Time Series Data Analysis: Logical segmentation of phases of data
3 visualizaciones (últimos 30 días)
Mostrar comentarios más antiguos
I have high speed mixing data of various materials and during this mixing phase, the materials go through distinct phase changes. The super fancy mixer I have controls the rotor speed and trends the power required to achieve that speed.
During the last phase, you can see from the data attached, there appears to be 3 distinct phases in the required rotor power. The first phase, the material starts to clump, so the rotor requires a little more power, and as the material breaks up, we see a distinct decrease in power. As this is happening, the temperature is going up and at some point, the material densifies and granulates. When this happens, we see a sharp increase in the rotor power. This is very consistent but depending on what else I add to the mix, I can get different slopes, as some things delay granulation and other things speed them up.
So basically I have 100s of experiments that show these three distinct phases but all have very different time lengths. What I am trying to do is fit these three phases with very simple linear or polynomial fits so I can put more basic numbers into a design of experiment.
Since I haven't had a lot of time to look around I thought I would just throw this out there.
Attached is the data and some code that simply creates X linear segments and I am hoping some of you have some ideas on how I could create more logical code that would be able to find these three segments, for lack of a better term, that would then allow me to extract some more basic information.
Data and code attached. Sorry I haven't added any comments but you should be able to run this and see what I am trying to do.... Thanks
clc; clear all;
load('EL1_Data.mat')
tic
% dSeg is how many brackets we are going to make
dSeg = 6;
% Create segments to linearly fit
brack = linspace(min(timeLast),max(timeLast), dSeg);
% Cell to store the different brackets
iX = cell(dSeg-1);
dataFits = zeros(dSeg-1,2);
%errors = struct();
figure(1)
plot(timeLast,rPowLast,'.')
hold on
for i = 1:dSeg-1
iX{i} = find(timeLast >= brack(i) & timeLast <= brack(i+1));
[p,s] = polyfit(timeLast(iX{i}),rPowLast(iX{i}),1);
dataFits(i,:) = p;
errors(i) = s;
[yFit, yErr] = polyval(p,timeLast(iX{i}),s);
plot(timeLast(iX{i}),yFit,'-')
hold on
end
hold off
xlabel('Time (s)')
ylabel('Rotor Power (W)')
toc
1 comentario
dpb
el 2 de Abr. de 2015
Look at "piecewise linear regression". I demonstrated/provided code for the case of two segments not very long ago here (but, one of the problems Answers hasn't solved but was supposed to) I can't seem to find it easily again at the moment.
I've not, at least recently, moved beyond the two segments but it's just additional conditions on the model.
Respuestas (0)
Ver también
Categorías
Más información sobre Statistics and Machine Learning Toolbox en Help Center y File Exchange.
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!