PSI - Issue 38

Available online at www.sciencedirect.com Available online at www.sciencedirect.com Available online at www.sciencedirect.com

ScienceDirect

Procedia Structural Integrity 38 (2022) 159–167 Structural Integrity Procedia 00 (2021) 000–000 Structural Integrity Procedia 00 (202 ) 000–000

www.elsevier.com / locate / procedia www.elsevier.com / locate / procedia

Fatigue Design 2021, 9th Edition of the International Conference on Fatigue Design A data-driven approach for approximating non-linear dynamic Fatigue Design 2021, 9th Edition of the International Conference on Fatigue Design A data-driven approach for approxi ating non-linear dyna ic

systems using LSTM networks L. Heindel a, ∗ , P. Hantschke a,b , M. Ka¨stner a,b a Technische Universita¨ t Dresden, Institute of Solid Mechanics, 01062 Dresden, Germany b Dresden Center for Fatigue and Reliability (DCFR) 01062 Dresden, Germany syste s using LST networks L. Heindel a, ∗ , P. Hantschke a,b , M. Ka¨stner a,b a Technische Universita¨ t Dresden, Institute of Solid Mechanics, 01062 Dresden, Germany b Dresden Center for Fatigue and Reliability (DCFR) 01062 Dresden, Germany

© 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the Fatigue Design 2021 Organizers Our approach is tested on a large, non-linear experimental dataset, obtained from a servo hydraulic fatigue test bench. To enable a comprehensive evaluation of the model quality, both hybrid modeling approaches are compared on virtual sensing and forward prediction tasks using multiple error metrics relevant to fatigue analysis. 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND licen e (https: // reativecommons.org / licenses / by-nc-nd / 4.0) Peer-review under responsibility f the scien ific ommittee of h Fa igue Design 2021 Organizers. Keywords: LSTM; non-linear dynamic systems; Virtual Sensing; Forward Prediction; hybrid modeling Virtual sensing techniques aim to replace physical sensors in a system by using the data from available sensors to estimate additional unknown quantities of interest. Forward prediction estimates the system response for a given drive signal during the commission of component test rigs. Data-driven approaches can be convenient tools for forward prediction and virtual sensing, as they only require a su ffi ciently large dataset of the desired input and output quantities for the purpose of model parametrization. The presented approach explores two hybrid modeling strategies, which combine Frequency Response Function models with Long Short-Term Memory networks to approximate the behavior of non-linear dynamic systems with multiple input and output channels. The proposed method utilizes short subsequences of signals to carry out model training and prediction. Long sequence estimations are generated by combining the individual subsequence predictions using a windowing technique. Our approach is tested on a large, non-linear experimental dataset, obtained from a servo hydraulic fatigue test bench. To enable a comprehensive evaluation of the model quality, both hybrid modeling approaches are compared on virtual sensing and forward prediction tasks using multiple error metrics relevant to fatigue analysis. © 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https: // creativecommons.org / licenses / by-nc-nd / 4.0) Peer-review under responsibility of the scientific committee of the Fatigue Design 2021 Organizers. Keywords: LSTM; non-linear dynamic systems; Virtual Sensing; Forward Prediction; hybrid modeling Abstract The analysis of sensor data in nonlinear dynamical systems plays a fundamental role in a variety of modern engineering problems like Fatigue Analysis and Predictive Maintenance. In many applications, the accurate approximation of sensor signals supports and complements experimental e ff orts in order to accelerate the development process and conserve resources. Virtual sensing techniques aim to replace physical sensors in a system by using the data from available sensors to estimate additional unknown quantities of interest. Forward prediction estimates the system response for a given drive signal during the commission of component test rigs. Data-driven approaches can be convenient tools for forward prediction and virtual sensing, as they only require a su ffi ciently large dataset of the desired input and output quantities for the purpose of model parametrization. The presented approach explores two hybrid modeling strategies, which combine Frequency Response Function models with Long Short-Term Memory networks to approximate the behavior of non-linear dynamic systems with multiple input and output channels. The proposed method utilizes short subsequences of signals to carry out model training and prediction. Long sequence estimations are generated by combining the individual subsequence predictions using a windowing technique. Abstract The analysis of sensor data in nonlinear dynamical systems plays a fundamental role in a variety of modern engineering problems like Fatigue Analysis and Predictive Maintenance. In many applications, the accurate approximation of sensor signals supports and complements experimental e ff orts in order to accelerate the development process and conserve resources.

1. Introduction 1. Introduction

Measurement data are essential in the design or operation of most mechanical systems. They also form an impor tant prerequisite for component testing and fatigue assessment. The acquisition of measurement data is accomplished by sensor setups, which must satisfy technical and economic constraints. It is therefore desirable to develop strategies Measurement data are essential in the design or operation of most mechanical systems. They also form an impor tant prerequisite for component testing and fatigue assessment. The acquisition of measurement data is accomplished by sensor setups, which must satisfy technical and economic constraints. It is therefore desirable to develop strategies

∗ Corresponding author. Tel.: + 49 351 463-37993 ; fax: + 49 351 463-37969. E-mail address: leonhard.heindel@tu-dresden.de ∗ Corresponding author. Tel.: + 49 351 463-37993 ; fax: + 49 351 463-37969. E-mail address: leonhard.heindel@tu-dresden.de

2452-3216 © 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the Fatigue Design 2021 Organizers 10.1016/j.prostr.2022.03.017 2210-7843 © 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https: // creativecommons.org / licenses / by-nc-nd / 4.0) Peer-review under responsi bility of the scientific committee of Fatigue Design 2021 Organizers . 2210-7843 © 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https: // creativecommons.org / licenses / by-nc-nd / 4.0) Peer-review under responsi bility of the scientific committee of the Fatigue Design 2021 Organizers .