Main Content

Diagnostic Feature Designer

Interactively extract, visualize, and rank features from measured or simulated data for machine diagnostics and prognostics

Description

The Diagnostic Feature Designer app allows you to accomplish the feature design portion of the predictive maintenance workflow using a multifunction graphical interface. You design and compare features interactively and then determine which features are best at discriminating between data from nominal systems and from faulty systems. The most effective features ultimately become your condition indicators for fault diagnosis and prognostics.

Using this app, you can:

  • Import measured or simulated data from individual files, an ensemble file, or an ensemble datastore that references files external to the app.

  • Interactively visualize data to plot the ensemble variables you import or that you compute within the app. Group data by condition label in plots so that you can clearly see whether member data comes from nominal or faulty systems.

  • Derive new variables such as time-synchronous averaged signals or order spectra. The app executes processing on all ensemble members with one command.

  • Generate features from your variables and visualize their effectiveness using histograms. Features include signal statistics, nonlinear metrics, rotating machinery metrics, and spectral metrics.

  • Rank your features to determine which ones are best at discriminating behavioral differences in the data.

    • Use supervised ranking with labeled features to determine which features are most likely to discriminate between nominal and faulty behavior.

    • Use unsupervised ranking when your data has no condition variables or labels to determine which features exhibit the best clustering with other features and are most likely to indicate different fault or operating conditions.

    • Use prognostic ranking with features extracted from run-to-failure data to determine which features are most likely to indicate remaining useful life (RUL).

  • Export your most effective features directly to Classification Learner for more insight into feature effectiveness and for algorithm training.

  • Generate code for the features you choose so that you can reproduce, customize, and automate the feature computations in a MATLAB® function.

To get started with the app, you must have data from your systems available in your MATLAB workspace. For information about organizing your data for import into the app, see Organize System Data for Diagnostic Feature Designer.

For more information about condition indicators for predictive maintenance, see Condition Indicators for Monitoring, Fault Detection, and Prediction.

Diagnostic Feature Designer app. A toolstrip is on the top. The data browser is on the left. A set of histograms that illustrate group separation is in the main plot area.

Open the Diagnostic Feature Designer App

  • MATLAB toolstrip: On the Apps tab, under Control System Design and Analysis, click the app icon.

  • MATLAB command prompt: Enter diagnosticFeatureDesigner.

Parameters

Feature Designer Tab

Initiate a new app session by importing source data into the app from your MATLAB workspace. You can import data from tables, timetables, cell arrays, or matrices. You can import data from a single source that combines the data of multiple ensemble members or import individual ensemble members from separate sources. You can also import an ensemble datastore that contains information that allows the app to interact with external data files. Your files can contain actual or simulated time-domain measurement data, spectral models or tables, variable names, condition and operational variables, and features you generated previously. Diagnostic Feature Designer combines all your member data into a single ensemble data set. In this data set, each variable is a collective signal or model that contains all the individual member values.

For more information about importing data, see Import Data into Diagnostic Feature Designer.

For more information about terms related to data ensembles, see More About.

For more information about organizing your data for import into the app, see Organize System Data for Diagnostic Feature Designer.

Specify default plotting options for all plots that you generate during your app session. You can set these options before you generate your first plot, or at any time during your session. New settings apply only to plots that you generate after setting the options and not to plots that you generated earlier. You can temporarily override the Plot Options settings for individual plots without changing your specified defaults for subsequent plots. When you click Plot Options, you open a dialog that allows you to set options in the following panes.

  • General — These options apply to all signal and spectrum plots.

    • Group by — Group data by a condition variable label. The app uses color to distinguish label groups. For example, if your condition variable is faultCode with labels healthy and degraded, the app uses one color for member data with the healthy label and another color for member data with the degraded label

    • Number of curves — Specify the number of members to plot. Set this option when you have a large number of ensemble members and you want to plot only a subset of the members. Using this option reduces the plotting time and allows you to assess individual member behavior more easily.

  • Spectrum — These options apply only to spectral plots.

    • Number of peaks to mark — Specify the number of peaks to mark. Set this option to limit the number of spectral peaks that are marked to highlight only the most significant peaks.

