Prediction of remaining useful life of an asset using conformal mathematical filtering
A system determines that an asset of an engineering system has transitioned from a quasi-steady degradation stage to an accelerated degradation phase based on sensor measurements received from an asset. During the accelerated degradation phase, features are extracted from the sensor measurements that are indicative of wear of the asset. A conformal mathematical filter is applied to the features that causes the features to conform to a wear curve formulation associated with the asset. An output of the filter is resampled to form a noise-reduced signal. The noise-reduced signal is input into a sequence machine learning model. A loss function of the sequence machine learning model uses an increased penalty to overprediction and a relaxed penalty for underprediction. An output of the sequence machine learning model is used to predict a remaining useful life (RUL) of the asset.
1 . A computer-implemented method comprising:
based on sensor measurements received from an asset of an engineering system, determining that the asset has transitioned from a quasi-steady degradation stage to an accelerated degradation phase; and
during the accelerated degradation phase:
extracting features from the sensor measurements that are indicative of wear of the asset;
applying a conformal mathematical filter to the features, the conformal mathematical filter causing the features to conform to a wear curve formulation associated with the asset, an output of the filter being resampled to form a noise-reduced signal;
inputting the noise-reduced signal into a sequence machine learning model, a loss function of the sequence machine learning model using an increased penalty to overprediction and a relaxed penalty for underprediction; and
using an output of the sequence machine learning model to predict a remaining useful life (RUL) of the asset.
2 . The method of claim 1 , wherein the wear curve formulation is based on a physics-based behavioral model.
3 . The method of claim 2 , wherein the physics-based behavioral model is a crack propagation formulation.
4 . The method of claim 2 , wherein the physics-based behavioral model is a tool degradation formulation.
5 . The method of claim 2 , wherein the physics-based behavioral model is a chemical reaction formulation.
6 . The method of claim 2 , wherein the physics-based behavioral model is a corrosion formulation.
7 . The method of claim 2 , wherein the physics-based behavioral model is based on one of adhesive wear, surface fatigue, fretting wear, erosive wear, corrosion wear, oxidation wear, impact wear, cavitation wear, and diffusive wear.
8 . The method of claim 1 , wherein the sequence machine learning model comprises a recurrent neural network.
9 . The method of claim 1 , wherein the sequence machine learning model comprises a physics-informed neural network.
10 . The method of claim 1 , wherein the sequence machine learning model comprises an ordinary differential equation (ODE) neural network.
11 . The method of claim 10 , wherein the ODE neural network comprises an encoder-decoder block that transforms the noise-reduced signal into a latent space, and wherein learning occurs in the latent space.
12 . The method of claim 1 , further comprising using the output of the sequence machine learning model to predict fitting constants for an RUL mathematical model, the RUL mathematical model being used to predict the RUL.
13 . The method of claim 1 , wherein the sequence machine learning model is trained using a data set exhibiting a wear trend predicted to be similar to the wear curve of the asset.
14 . The method of claim 1 , wherein determining that the asset has transitioned from the quasi-steady degradation stage to the accelerated degradation phase comprises monitoring derived features from the sensor measurements and detecting an elbow point using a binary classifier, the detecting of the elbow point indicating a start of the accelerated degradation phase.
15 . A system, comprising:
an engineering system comprising:
one or more assets; and
one or more sensors that provide prognostics data about the one or more assets; and
a monitoring system coupled to the engineering system, the monitoring system comprising:
an external data interface that receives the prognostics data from the one or more sensors;
a processor coupled to the external data interface and operable to perform, during an accelerated degradation phase of the one or more assets:
extracting features from the prognostics data that are indicative of wear of the asset;
applying a conformal mathematical filter to the features, the conformal mathematical filter causing the features to conform to a wear curve formulation associated with the asset, an output of the filter being resampled to form a noise-reduced signal;
inputting the noise-reduced signal into a sequence machine learning model, a loss function of the sequence machine learning model using an increased penalty to overprediction and a relaxed penalty for underprediction; and
using an output of the sequence machine learning model to predict a remaining useful life (RUL) of the asset; and
a user interface coupled to the processor that communicates the prediction of the RUL to a user.
16 . The system of claim 15 , wherein the wear curve formulation is based on a physics-based behavioral model.
17 . The system of claim 15 , wherein the sequence machine learning model comprises one of a recurrent neural network, a physics-informed neural network, and an ordinary differential equation neural network.
18 . The system of claim 15 , wherein the sequence machine learning model is trained using a data set exhibiting a wear trend similar to the wear curve.
19 . The system of claim 15 , wherein the processor is further configured to determine that the asset has transitioned from a quasi-steady degradation stage to the accelerated degradation phase.
20 . The system of claim 19 , wherein determining that the asset has transitioned from the quasi-steady degradation stage to the accelerated degradation phase comprises monitoring derived features from the sensor measurements and detecting an elbow point using a binary classifier, the detecting of the elbow point indicating a start of the accelerated degradation phase.