Recognise specific pattern in timetable

I have a timetable logfile from an electrical motor.

This motor runs a variable rounds per minute, according to manual adjustment.

if i plot the entire timeline vs rpm i get what like in attached picture "1".

The intresting part of that data is where you see the rpm drops from ~2250 rpm to ~1300rpm. please see picture "2".

Is it possible to make a code that is recognising this specific pattern automatically, and isolate the time where this is happening? in this case it must be isolated from ~60s to ~100s to isolate this specifik range.

Thanks!

[Star Strider]

The findsignal function is the only method that I am aware of that would easily do this sort of pattern-matching.

[Mads Petersen]

Thank you that is a very cool function i did not know. Do you have experience with it? got some pattern recognition, but a lot of failures also. I dont think the matlab documentation i explaining clear enough what the different steps do.

i made a vector that look a lot like the pattern i am intrested in. see image3

got pattern script setup to find pattern in data. see here:

looks promising, but is finding more pattern than it should. following matlab documentation more. looks better

i still have a lot of errors. Specially when finding pattern of different log files, where typical the time and lograte (hz) is changed. Rpm pattern starting from ~2200 and down to ~1400 is always the same. see code:

subplot(2,1,1)

plot(data)

title('Data')

subplot(2,1,2)

plot(signal)

title('Signal')

figure

findsignal(data,signal)

dt = data;

dt(t>24&t<25) = 1900;

dt(t>25&t<26) = 800;

findsignal(dt,signal,'Metric','absolute')

findsignal(dt,signal,'TimeAlignment','dtw','Metric','absolute') %Let the x-axes stretch if the stretching results in a smaller absolute distance between the closest data segment and the signal.

dt(t>0&t<0.1) = 2400;

dt(t>0.1&t<0.12) = 1900;

Can yo explain more on how this is working?

Answer 1

[Mitch Lautigar]

After you grab the data, run a simple check to see if the rpm's drastically decrease. This can look something like the following

%rpms_val is the name i'm use for your "y" data points.

%rpms_time is the name i'm going to use for your "x" data points.

rpms_slope = [];

for i = 1:length(rpms_val)-1

rpms_slope(i) = (rpms_val(i+1) - rpms_val(i)) / (rpms_time(i+1)-rpms_time(i));

end

rpms_check = find(rpms_slope < -5) %-5 can be whatever tolerance you want.

from here, you can code in any flags you wish. You can even change the color of the data that is negative.

[Mads Petersen]

Thank you for comment. I keep getting this error using you script. am i missing something?

rpms_val = 12×1

103 ×

1.6937

2.1374

2.1364

2.0351

1.8646

1.6375

2.2471

0.8924

0.8503

0.8445

rpms_time = 12×1 duration

0 sec

4 sec

8 sec

12 sec

16 sec

20 sec

24 sec

28 sec

32 sec

36 sec

%rpms_val is the name i'm use for your "y" data points.

%rpms_time is the name i'm going to use for your "x" data points.

rpms_slope = []

for i = 1:length(rpms_val)-1

rpms_slope(i) = (rpms_val(i+1) - rpms_val(i)) / (rpms_time(i+1)-rpms_time(i))

end

rpms_check = find(rpms_slope < -5) %-5 can be whatever tolerance you want.

The denominator in matrix division for duration arrays must be a real scalar double value.

[Mitch Lautigar]

rpms_slope(i) = (rpms_val(i+1) - rpms_val(i)) ./ (rpms_time(i+1)-rpms_time(i))

forgot it was vector math. Should fix your issue.

Answer 2

[Les Beckham]

One approach without requiring the Signal Processing Toolbox:

load('answers.mat') % an approximation of your data
x = data(:,1);
y = data(:,2);
startidx = find(y>2000, 1, 'first'); % you may need to adjust this
% The following line may also need to be adjusted if the desired region is
% not strictly decreasing
stopidx = find(diff(y(startidx:end))>1, 1, 'first') + startidx - 1

stopidx = 26

plot(x,y)
hold on; 
plot(x(startidx:stopidx), y(startidx:stopidx), 'r.')

Answer 3

[Mads Petersen]

Thank you very much for all the nice answers. All of them are very intresting. i went with the diff. solution which i simple and works for now.

I thnik the findsignal is more intresting but required more fine tuning.

Thanks everyone