  • Ensemble Summary — These options apply only to the ensemble summary plot, which is a special plot that displays the mean and standard deviation of the ensemble as a whole.

    • Number of standard deviations — Specify the number of standard deviations that the ensemble summary plot displays.

    • Show min and max boundaries — Specify whether to display the boundaries of the actual minimum and maximum values of the ensemble.

Generate a plot of an ensemble variable or feature table. To generate a plot, first select a variable or feature table from the Data Browser. The plot gallery shows icons for the compatible plot types. The following table describes the plot types for each type of selection.

Input TypePlot TypeDescriptionCustomize With
SignalSignal TraceEnsemble signal data plotted by time or another independent variable that does not represent frequency.Signal Trace Tab
Ensemble SummaryMean, standard deviation, and min/max boundary for the ensemble as a whole.Ensemble Summary Tab
SpectrumPower SpectrumEnsemble signal power plotted by frequency.Power Spectrum Tab
Order SpectrumEnsemble signal power plotted by order, which is the ratio of a specific frequency to the main rotational frequency.Order Spectrum Tab
Feature TableHistogramFeature effectiveness, as visualized by a bar chart with color coding for the condition label. Effective features separate conditions cleanly.Histogram Tab
Feature Table ViewTable containing feature values and their condition labels for each ensemble member.N/A
Feature TraceFeature values for each member. This plot is especially useful for prognostic features (for RUL) computed from frame-based data.N/A

Specify a frame policy when you want to perform data processing on sequential segments of a signal rather than the full signal at once. A frame policy consists of a frame size and a frame rate. The frame size is the interval over which the frame data is collected. The frame rate is the time interval between frame start times.

For more information on frame-based processing, see Data Handling Mode and Frame Policy.

Specify options when you want to modify one or both the following settings.

  • Independent Variable — Independent variable (IV) to use. When you import your data, you can specify more than one independent variable for a signal. For example, if your signal is time based, you might want to also have an independent variable for sample index. After you complete the import, you can change the independent variable that the app uses for a specific plot or computation. When you select Options > Independent Variable, the app displays a list of the available independent variables. Your selection changes the IV of all the applicable signals or spectra. For more information on specifying an alternate IV, see Specify Sample Index as Alternate IV in Import Data into Diagnostic Feature Designer.

  • Use Parallel Computing — Process ensemble members in parallel. Using parallel computing can significantly decrease processing time for large ensembles. This option is available only when you have Parallel Computing Toolbox™ installed and licensed.

Select options for processing your data into new signals. Use these new signals as inputs to other processing options or as inputs to feature generation. Most processing options operate on each ensemble member. You can also perform ensemble-level processing to view how the ensemble behaves as a whole. Each option selection opens a new tab for your specifications. Selection of an option also opens a general Data Processing tab if that tab is not already open. The Data Processing tab provides information about the input signal.

To specify a signal to process, select a variable from the data browser prior to selecting the data processing option. To change the signal after opening the option tab, close the option tab and select a new signal either in the data browser or from the Signal list in the Data Processing tab.

For more information about the processing options and the parameters that you can set for each option, see:

Compute features in the time domain. Signal Features apply to any signals. Rotating Machinery Features are specialized metrics that apply to gearing. Nonlinear Features provide metrics that characterize chaotic behavior in vibration signals. Each selection opens a dialog box for your source signal and feature specifications.

To specify a signal source for your features, select a signal variable from the data browser prior to selecting the time-domain features option. To change the signal after opening the option tab, close the option tab and select a new signal either in the data browser or from the Signal menu in the Time-Domain Features tab.

For more information about time-domain features options and the parameters that you can set for each option, see:

Compute features in the frequency domain. Spectral Features are general metrics that apply to any spectrum, such as the peak amplitude across the full specified frequency range. Bearing Faults Features, Gear Mesh Faults Features, and Custom Faults Features are specialized metrics for rotating machinery that focus on spectral behavior within specific fault bands that bound characteristic frequencies, where faults can occur, of the components of the system.

For more information about the frequency-domain features, see

Open the feature ranking tab to perform supervised, unsupervised, or prognostic ranking for the feature table that you select. For more information, see Feature Ranking Tab.

Export features, or your entire data set, to use them or share them outside of the app. Generate code to reproduce your feature computations in a MATLAB function.

