A framework for detecting defects in sequential inputs to modeling and simulation using machine learning techniques

Abstract

This work presents a method for detecting defects in sequential data inputs for digital twins (DT) during simulations, emphasizing the importance of input validation for ensuring the accuracy and reliability of the simulation results. By thoroughly validating input data, researchers and practitioners can have confidence in the validity of their models, ultimately leading to better decision-making processes and outcomes that are more successful. As DTs continue to expand in complexity, there are an increasing number of mechanisms that may produce undesirable output. An external data stream is just one such potential source of faulty DT behavior and must be analyzed during simulation execution. The proposed framework for validating inputs in real time offers a way to improve the quality and credibility of DTs, guiding future research in the evolving field of modeling and simulation. The case study described in this paper involves using second-order polynomial regression, autoregressive integrated moving average (ARIMA), artificial neural networks (ANN), and a meta-algorithm to detect defects in rocket trajectory data streams, highlighting the effectiveness of validation techniques. This method shows promise, as it successfully identified defects in trajectories in real time using only historical data without knowledge of the future of the data stream. The novelties in this work include 1) using machine learning to validate DT inputs, and 2) performing this validation on sequential data in real time to protect modeled results. This research contributes valuable insights to the field, emphasizing the significance of input validation for enhancing the quality and accuracy of simulation models.