Condition Indicators for Monitoring, Fault Detection, and Prediction
A condition indicator is a feature of system data whose behavior changes in a predictable way as the system degrades or operates in different operational modes. 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 a statistical moment describing changes in the spectrum over time
Model-based analysis of the data, such as the maximum eigenvalue of a state space model which has been estimated using the data
Combination of both model-based and signal-based approaches, such as using the signal to estimate a dynamic model, simulating the dynamic model to compute a residual signal, and performing statistical analysis on the residual
Combination of multiple features into a single effective condition indicator
The identification of condition indicators is typically the third step of the workflow for designing a predictive maintenance algorithm, after accessing and preprocessing data.
You use condition indicators extracted from system data taken under known conditions to train a model that can then diagnose or predict the condition of a system based on new data taken under unknown conditions. In practice, you might need to explore your data and experiment with different condition indicators to find the ones that best suit your machine, your data, and your fault conditions. The examples Fault Diagnosis of Centrifugal Pumps Using Residual Analysis and Using Simulink to Generate Fault Data illustrate analyses that test multiple condition indicators and empirically determine the best ones to use.
In some cases, a combination of condition indicators can provide better separation between fault conditions than a single indicator on its own. The example Rolling Element Bearing Fault Diagnosis is one in which such a combined indicator is useful. Similarly, you can often train decision models for fault detection and diagnosis using a table containing multiple condition indicators computed for many ensemble members. For an example that uses this approach, see Multi-Class Fault Detection Using Simulated Data.
Predictive Maintenance Toolbox™ and other toolboxes include many functions that can be useful for extracting condition indicators. For more information about different types of condition indicators and their uses, see:
You can extract condition indicators from vectors or timetables of measured or simulated data that you manage with Predictive Maintenance Toolbox ensemble datastores, as described in Data Ensembles for Condition Monitoring and Predictive Maintenance. It is often useful to preprocess such data first, as described in Data Preprocessing for Condition Monitoring and Predictive Maintenance.