  • For feature export, both options open a list of features.

    • If you have not yet ranked your features, the app sorts this list by name, and marks all features by export by default. You can refine the selection if you want to export only specific features.

    • If you have ranked your features, the app sorts this list by your Sort by specification in the Feature Ranking Tab. Use Select top features to export only the most highly ranked features, based on the number of features that you specify. You can change the sorting order to alphabetical by selecting Name in the Features sorted by list. With either sorting order, you can individually select or clear features to export.

    When you export to the MATLAB workspace, you can use command-line techniques with the features. When you export to Classification Learner, you open a Classification Learner session that uses your selected features as input.

  • For code generation, the first option, Generate Function for Features, lets you generate MATLAB code with a simple set of specifications for feature table, ranking algorithm, and number of features. Use this option when you want to generate code for features based solely on ranking, or when you want to generate code for all your features.

    The second code generation option, Generate Function for..., allows you to customize your selection of items to include in the function. For example, you can filter your selection based on criteria such as input or output text. You can include signals and spectra that are not used in the features you select. Selecting Generate Function for... opens a selectable list of all the signals, features, and ranking tables that you have generated. Generate Function for... also opens the Code Generation tab, which provides filtering capability for the list. Use a filter to view only the items that meet the filter criterion. You can use different filters to select the outputs you want. To review all your selections regardless of filter, click Sort by Selection. This option lists all the available outputs with items that you selected on top. For more information, see Code Generation Tab.

    If you have specified frame-based data (see Options), clicking Generate Function for... first opens a list with selections for the frame specifications that you have used. The items in your generated code must either all operate on the full signal or all use the same frame specification.

    For more information on how to generate code in the app, see Automatic Feature Extraction Using Generated MATLAB Code and Generate a MATLAB Function in Diagnostic Feature Designer.

For more information about the Export options, see:

Signal Trace, Power Spectrum, and Order Spectrum Plot Tabs

Use the Panner to focus on data segments in the x-axis range that you specify and to change the plot scale. The Panner provides a strip plot beneath the main plot. To focus on a section of the main plot, move the handles. To change the scale of the plot, select one of the options in Scale.

Use Group By and Number of Curves when you want to display ensemble label grouping and limit the number of members that you plot. Use Peaks to Mark to specify how many spectral peaks to highlight. The default values correspond to the settings in Plot Options. When you change these settings in the plot tab, you change them only for the current plot. For more information on these parameters, see Plot Options.

Use Normalize Y Axis when you are plotting multiple variables and want to view the variables on the same [-1, 1] scale. The relative signal or spectrum amplitudes within a variable do not change.

In a signal or spectrum plot, you highlight an individual member by positioning your cursor on the member trace. Select Show Signal Information or Show Spectrum Information to display both the variable member that you highlight and the condition label for that member in the lower right corner.

Specify how to plot multiple variables together.

  • Select to create a single plot that overlays all traces and uses a single y-axis scale.

  • Clear to create separate plots displayed vertically, each with a unique y-axis scaling.

Select Data Cursors to display values of key points in your signal. Data Cursors are horizontal and vertical bars that you position over a point of interest, such as a peak value. The cursors display the x and y positions. To display the distance between the cursors, select Show Signal Information. To lock the bars so that they move together, select one of the Lock Spacing options.

Histogram Tab

Click Select Features to open a selectable list of features to plot. Use Select Features, for example, when you have generated many features but you want to focus on a subset in a single plot panel.

Select the condition variable to base feature histograms on. The feature histograms use color to visualize the separation of data groups with different labels for that condition variable.

Example: faultCode

Specify histogram resolution using Bin Width, Bin Method, Number of Bins, and Bin Limits. The bin settings apply to all the histograms for the feature table.

The bin settings are not independent. The app histogram algorithm uses an order of precedence to determine what to use:

  • The Binning Method is the default driver for the bin width.

  • A Bin Width specification overrides the Binning Method.

  • The bin width and the independent Bin Limits drive the number of bins. A Number of Bins specification has an effect only when the value of Group By is none.

    For more information on interpreting and customizing histograms, see Generate and Customize Feature Histograms.

