Determine abnormal functioning of an industrial machine

I want to identify any problems that can arise in the operation of an industrial machine. I mounted an accelerometer on the machine arm and measured the values both in normal operation and in abnormal operation. Values can be seen in the attached picture. How can I determine periods of abnormal functioning and warn a worker about this? Is TTF a solution? The attached image was created with the Processing application and is based on the real data sent to Arduino through the serial port.
Thank you!

you should be able to look at the values that are below the average. then, if you have too many too frequently, you can set off the alarm

there is no time scale and you do not mention if there are other times when the signal goes low.

Timber Trade Federation, The Time Frequency ?

Start by looking at the graphs yourself. How can you tell if the machine is operating normally or not?
Once you know how to differentiate the two states start thinking about how to implement that as an algorithm?

If you cannot solve the problem yourself then you will not be able to write a program to solve it.
Having said that training a neural net might work but that is not an Arduino solution.

Is TTF a solution?

Did you mean FFT?

AWOL: Did you mean FFT?

YES! sorry, my mind was away!

there is no rule on the appearance of the signals in time, and they do not have a definite repeatability....

The Time Frequency is a Scottish electronic dance group :-)

FFT might work. The "normal" sections of your graph seem to have a regular beat and to go negative. The "abnormal" section is messy with only one negative spike. Maybe FFT will show clear differences in the frequency components.

Before delving into coding though I still think you should try solving the problem yourself. Take several samples of normal operation and several samples of abnormal operation analyze them and see if you can tell them apart in any way. If you want to use FFT there are online calculators that might let you try that.

ThePuiu: I mounted an accelerometer on the machine arm and measured the values both in normal operation and in abnormal operation.

As well as what other people have said, if you want useful advice it would be a good idea to

  • describe the machine,
  • show a diagram or a picture of it,
  • describe the causes of the abnormal function,
  • show where the accelerometer was placed,
  • describe what the accelerometer is intended to detect
  • describe the type of accelerometer - for example how many axes does it detect and which axis does your data come from
  • provide a link to the datasheet for the accelerometer
  • and everything else that would be relevant for someone trying to diagnose a problem.

...R

  • describe the machine,
  • I don't have much details about the machine (I only saw it twice). It is used in a small leather shoe factory and I think it has the role of cutting some shapes with a punching knife. The machine has a metal arm actuated by a motor at one end, and at the other end is the knife

  • show a diagram or a picture of it,

  • don't have...

  • describe the causes of the abnormal function,

  • normally, the piece of cutted leather must remain on the table, but sometimes it remain inside the knife, the next so on....etc the addition of several pieces of leather drive to the blocking of the machine

  • show where the accelerometer was placed,

  • i don't know right now...i think the best place is in the end of the arm, where knife is

  • describe what the accelerometer is intended to detect

  • when it does not work properly, the machine has some stronger vibrations

  • describe the type of accelerometer - for example how many axes does it detect and which axis does your data come from

  • MPU6050 - 3 axes

  • provide a link to the datasheet for the accelerometer

  • https://www.invensense.com/wp-content/uploads/2015/02/MPU-6000-Datasheet1.pdf

Unfortunately, I don't have a lot of information about the machine (is very old...maybe 40 years!) , but from what I've seen on the factory, that would be the only way to determine improper operation.

ThePuiu: - describe the causes of the abnormal function, - normally, the piece of cutted leather must remain on the table, but sometimes it remain inside the knife, the next so on....etc the addition of several pieces of leather drive to the blocking of the machine

I suspect measuring (peak) current will be easier and more effective. And may even have the side-effect of detecting a dull blade / failing motor.

ThePuiu: but from what I've seen on the factory, that would be the only way to determine improper operation.

Thank you for your extensive answer. I wish others were so responsive.

However if you don't have free access to the machine - which would enable you to take pictures or make a diagram - I reckon there is a very low probability of this succeeding.

From what you say, if the pieces of leather are getting stuck in the machine then counting them may be more useful.

...R

I think what Robin2 was hoping for, was a good description of what generates an abnormal condition and what causes it to cease being abnormal.

It would go a long way to help us to understand the nature of the problem.

in addition, is there an acceptable time for an abnormal condition to exist ? like hitting a bump with a car, it shutters, but recovers without any exterior intervention.

=================

based on the normal in your picture, the normal almost never goes negative. or does so very occasionally. Green being Abby Normal, and red being normal.

the abnormal has a high frequency of the signal going negative, almost a sine wave.
but, it also has a grouping of negatives that occur in what appears to be intervals.

assuming the machine has the negative spikes when it goes abnormal, you can set up a timers, like a debounce.

SCENARIO : HIGH FREQUENCY OF NEGATIVE PULSES

if you set the negative spike to start a timer.
Looking at your readouts, the abnormal produces a rather high frequency of negative spikes.

you could set the timer for the longest occurrence of a negative.

if (a negative occurs)
debounce to eliminate duration #1 from counting more than one occurrence

start timer
if duration #2 exists then
abby ++ ; // adds a 1 to the counter
if a second (or third) negative occurs while the timer is running abby will count up.

if abby >= 5 // picking a random value
then do something, lights, horn, etc

=====================
this is not code, just an idea.

if the machine self clears after 3 or 4 negatives, then you do not want to intervene.

if the machine has not cleared after 5, you might want to intervene.

what this does not have, but might be useful
is a third counter. one that counts how often the first counter got past, say. 3
that way if the cutter is getting dirty, or dull, or some such and it is setting itself up for a failure, you might light a yellow light to indicate a cleaning or blade change or some service is needed. the 'squeaky wheel' so to speak.

SCENARIO : HIGH FREQUENCY OF LONG DURATION OF NEGATIVE PULSES

if the problem is a negative pulse that lasts more than a brief time, duration #1 too long
then the could would look at the duration of the 'event'
the 'event' would be the entire duration of a negative that created the grouping of a negative pulse.

ardly:
The Time Frequency is a Scottish electronic dance group :slight_smile:

the repetitious vibrations have a similar pattern to dancers all being in step with each other.

Maybe we can get an audio clip with the graph and a video of the dancers of Scottish dance troupe TheTimeFrequency ?