Listen Up: iPhone Hack Diagnoses HVAC

We all know that guy (or, in some cases, we are that guy) that can listen to a car running and say something like, “Yep. Needs a lifter adjustment.” A startup company named Augury aims to replace that skill with an iPhone app.

Aimed at commercial installations, a technician places a magnetic sensor to the body of the machine in question. The sensor connects to a custom box called an Auguscope that collects vibration and ultrasonic data and forwards it via the iPhone to a back end server for analysis. Moving the sensor can even allow the back end to determine the location of the fault in some cases. The comparison data the back end uses includes reference data on similar machines as well as historical data about the machine in question.

We couldn’t help but wonder why the hacker community hasn’t been building things like this for a long time. We have plenty of CPUs with enough horsepower to collect audio (or other sensor data) and process it. If you wanted to diagnose cars, for example, you can find sounds readily enough for training.

There are a few academic papers on the subject, but nothing we could find in the way of hacker-style projects. Maybe the comments will flesh out some. If not, this seems like a ripe area for experimentation. Acoustic analysis of a PC could identify fan, power supply, and hard drive problems (as long as you aren’t using a solid state drive). Anything with moving parts could be a candidate.

About the closest thing we found in the archives was a hybrid magnetic/audio system to plot motor balance. We’d love to see some more work in this area over on Hackaday.io.

Images courtesy of Augury.

29 thoughts on “Listen Up: iPhone Hack Diagnoses HVAC

  1. methods like these are used often on ships and in wind turbines.

    more than a decade since they started mounting multiple dedicated microphones in every gear and bearing in turbines.
    automated software will then alert the service company should anything be wrong.

    1. Brüel & Kjær had (very expensive) accelerometer-based diagnostic systems like this in the 1980s for pumps, turbines etc. – nothing as smart as these – you were basically looking at spectral analyzers of the raw data for shifts in particular frequencies, but the principle was the same. If I’m not mistaken, this is also a big part of GE’s engine analysis software that keeps an eye on all their engines in service.

    2. Have seen something similar in use in a petroleum cracking plant, maybe 15 years ago. Microphones were mounted on over a dozen critical mechanical components. A computer ran Fourier analysis, and alerted a technician if there were significant differences in spectrum from a baseline measurement.

      That much would be quite easy to recreate. Automatic diagnosis would be harder, but getting early warning of mechanical issues is a big enough advantage on its own.

      1. Depending on what you’re looking for, a fourier analysis can still likely be the core of this, could it not? Simply match which frequencies are not in line with the baseline, and then use those measurements to estimate any issues. Some of the techs I work with can diagnose general issues (bad bearing, coil issues, etc) just on the sound alone without tools.

    1. The advantage of iDevices over cheap android etc. is consistency and availability..

      In a hardware monoculture, you can be sure if you test on models X, Y, Z it’ll work consistently. With Android, the various manufacturer variations, software and hardware, weird stuff creeps in and the companies just couldn’t be bothered debugging that in some cases..

      Don’t get me wrong, I’m not an Apple fan, but I see why a third party device manufacturer would find their uniformity enticing.

      1. that uniformity is non existant, several production runs of iphones have slightly different hardware as well and most of the components are brand agnostic, they are in both androids and iphones.

        1. Why are you guys discussing iPhone’s. In the video it doesn’t even mention the iPhone, just a “smartphone”. It doesn’t matter what phone you have, you just need the app that interfaces with their proprietary hardware.

          It’s not about Android or iPhone, heck they could of incorporated an esp8266 as an AP and made a small server that wasn’t OS dependable and just hosted some small HTML to expose the data.

      2. (Nokia fan boy alert) You have different generations of iphones as well to think about, anyway my nokia has a digital mic, I can record a jet taking off, crystal clear, try that with any other phone…

        1. Not a problem, I have done it with iPhones as well as a HTC ONE M8. Nobody makes a 100% digital microphone as it’s impossible, you still have an analog electret mic element going to an analog to digital converter. Just because they put it all in the capsule does not make it any better than an electret mic hooked up to an ADC it just makes it smaller so that it is easier to fit inside a phone.

          1. There are digital MEMS modules with a MEMS sensor element packaged with a preamp and ADC. Still has some analog components but the analog all stays inside the module.

  2. The only innowation I see here is using cloud to store and analyze data. Electric engines are analyzed using sound for many years now. One does not trully need internet connection to figure out that something is off with the engine.

  3. Car engines have had microphones on for years – they’re called knock sensors. Analysed in realtime by the ECU, quite a knotty DSP problem determining what’s a bad knock and what’s the noise of al the other clanking, banging and explosions going on in an engine.

