How to filter a signal?

Hello everyone,

I've this data (3 forces, a power and a current), that I've imported from a .tdms file.

I need to detect the peaks on the lower peaks which represent the contact between a cutting tool and a workpiece. After escluding the initial and the final zone where the engagement it's not constant, using "getcursormode", I've used this part of code:

avg = mean(-Matrix_Data(trim(1,2)+pre_engag:trim(1,3)-post_engag,2));       % -Matrix_Data for detecting only negative peaks
[pks,locs] = findpeaks(-Matrix_Data(trim(1,2)+pre_engag:trim(1,3)-post_engag,2),'MinPeakHeight',avg,'MinPeakProminence',avg/2.5,'MinPeakDistance',19);

Basically my problem is that the detected peaks are not as perfect as I'd like.

I need to pre-filter the signals for having a better waveform to anlyse later. Any suggestions? I haven't the particular need of cutting any frequence but I'd have a filter that would allow me to follow existing data very precisely by eliminating any very close peaks. It's also possible to change the findpeaks parameter but I've already tried all the possible combinations.

Thank you some much for anyone will help me.

[Yazan]

The peaks of which column of your data do you need to detect?

[riccardo favaro]

column 2

[Yazan]

Can you indicate which peaks we're talking about here?

[Star Strider]

The peaks are more apparent in the filtered signnal in Column #3, as I filtered them.

Answer 1

[Star Strider]

First, the data are really noisy, and exhibit broadband noise that will be impossible to eliminate with a frequency-selective filter. The posted code fails because the variables are not provided.

I mamaged to use the Savitzky-Golay filter on the first 3 signals, however it is not doing much for the last 2.

LD = load('data.mat');
Matrix_Data = LD.Matrix_Data;

Ts = mean(diff(Matrix_Data(:,1)));
Tsd = std(diff(Matrix_Data(:,1)));

sm = Matrix_Data(:,2:end) - mean(Matrix_Data(:,2:end));
np = size(Matrix_Data,2)-1;

figure
for k = 1:np
    subplot(np,1,k)
    plot(Matrix_Data(:,1), sm(:,k))
    grid
end
sgtitle('Original Data')


figure
tic;
for k = 1:np
    fprintf('\tk = %d\n',k)
    subplot(np,1,k)
    smfilt = sgolayfilt(sm(:,k), 3, 501);
    plot(Matrix_Data(:,1), smfilt)
%     plot(Matrix_Data(:,1), sm(:,np))
    grid
    toc
end
sgtitle('Savitzky-Golay Filtered Data')

L = size(Matrix_Data,1);
Fs = 1/Ts;
Fn = Fs/2;
NFFT = 2^nextpow2(L);
sm = Matrix_Data(:,2:end) - mean(Matrix_Data(:,2:end));
FTs = fft(sm,NFFT)/L;
Fv = linspace(0, 1, NFFT/2+1)*Fn;
Iv = 1:numel(Fv);

np = size(FTs,2);

figure
for k = 1:size(FTs,2)
    subplot(np, 1, k)
    plot(Fv, 20*log10(abs(FTs(Iv,k))*2))
    grid
    ylim([-150, 0])
end
sgtitle('Fourier Transforms of Original Data')

I have no idea what peaks need to be detected, however the result of the Savitzky-Golay filtering step should make that a bit easier. I did not save the ‘smfilt’ results to a matrix, so it might be necessary to add that if desired:

    smfilt(:,k) = sgolayfilt(sm(:,k), 3, 501);
    plot(Matrix_Data(:,1), smfilt(:,k))

The rest of the code is unchanged.

EDIT — (9 Aug 2021 at 16:15)

Added the ‘Savitzky-Golay Filtered Data’ plot this code produced —

.

[riccardo favaro]

it already gives me an error in line 3:

"Array indices must be positive integers or logical values.

Error in 'name' (line 3)

Ts = mean(diff(Matrix_Data(:,1)));"

[Yazan]

Make sure to clear previous variables from the workspace. The code works just fine.

[Star Strider]

@riccardo favaro — I cannot improve on Yazan’s suggestion.

@Yazan — Thank you!

.

[riccardo favaro]

yeah, now it works. But actually it doesn't solve my entire problem. Nevermind.

Thank you all

[Yazan]

To get better help or suggestions from the community, you need to improve your description of the problem at hand in my opinion.