Predictive maintenance allows equipment users and manufacturers to assess the working condition of machinery, diagnose faults, or estimate when the next equipment failure is likely to occur. When you can diagnose or predict failures, you can plan maintenance in advance, better manage inventory, reduce downtime, and increase operational efficiency.
A key step in predictive maintenance algorithm development is identifying condition indicators, which are features in your system data whose behavior changes in a predictable way as the system degrades. A condition indicator can be any feature that is useful for distinguishing normal from faulty operation or for predicting remaining useful life. A useful condition indicator clusters similar system status together and sets different status apart. Examples of condition indicators include quantities derived from:
Simple analysis, such as the mean value of the data over time.
More complex signal analysis, such as the frequency of the peak magnitude in a signal spectrum, or the time-synchronous average of a signal from a rotating source.
In the Diagnostic Feature Designer app, you can develop features and evaluate potential condition indicators using a multifunction graphical interface.
The app operates on data ensembles. An ensemble is a collection of data sets, created by measuring or simulating a system under varying conditions. An individual data set representing one system under one set of conditions is a member. Diagnostic Feature Designer processes all ensemble members when executing one operation.
Within Diagnostic Feature Designer, you can interactively:
Explore your data ensemble visually by plotting and interacting with your ensemble members together.
Convert your data into different forms for further exploration. For example, you can create a power spectrum of your signal to evaluate its frequency-domain behavior. Or you can perform time-synchronous averaging, which filters out any noise or disturbance that is not associated with your machine rotation.
Generate features of various types, and plot histograms that visualize the effectiveness of each feature in separating data from systems with different conditions.
Rank the generated features by using ranking algorithms that use specific criteria to establish which features are most effective.
The following three-part tutorial takes you through the Diagnostic Feature Designer workflow for a transmission system model, from your initial data inport to the export of your chosen features.
The full workflow for a predictive maintenance program includes multiple steps that begin with data acquisition and end with deployment and integration of a condition monitoring algorithm. For more information, see Designing Algorithms for Condition Monitoring and Predictive Maintenance.