PCA-based Fault Detection for 2D Multivariate Process Data

Fault detection in a simple process using PCA and Kernel Density Estimation

% PCA-based Fault Detection
%
% Inputs: z0 [N x 2] = training data
% z1 [N x 2] = test data
% where: N = number of samples
%
% This code visualizes how PCA can account
% for multivariate data in fault detection.
% It also uses MATLAB's ksdensity for
% estimating the data PDF, so as to compute
% a T^2-based upper control limit.
%
% simpledata.mat has sample temperature [K]
% and concentration [mol/L] data from
% the contents of a simulated CSTR.
%
% The output are plots of the raw data,
% normalized data, and PCA projected data.
% Also, rings representing the T^2-based
% upper control limits at different user-
% defined confidence levels are plotted.
%
% You can edit confidence limits at Line 77.
%
% This code is intended for educational purposes.
%
% Load simpledata.mat and run the following:
% >> pcabased_fault_detection(train,test)