RUL Estimation Using Identified Models or State Estimators

When you have an identified dynamic model that describes some aspect of system behavior, you can use that model to forecast future behavior. You can identify such a dynamic model from system data. Or, if you have system data that represents the operation of your machines with time or usage, you can extract condition indicators from that data and track the behavior of the condition indicators with time or usage. You can then identify a model that describes the behavior of the condition indicator, and use that model to predict future values of a condition indicator. If you know, for example, that your system needs repair when some condition indicator exceeds some threshold, you can identify a model of the time evolution of that condition indicator. You can then propagate the model forward in time to determine how long it will be before the condition indicator reaches the threshold value.

Some functions you can use for identification of dynamic models include:

• ssest — Estimate a state-space model from time-domain input-output data or frequency-response data.

• arx, armax, ar — Estimate an autoregressive or moving-average (AR or ARMA) model from time-series data.

• nlarx — Model nonlinear behavior using dynamic nonlinearity estimators such as wavelet networks, tree-partitioning, and sigmoid networks.

You can use functions like forecast to predict the future behavior of the identified model. The example Condition Monitoring and Prognostics Using Vibration Signals uses this approach to RUL prediction.

There are also recursive estimators that let you fit models in real-time as you collect and process the data, such as recursiveARX and recursiveAR.

RUL estimation with state estimators such as unscentedKalmanFilter, extendedKalmanFilter, and particleFilter works in a similar way. You perform state estimation on some time-varying data, and predict future state values to determine the time until some state value associated with failure occurs.

Related Topics

• Nonlinear State Estimation of a Degrading Battery System
• Condition Monitoring and Prognostics Using Vibration Signals
• Models for Predicting Remaining Useful Life
• Feature Selection for Remaining Useful Life Prediction