Analog pressure meter recognize

Hello everyone! I want to make a project on which I will use an Arduino with a camera, that it will "reads" an analog pressure meter and it will sent me a notification if the pressure raises or drops.
The pressure meter will be something like this...


Do you believe that this is possible?

Yes it is possible but a pressure sensor would much more cost effective and accurate at the same time.

2 Likes

I want to use it at a system that has the meter already installed and it isn't easy (or safe...) to install a (cheap) digital meter, because the working pressures are over 150bars.
So could you give me some advice?

Disclaimer - I've never done anything like that.

That would require advanced programming and one of the more advanced Arduinos (if any Arduino can do it at all.)

So, don't use a cheap one! :stuck_out_tongue: I'm sure there are "industrially-rated" high-pressure sensors. Probably made or distributed by the same people as the analog gauges.

2 Likes

Probably, but I'd forget about Arduino and look at single-board computers and machine vision solutions. Arduino's are generally good at low-level tasks, such as reading out dedicated/electronic pressure sensors. Something like a recent model Raspberry Pi (or comparable) can be good at far more complex tasks, that are also programmed in a different way, using more higher-level and scripting languages than the relatively low-level stuff we generally do on microcontrollers.

That only effects the sensor and its plumbing, the electronics does not care what the pressure is. Check this out? Arduino Pressure Sensor Tutorial | MPS20N0040D

If you do not have experience with these pressures do not mess with them.

2 Likes

Post a photo of the gauge installation (good lighting and focus).

https://www.ifm.com/us/en/product/PL5401

Possible - perhaps.
Realistic, efficient, safe ... NO

1 Like

Could be possible, but not easy, with a more powerful MCU, like the ESP32, or a small SBC like a Rasperry Pi.

You should learn many concepts about image recognition, machine learning, neural networks, etc. The process would be more or less:

  • Take a lot of photos and label them with the shown preasure, for many measures and different light conditions, positions, etc.
  • Setup and train a neural network in your computer with these images until the result is good enouth.
  • Deploy the trained model along with all the needed software to the MCU. Not easy.

Learning how to install, program and use all this is not easy and not fast. I have done that but not with a MCU. Search for: ESP32 image recognition.

There are some online cloud services that could make things a bit easier (I haven't tried it), like this one: Adding sight to your sensors | Edge Impulse Documentation

Of course much easier would be to install a second sensor like the one proposed in post #7

1 Like

People have already done this, mostly with RPis.

Here's one:

Another way would be to take a b&w photo, save in BMP format, and write code to find the needle based on where a normally white pixel (or pixels) is black. Depending on photo resolution, an ESP could do it.

thanks for your reply! The links you provided seems very very interesting...!!

Thank you very much for your reply @DaveEvans!!
i believe that I will make it to set up something that fits my needs...

Here is the actual gauge.

Your pictured gauge is a 0 to 200 bar or 0 to 3,000 PSIG gauge. These gauges are typically +/- 3% limits of error. The best resolution I see in bar is 2.0 bar and in PSIG 50. While hardly low pressure these ranges are not high or dangerous to anyone know legible who knows what they are doing. I would just insert a T adapter below the gauge and add a pressure sensor and make it one with an I2C output. Pressure sensors have come a long way.

You make no mention of uncertainty (accuracy) you want? I really do not see imagery as a viable solution.

Ron

I certainly agree that an I2C (or whatever) pressure sensor would be the best way to go. But say an image is 1600px x 1200px and the dial area of interest is made to fit in a 1000x1000 box. Then the indicator sweeps an arc of length approximately 2 x pi x 500 px x 75% (since full scale is about 270 degrees), or about 2300 pixels. But due to the discrete nature, say the arc is only 1500 pixels long. Then 3000 psi full range is about 2 psi per pixel. And at its tip, the indicator appears to be about 25 psi wide, and at the psi scale, is it about 40 psi wide. So it seems imagery analysis ought to get you at least as good a reading as an eyeball. But, an inline pressure sensor would make all that moot...

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.