How to split data points into new numeric matrixs based on peaks

I am trying to analysis finish dimension data of a machined part and create a control chart. However, the data is impacted by tool wear and must be analysed using linear regression. To do this I have to split data points between tool changes which can be seen in the data (based on troughs and peaks). See image of scatter plot below.

Obviously it would be time consuming to split the data in excel manually for 4 data sets each with 8000 points. I am hoping to code a way to identify the change between tool changes and place that data segment into a new numeric matrix and test for regression. For example dimensions will climb from 11.955 to 12 at which will drop back down to 11.949 after a tool change.

I know of the

   findpeak(data)

But not sure how to split the data and place it into a new matrix.

How can I achieve this in MATLAB?

[Mark Saad]

If your goal is to make a separate matrix of data after every peak, you could try to first get the number of peaks:

[pks, locs] = findpeaks(data);
numpeaks = length(pks);

Then make an array for each peak:

start_point = 1;
for i = 1:numpeaks
    end_point = locs(i);
    for j = start_point:end_point
        split_data = data(start_point:end_point, :);
    end
    start_point = end_point + 1;
end

start_point and end_point set the range of data that you want to collect, so initially, start_point begins at the first point and data is collected until the first peak, which is end point. Then in the second iteration, start_point is now the point after the first peak, and end_point is now at the second peak, and so on.

This makes a new array through each iteration, so you could just make a larger variable before the loops to store all of the split data arrays.

[Aaron Holiday]

The find peaks is too sensitive, It returns too many results. I'll need a way to test for a extreme change in y values, signifies tool changes. For example, a drop from 12.10 to 11.50. Any ideas on how to do this?

Answer 1

[Guillaume]

You may need to tweak the options of findpeaks for you data (and finding the minimums by inverting your data may be more reliable), but once you've got the location of the peaks it trivial to split your vector into subvectors stored in a cell array:

 [~, locs] = findpeaks(data);  %maybe findpeaks(-data), or maybe use options
 splitdata = mat2cell(data, 1, diff([1, locs, numel(data)+1])); %assumes data is a row vector

[Aaron Holiday]

The find peaks is too sensitive, It returns too many results. I'll need a way to test for a extreme change in y values, signifies tool changes. For example, a drop from 12.10 to 11.50. Any ideas on how to do this?

[Guillaume]

As I said, you may need to tweak the options of findpeaks so that it works with your data and it may be that finding the peaks of the negative of your signal may also be more efficient.

Otherwise, you could try detecting large gradients in your data. Possibly:

locs = abs(gradient(data)) > 0.02

[Aaron Holiday]

Ok, so this works. I get two 1's at each max peak and max trough at each tool change. However, your splitdata code throws an error. Why is mat2cell being used? I've never used MATLAB to this level before.

Error using horzcat Dimensions of arrays being concatenated are not consistent.

Error in Data (line 5)
splitdata = mat2cell(LatheB_BPTO, 1, diff([1, locs, numel(LatheB_BPTO)+1]));

[Guillaume]

locs is a column vector. Hence you need to use semicolons for concatenation:

splitdata = mat2cell(LatheB_BPTO, 1, diff([1; locs; numel(LatheB_BPTO)+1]));

If data is also a column vector:

splitdata = mat2cell(LatheB_BPTO, diff([1; locs; numel(LatheB_BPTO)+1]), 1);

[Aaron Holiday]

Ok, Im not sure about this. However, I have a column vector that list the start and end points of each segment. I got this using the find function on the gradient results. All i really need is to either look at a certain range of indices to perform a mathematical function on or create new column vectors for each range.

For example, the first tool change in the LatheB_BPTO data is between index 88-246, the next is 247-359. etc... I need to perform a regression analysis on each of these. that is the end goal.

Answer 2

[Walter Roberson]

blocks = mat2cell(data, 1, diff(find([true,diff(data)<0,true])));