    Also, “iPhone does XYZ” is misleading, the iPhone is doing nothing here that any other phone or tablet couldn’t do.

  4. You could also check an SSD or any other steady part. Just look at acoustic cryptanalysis.

    In 2014 the Pwnie Award for the most interesting research went to guys cracking a 4096 bit RSA GnuPG key within one hour simply by analyzing the sounds coming from the capacitors of the CPU while it was decrypting some chosen ciphertexts. The sounds needed were above 35 kHz. But it worked even with a normal smartphone microphone, normally recording at 48.1 kHz, so its frequency response along with aliasing only goes up to 24 kHz (even less, regarding sensitivity and internal filtering). I don’t really understand how they worked it out but they did. So you don’t even really need an ultrasonic microphone for it.

    So everything produces a sound, this way you can even analyse CPU, RAM, SSD, GPU etc. You just have to specify the parts you’re using, run some specific instructions on your targeted device, et voilà, your smartphone will be a bit more useful.

    1. Pretty sure that there is a lowpass filter before the ADC of the microphone to prevent anti-aliasing problems with ultrasonics. Same for the speaker, but backwards.

      As for hearing what the caps “say” – then whining/noise you can hear (some are significantly louder then others, but almost none are completely quiet) is discontinuous mode of the PSU…

      1. Which effectively makes it a one-step-removed powerline monitoring attack. An ammeter can tell you when and how long a CPU “struggles” over a cryptographic problem, and discernible patterns there are what enable this particular attack. Knowing the mode and/or duty cycle of the switching element in a switching supply is almost the same thing.

  5. in the automotive trade we do this with our ears and a long screwdriver (one side touching the casing of the bearing in question, other resting on your ear) a little experience and diagnosing a troubled bearing becomes really easy. A good trick with wheelbearings is the road springs often resonate at the frequency of the rumble produced by a failing bearing, so you rotate the wheel and touch the middle of the road spring nearest that wheel, a failed bearing is really really obvious in the vibrations in the spring. the posher tool for this is a stethoscope with a long metal rod which is touched on the casing of the bearing in question

  6. I have used a freq minitor on submarines, normal cellphones would not normally work because the standard analyzer records to over 40,000Hz. Don’t know of many standard mirophones that go that high.

  7. You could probably use a crystal transducer with a pre amp or even with using a cell phone most come with a accelarometer that you could place the phone on the object to measure data.

  8. I worked on something reminiscent of this in the late 1990s. Since mechanical parts sound different as they wear, I figured taking a baseline vibration FFT (say, 20hz-22000hz) and then periodically comparing the reading to the baseline would have the potential to tell you at least when mechanical components were going bad. At the time I was using GoldWave and CoolEdit to look at waveforms, fft, and spectrograms. I was thinking about whether I could track fluid quality based on pump sounds (oil, power steering, A/C, water) but was distracted by wanting to graduate.

  9. This is called Signal Conditioning and, while it is nothing new, it has gained popularity with time and cloud computing availability.

    Basically, you have several accelerometers and microphones coupled to a machine. You constantly monitor the spectrum and you can easily determine when a bearing is about to fail. This way, you can schedule a maintenance stop at the most convenient time without affecting the production.

    Bearing noise lies basically in the ultrasonic realm of the spectrum. An ultrasonic whine means that something is bad with the bearings of your machine and they should be replaced. Whine frequency decreases with time as the bearing looses stiffness and eventually it will enter in the audible spectrum as a very subtle high pitched whine.

    FFT gives us more information: If there is a peak at the operating frequency of the machine that we identify as the fundamental frequency, the magnitude of the first harmonic will indicate the magnitude of a misalignment.

  10. There’s a “sensorless” refrigeration compressor VFD (Shannon Liu Quadrature Drive) that exploits the torque pulsations of a reciprocating compressor to derive the rotor speed. From my understanding, the DSP used to control it just runs the current sensing values through a downconversion, then uses FFT to extract the slip speed and uses that along with the (known) drive frequency to derive the rotor speed.

    I have actually once digitized the current signal of a rotary compressor and was able to extract the rotor speed using FFT, but it was much a weaker signal than what a reciprocating compressor provides. That trick probably won’t work very well if at all with a scroll compressor, but a piezo sensor glued to the casing to sense the vibration would probably work well.

  11. I think this company is on the right track. Using fault frequencies with a accelerometor to find defects in rotating machinery is a known Science. Now that we have entered the tablet age it is the time to add sensors to our smart devices. GTI Predictive technology is useing the same technology for not only rotating machinery but robots as well. This company is on the forefront of technology!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s