less sure about the Xiaomi (since they’re a company that puts out 101 different devices). Having said that, I have no doubt it’ll be ‘reasonably’ accurate, but accurate enough to ensure your air is clean 100% of the time… that’s a difficult one to answer.

  6. Hello,
    just read the article out of curiosity, and well some points are somehow weird. I belive the sensor is quite accurate, if they build in what they claim to use. Purely from the technical PoV.
    But going by your data it is more likely, that the software used is wrongly programmed. They write to use shinyei sensors, which do not give an output in any unit but in an voltage equivalent between around 0.5 and 5V. So going by this, the Data you get is most likely the output Voltage of the sensor. Most likely the developer for the purifier took that output directly and fed it to the software as input.
    And herein lies the problem, going by your data he most likely used the wrong stepping for the adc in the control unit, which resultet in a voltage output in far too high steps.
    Well to say it in an easy way, the got a dounce to programm the software and set the electronics, and in the pro version they most likely simply used a sensor which gives digital output and therefore does not have that problem.

    1. Rolf, you’re right that the software-side of particle sensors is often the most important, since there are so many variables at play especially when you’re converting from a particle count (which the sensor gives) to a concentration or AQI (which is what we’re most often used to). In some of the more accurate air quality monitors we tested, they all use fine-tuned software to adjust for different variables.

      Having said that, there are other issues with the sensor in the Xiaomi, firstly that it’s located right next to the purifier. This means you’re only getting an indication of the air around the purifier, as opposed to on the other side of the room. This can be dangerous if you’re using the Xioami in auto-mode, where the sensor is reading clean air (near the machine) but the truth is the air on the other side of the room is not clean.

  7. Moin Moin,

    thank you very much Thomas for your interesting and well written articles. I share the same observations with my Mi2 purifiers but comparing the Mi2 Pro to my laser egg and Xioami particle counter, the results fit much better and also the MI Pro goes down to 1.

    A tip for all Mi purifier users: I only use my purifiers in manual mode! How does it work? You just have to program under automation: If the PM2.5 concentrations is below 500 (which means always), switch to manual mode. This is how you can prevent them to switch to auto mode after some hours automatically!

发表评论

电子邮件地址不会被公开。 必填项已用*标注