Feature Ranking Tab

Select a supervised classification ranking method to assess how effectively each feature separates data with different condition labels. If you have already ranked your features, you can rank again with a different method and display the resulting rankings together. Each method uses a different statistical approach.

The menu differentiates between two-class and multiclass ranking methods.

  • Two-Class Methods — Use when your condition variable (CV) has only two labels, such as healthy and faulty. The default value for two-class methods is T-Test.

  • Multiclass methods — Use when your condition variable has two or more labels, such as healthy, faultCode1, and faultCode2. The default value for multiclass methods is One-way ANOVA

The default ranking method for two-class condition variables, T-Test, is the simplest method, as it only considers whether the means of the two labeled groups are equal or not. T-Test is primarily useful for identifying ineffective features to discard.

The table lays out the influence of specific criteria on ranking-method selection.

CriterionRanking Method
Condition Variable Type
  • Multiclass CV — One-way ANOVA, Kruskal-Wallis

  • Two-Class CV — T-test, Entropy, Bhattacharyya, Wilcoxon, ROC

Feature Scoring Criterion
  • Mean Difference — T-Test (primarily for discarding ineffective features)

  • Distribution Overlap — All others

Distribution Shape
  • Gaussian — T-Test, Entropy, Bhattacharyya, One-way ANOVA

  • Non-Gaussian — ROC, Wilcoxon, Kruskal-Wallis

Desired Method Basis
  • Hypothesis Test — T-Test, One-way ANOVA, Wilcoxon, Kruskal-Wallis

  • Distance Measurement — Entropy, Bhattacharyya, ROC

Selecting a method activates a new tab with a name that matches the ranking method. For more information on this method-activated tab, see Ranking Method Tabs.

For more information on the ranking methods, see:

Select an unsupervised classification ranking method to assess how effectively each feature performs when you do not have labeled data. The app provides two unsupervised ranking options:

  • Laplacian Score — Scores reflect how well features cluster with other features to form distinct groupings.

  • Variance — Scores reflect feature variance. Features with low variances tend to add less useful information to a model.

Selecting a method activates a new tab with a name that matches the ranking method. For more information on this tab, see Ranking Method Tabs.

For more information on the unsupervised ranking scoring, see:

Unsupervised ranking is available in Diagnostic Feature Designer, but not in Classification Learner. If you plan to export your features to Classification Learner to train a model, you must use ensemble data that includes labels.

Select a prognostic ranking method to assess how effectively each feature tracks the degradation of your ensemble members when you have run-to-failure data. The top-ranking features are best at predicting the remaining useful life (RUL).

The app provides three prognostic ranking methods, all of which score features on a scale ranging from 0 through 1. One method, Monotonicity, is always available. The other two methods, Trendability and Prognosability, are available only when you are using frame-based data. The smaller data segments in frame-based data allow the tracking of small changes that are induced by degradation.

  • Monotonicity characterizes the trend of a feature as the system evolves toward failure. As a system gets progressively closer to failure, a suitable condition indicator has a monotonic positive or negative trend. For more information, see monotonicity.

  • Trendability provides a measure of similarity between the trajectories of a feature measured in multiple run-to-failure experiments. Trendability of a candidate condition indicator is defined as the smallest absolute correlation between measurements. For more information, see trendability.

  • Prognosability is a measure of the variability of a feature at failure relative to the range between its initial and final values. A more prognosable feature has less variation at failure relative to the range between its initial and final values. For more information, see prognosability.

Selecting a method activates a new tab with a name that matches the ranking method. For more information on this method-activated tab, see Ranking Method Tab.

For an example of using prognostic ranking in the app, see Perform Prognostic Feature Ranking for a Degrading System Using Diagnostic Feature Designer.

Select the condition variable that provides the labels for the classification ranking algorithm to use.

Specify the ranking method to sort by when comparing the results of different ranking methods. When you use a single ranking method, the app displays the results in order of importance, as indicated by the ranking score for that method. When comparing the results for multiple methods, change Sort By to change the method that drives the sorting order.

Specify this parameter to eliminate ranking scores for a specific method. Use this parameter, for example, when comparing the results of multiple rankings, and you want to simplify the display by eliminating rankings that do not influence your feature selection.

