Hello dear friends, well actually im a newbie in node-red and still learning, however im getting a project which going to use node-red machine learning.. for now im kind of clueless but i would explain regarding my project so i could get help from the members in the community as the community is really helpfull !
basically im taking an accelerometer data from smartphone trasferred to MQTT cloud server and to node-red ..which im using node-red to store data in form of csv file..and now what am i try to do is to get the data being classified using node-red machine learning ..im trying to classified the data obtain as good or faulty (the phone is taking reading from machine vibration of machine in form of accelerometer)..the thing im trying to do is the node-red will classified the data in form of Hz as faulty or good..for example 10hz is good data but then the frequency drop to 8hz and then we know that there is problem and classified as faulty
i hope i explain well(im sorry if not)...is there any example or tutorial to do node-red machine learning..or any material or video i could study to build one..
For machine monitoring, you may also want to look at time domain waveform and acceleration rms amplitude change, true peak change, kurtosis value, etc.
What kind of application do you have with variable frequency?
If you know the frequency threshold already, then you don't need machine learning. You just need a simple function or a switch node to label the machine conditions.
helllo there sir davidz,
for now i just obtained the data from MQTT cloud broker as below
and transfered to node red(credit to node-red community that help me out)
for now this is the only data i have, im not sure how to make it to frequency ..and im looking for general use ..but for experiment im using 1000rpm rotating machine ... its like turn on the device and the data is transfered untill reaches node-red machine learning will decide the condition... then i will make a faulty condition on the machine(such as add load etc) and the node-red machine learning will detect the abnomalities from the frequency and give warning..this is for now..
please suggest or give advice really appreciate it, im sorry if i did not answer some question due to limited data obtained and knowledge
To convert the time domain data to frequency, you need FFT.
From FFT curve, you can look at the peak position and tell if the machine is working well or not. Of course this is only the basic level vibration analysis.
For 1000rpm, you are looking at 16.67Hz frequency. You need at least 50Hz sampling rate to tell if the machine is working well for the basic analysis. What is your sampling rate at the smart phone?
well i did get ur point ..last time the data is analyse through matlab , it has preprocessing (to remove outliers) and then went through FTT process...but its not real-time data..the data need to be saved first the only can be analyse(if im not mistaken)
does node-red able to do FTT?in real time manner
well for sampling rate i have done few calculation ,which i think accelerometer from phone is good enough..the frequency is around 50Hz which generate around 2000+ data perminutes (tested )
Since the machine frequency is probably (my previous calculation,which maybe wrong)25Hz(for 1k RPM) and my sensor is capable up to 50Hz .it should capture the data quite sharp and should produce less noice and more accurate
There is a Node-RED FFT node but it is way outdated.
So you may have to write your own FFT code for this part.
A smartphone can do a quick diagnosis for low-frequency machine issues.
thanks for the ftt node, yes i just get some information and it is quite inaccurate..or does it considerable for 50hz accelerometer to capture 16.67hz machine?(which i think maybe the data is quite sharp and less noise)
the other option that i have now is 1D CNN( 1 dimensional convolutional neural networks) for time series classification but its developed in python..which is post processing .. because i think 1D CNN is still not available in node-red..or am i wrong? ..do u have any suggestion for this issue
That FFT node is pretty preliminary. To get better results, you may look into some library FFTs such as:
Actually I just found a machine learning node. It is also kind of outdated but you may try it out:
Thanks alot! i guess i will look into this one firstand get some idea
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.