Optimal RUL Estimation: A State-of-Art Digital Twin Application

Abstract

A real world Industrial IoT set up has paved way for simultaneous monitoring of several sensors at their unique sampling rates. This has realized the need for artificial intelligence tools for robust data processing. However, the large size of input data requires real time monitoring and synchronization for online analysis. As the star concept behind the Industry 4.0 wave, a digital twin is a virtual, multi-scale and probabilistic simulation to mirror the performance of its physical counterpart and serve the product lifecycle in a virtual space. Evidently, a digital twin can proactively identify potential issues with its corresponding real twin. Thus, it is best suited for enabling a physics-based and data-driven model fusion to estimate the remaining useful life (RUL) of the components. Traditional RUL prediction approaches have assumed either an exponential or linear degradation trend with a fixed curve shape to build a Health Index (HI) model. Such an assumption may not be useful for multi-sensor systems or cases where sensor data is available intermittently. A common constraint in the industry is irregular sensor data collection. The resulting asynchronous time series of the sporadic data needs to be an accurate representation of the component’s HI when constructing a degradation model. In this paper, we extend the Long-Short Term Memory (LSTM) Recurrent Neural Network (RNN) technique to generate RUL prediction within a digital twin framework as a means of synchronization with changing operational states. More specifically, we first use LSTM encoder-decoder (LSTM-ED) to train a multilayered neural network and reconstruct the sensor data time series corresponding to a healthy state. The resulting reconstruction error can be used to capture patterns in input data time series and estimate HI of training and testing sets. Using a time lag to record similarity between the HI curves, a weighted average of the final RUL estimation is obtained. The described empirical approach is evaluated on publicly available engine degradation dataset with run-to-failure information. Results indicate a high RUL estimation accuracy with greater error reduction rate. This demonstrates wide applicability of the discussed methodology to various industries where event data is scarce for the application of only data-driven techniques.