Export features to use them or share them outside of the app. Both options open a ranking-sorted selectable list to choose from. When you export to the MATLAB workspace, you can use command-line techniques with the features. When you export to the Classification Learner, you open a Classification Learner session that uses your selected features as input.

If you want to export your entire data set from the app, use Export from the Feature Designer tab.

You can also generate code that reproduces the computations for the variables and features you select. For more information, see the code generation options description in the Export section in the Feature Designer tab. When you generate code using Generate Function for Features from the Feature Ranking tab, Ranking Method defaults to the method you specify in Sort By.

Ranking Method Tabs

The correlation importance setting allows you to screen out features that convey similar information to higher ranked features. This screening provides a more diverse feature set in the upper ranks.

The criterion for the screening is the set of cross-correlation coefficients a feature has with higher ranked features. High cross-correlation between two features implies that both features are separating condition groups similarly and provide redundant information. With the default value of 0, the app does not incorporate feature redundancy into ranking scores. As you increase the correlation importance value, the app increases the influence of feature cross-correlation on the feature ranking score. This increasing influence progressively lowers the score of redundant features.

The normalization scheme performs independent normalization across the members for every feature. Normalization allows more direct comparisons among features. The app displays the defining equation for the scheme you select directly beneath your selection.

This option is available only for supervised ranking and unsupervised Laplacian ranking methods.

The Laplacian parameters define key values for calculating the Laplacian score <to be updated>.

  • Number of neighbors — <to be updated>

  • Distance Metric — <to be updated>

  • Kernel Scale — <to be updated>

This option is available only for the unsupervised Laplacian ranking method.

Click Apply to calculate a ranking with the specified parameters. The Feature Ranking tab in the plotting area displays the results both graphically and tabularly. This display also includes the results for the default ranking algorithm, and for any other ranking methods you calculated previously.

Once you calculate a ranking, the app disables Apply until you change a parameter. You can calculate a ranking within a tab multiple times. Each time you modify the parameters and calculate a ranking, the new results overwrite the previous results in the plotting-area tab.

Once you have completed your ranking within the ranking method tab, close that tab to return control to the Feature Ranking tab. The Feature Ranking is disabled while any ranking method tab is activated.

Code Generation Tab

This property is read-only.

The frame policy information reflects the choice you make when you select Export > Generate Function for... in the Feature Designer tab.

Set criteria to refine your options when selecting items for your generated function. All criteria allow you to overwrite selectable options with a string. String matching is case insensitive. Your filters apply to all output items, including signals, features, and ranking tables. Criteria include:

  • Output — String appearing in the output name, which is the name of the variable, feature, or ranking table to select for the generated function

  • Input — Input signal from which the output variable or feature was computed or feature table from which the ranking table was computed

  • Method — Computation that produced the output item, such as TSA or Kurtosis

  • Analysis Type — Data processing, feature processing, or feature ranking

To reset a single filter, delete the contents and click anywhere in the app. To reset all filters at once, click Reset Filters.

Display all selected items together. Use Sort Selected especially when you have used multiple filter combinations to assemble your codegen selections. All your selections appear together.

Specify whether to use parallel computing in the generated code. The default value is the value that is specified in Options. You can specify parallel computing even if you performed your interactive processing without using parallel computing. This approach helps your code to be more scalable if you plan to run the generated code on a larger ensemble than the ensemble you used to develop the features. You can also turn parallel computing off if you used it when you developed the features.

To take advantage of parallel processing in generated code, the user must have Parallel Computing Toolbox installed and licensed. However, the code will still run in serial mode on systems that do not have the toolbox.

Click the Generate Function button when you have completed configuring your selections. The app opens a function that contains computations used for all the output items you selected.

For more information about generating code in the app, see Automatic Feature Extraction Using Generated MATLAB Code.

Programmatic Use

expand all

diagnosticFeatureDesigner opens the Diagnostic Feature Designer app.

diagnosticFeatureDesigner(sessionFile) opens the app and loads a previously saved session. sessionFile is the name of a session data file on the MATLAB path. The data includes all of the variables and features that you either imported into the app or computed within the app. The data also includes your app settings and the processing information necessary to generate code.

To save a session, in the Diagnostic Feature Designer app, on the Feature Designer tab, click Save Session.

More About

expand all

Introduced in R2019a