Detect and Diagnose Faults

Train classifiers or regression models for condition monitoring

To design an algorithm for detecting and diagnosing faults, you use condition indicators extracted from system data to train a decision model that can analyze test data to determine the current system state.

When designing your algorithm, you might test different fault detection and diagnosis models using different condition indicators. Thus, this step in the design process is likely iterative with the step of extracting condition indicators, as you try different indicators, different combinations of indicators, and different decision models.

For an overview of the types of models you can use, see Decision Models for Fault Detection and Diagnosis

Functions

expand all

Feature Selection

`pca`	Principal component analysis of raw data
`pcares`	Residuals from principal component analysis
`sequentialfs`	Sequential feature selection using custom criterion
`fscnca`	Feature selection using neighborhood component analysis for classification
`tsne`	t-Distributed Stochastic Neighbor Embedding

Statistical Distribution Fitting

`ksdensity`	Kernel smoothing function estimate for univariate and bivariate data
`histfit`	Histogram with a distribution fit
`coxphfit`	Cox proportional hazards regression
`ztest`	z-test

Machine Learning

`fitcsvm`	Train support vector machine (SVM) classifier for one-class and binary classification
`fitcecoc`	Fit multiclass models for support vector machines or other classifiers
`fitcknn`	Fit k-nearest neighbor classifier
`fitclinear`	Fit linear classification model to high-dimensional data
`fitcnb`	Train multiclass naive Bayes model
`fitctree`	Fit binary decision tree for multiclass classification
`fitckernel`	Fit Gaussian kernel classification model using random feature expansion
`kmeans`	k-means clustering
`mle`	Maximum likelihood estimates
`TreeBagger`	Create bag of decision trees

Regression with Dynamic Models

`nlarx`	Estimate parameters of nonlinear ARX model
`ssest`	Estimate state-space model using time-domain or frequency-domain data
`arx`	Estimate parameters of ARX, ARIX, AR, or ARI model
`armax`	Estimate parameters of ARMAX model using time-domain data
`ar`	Estimate parameters of AR model or ARI model for scalar time series
`forecast`	Forecast identified model output
`translatecov`	Translate parameter covariance across model transformation operations

Control Charts

`controlchart`	Shewhart control charts
`controlrules`	Western Electric and Nelson control rules
`cusum`	Detect small changes in mean using cumulative sum

Changepoint Detection

`findchangepts`	Find abrupt changes in signal
`findpeaks`	Find local maxima
`pdist`	Pairwise distance between pairs of observations
`pdist2`	Pairwise distance between two sets of observations
`mahal`	Mahalanobis distance
`segment`	Segment data and estimate models for each segment

Topics

Decision Models for Fault Detection and Diagnosis

Use condition indicators extracted from healthy and faulty data to train classifiers or regression models for detecting and diagnosing faults.

Fault Diagnosis of Centrifugal Pumps Using Steady State Experiments

Use a model-based approach for detection and diagnosis of different types of faults in a pumping system.

Fault Diagnosis of Centrifugal Pumps Using Residual Analysis

Use a model parity-equations-based approach for detection and diagnosis of faults in a pumping system.

Multi-Class Fault Detection Using Simulated Data

Use a Simulink model to generate faulty and healthy data, and use the data to develop a multi-class classifier to detect different combinations of faults.

Analyze and Select Features for Pump Diagnostics

Use the Diagnostic Feature Designer app to analyze and select features to diagnose faults in a triplex reciprocating pump.

Fault Detection Using an Extended Kalman Filter

Use an extended Kalman filter for online estimation of the friction of a simple DC motor. Significant changes in the estimated friction are detected and indicate a fault.

Fault Detection Using Data Based Models

Use a data-based modeling approach for fault detection.

Detect Abrupt System Changes Using Identification Techniques

Detect abrupt changes in the behavior of a system using online estimation and automatic data segmentation techniques.

Chemical Process Fault Detection Using Deep Learning

Use simulation data to train a neural network than can detect faults in a chemical process.