Hello guys, i am actually doing my dissertation based on an predictive fault maintanace using arduino and some sensors( temperature, vibration, current, sound,...) and machine learning. I am instructed to make a prototype. The implementation part i can manage i guess. However, i dont know how to test the prototype on a machine for all of these sensors. For example, how to test for the temperature of a machine. Any ideas?
Use a temeperature sensor that gets good thermical contact with the motor.
It would help you if you posted what type of machine we are looking at and what sensors you have chosen. Also you need to define the points that need testing.
For each of these things, there are Arduino-compatible modules or sensors to measure quantities of interest.
To use Arduino with any of those sensors, an internet search like "arduino measure temperature" will find getting started tutorials and forum entries discussing problems and solutions.
Your two or more topics on the same or similar subject have been merged.
Please do not duplicate your questions as doing so wastes the time and effort of the volunteers trying to help you as they are then answering the same thing in different places.
Please create one topic only for your question and choose the forum category carefully. If you have multiple questions about the same project then please ask your questions in the one topic as the answers to one question provide useful context for the others, and also you won’t have to keep explaining your project repeatedly.
Repeated duplicate posting could result in a temporary or permanent ban from the forum.
Could you take a few moments to Learn How To Use The Forum
It will help you get the best out of the forum in the future.
Thank you.
Well doing a dissertation means you have already graduaded as an engineer.
Of course you are free to ask here a lot of different kinds of questions but:
Don't you think that doing a dissertation means using scientific methods to do research in a new field or a new aspect ?
And that doing this goes pretty far beyond asking in a public forum questions like
"what basic concept do you recommend?"
You haven't posted the exact subject of your dissertation in detail but from such a dissertation I would expect that the author will discuss the pro's and con's of different concepts. That the author will read quite a lot of scientific literature about his subject and judge different approaches by a self-defined list of criteria.
Machines can be very very different. My expectation is that you can't do one code for them all. There will be machine-specific things that "announce" a future fail of a part of the machine. And finding these signs that indicate a future fail will require evaluating existing machinedata of all kinds because you don't know yet what to look at.
I am far away from beeing an expert about machine learning. What I have read about it by scratching the surface one possible approach are neuronal networks in beeing able to discover hidden connections between different machinedata through training the neuronal network with example-data.
Using arduino is somehow still unspecific "arduino" could mean a "classical arduino uno with very limited memory
Arduino Uno: 32 kB of flash 2 kB of RAM 8 bit at a 16 MHz-clock
which surely has far less computational power than
a still Arduino-IDE programmable
Teensy 4.1: 8000 kB of flash-memory and 1000 kB of RAM 32 bit at 600 MHz-clock
I would ask manufacturers of standard machine-components like roller bearings, slide bearings, linear guide-systems, electric motors, pneumatic and hydraulic cyclinders
What are typical early signs of a future "wornout point" or a future component failure.
I can fully understand that this is very difficult. The reason is that machines can be so very very different. To one machine it might be acceptable that the motor heats up to 120 degrees and has a limited lifetime of 500 working hours because adding a cooling system is more expensive than changing the motor or adding a cooling system is even completely impossible for the reason "cooloing system needs to much space"
For another machine motor-temperature rising more than 5 degrees over environment temperature might be the very early sign for overheating within the next hour with the motor catching fire.
How should any kind of microcontroller be able to predict in a correct way with such great variability of possible parameters?
So one approach might be to install microcontrollers with sensors measuring machine-parameters that seem to promise delivering data with which you are able to predict future developments
As a first step to collect data of multiple machines of the same or similar type and then when wornout or a failure happends to look out for what parameters did take what kind of development that lead to wornout or failure.
And then as a second step to monitor these parameters for unusual developments or running outside
"normal parameters"
One tool for analysing might be calculating change-rates ( slope-values) of parameters
How fast does a parameter change over time
best regards Stefan
1 Like
I am not familiar with using machine learning algorithms. However I have some experience with machine failures and predictive fault maintenance. Depending on the failure mode, different sensors will give the earliest possible indicator. So, I would use as many different sensor types and locations as possible to ensure that my sensor array will be most likely to detect the first sensor indication of the impending failure.
As for the model to use, neither of the options you mentioned would be appropriate. Each sensor type and location should be evaluated individually and the entire sensor array will need to be evaluated as a whole data set.
Good luck with your dissertation. I hope you learn a lot while you are preparing for it.
1 Like
Hello, i have another query about the machine learning. So, machine learning is efficient when you have a dataset already available which already has an output. For example a historical dataset for a fault prediction system consists of its sensor values and its failure(fail or no fail). However, since im not using historical data and using real time sensor values, how do i implement this if i have only sensor values in my dataset. Help me please.
Then you should know that a set of good/fail sensor values is required to configure the system. You should find out how fake "fail" states can be generated for teaching.
Stop the bot.
Your other topic on the same subject deleted.
Please do not duplicate your questions as doing so wastes the time and effort of the volunteers trying to help you as they are then answering the same thing in different places.
Please create one topic only for your question and choose the forum category carefully. If you have multiple questions about the same project then please ask your questions in the one topic as the answers to one question provide useful context for the others, and also you won’t have to keep explaining your project repeatedly.
Repeated duplicate posting could result in a temporary or permanent ban from the forum.
Could you take a few moments to Learn How To Use The Forum
It will help you get the best out of the forum in the future.
Thank you.
DS18B30 gets better reviews than the obsolete cheap DS18B20.
Both are I2C devices, you get the value as text.
Have you heard of "heat guns" that blow hot air onto something? Heat can be directed to any part of a machine.
This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.