surrogate modeling based on deep learning for the trajectory of a launch vehicle

Purpose: in recent years, the analysis and design of systems through computer-based simulations have attracted considerable attention from researchers focused on predicting system performance. such engineering analyses depend on the execution of costly and complex computer codes. approximate methods have been extensively utilized to alleviate the computational burden of engineering analyses, and the advancement of modeling techniques allows for rapid, cost-effective, and accurate evaluations of engineering systems. this paper aims to explore the potential of deep learning models as an alternative approach for modeling the trajectory of a launch vehicle.design/methodology/approach: to enable analyses such as design optimization, reliability assessment, and others, there is a necessity for a simplified model that can efficiently represent the detailed and expensive product model. these simplified predictive models are also known as surrogate models. this approach employs experimental data to train neural networks and assesses its predictive accuracy against the response surface method (rsm). findings: the results demonstrated that the use of deep learning significantly enhances prediction accuracy and provides the capability to estimate performance under various conditions. consequently, at the comparison point for the total mass of the launch vehicle, the simulation code produced a value of 108,500 kg using the deep learning method, while the total mass with the rsm technique was 108,556.6 kg, indicating that the accuracy of the deep learning model is superior.practical implications: the deep learning model can identify complex nonlinear relationships among input variables, leading to more robust predictions. this advantage arises from the flexibility of deep neural networks, which adeptly learn intricate patterns within the training data. this model significantly decreases the time needed for predictions compared to traditional modeling techniques, which often require extensive iterative processes and parameter tuning. this reduction in computational load is vital for real-time aerospace applications, where swift decision-making is essential. originality/value: this paper analyzes the effectiveness of deep learning models in flight trajectory modeling, specifically concerning the two-stage launch vehicle kosmos 3m, aiming to reduce total mass while improving computational efficiency and accuracy in predictive modeling for optimal flight path design.

surrogate modeling based on deep learning for the trajectory of a launch vehicle

purpose: in recent years, the analysis and design of systems through computer-based simulations have attracted considerable attention from researchers focused on predicting system performance. such engineering analyses depend on the execution of costly and complex computer codes. approximate methods have been extensively utilized to alleviate the computational burden of engineering analyses, and the advancement of modeling techniques allows for rapid, cost-effective, and accurate evaluations of engineering systems. this paper aims to explore the potential of deep learning models as an alternative approach for modeling the trajectory of a launch vehicle.design/methodology/approach: to enable analyses such as design optimization, reliability assessment, and others, there is a necessity for a simplified model that can efficiently represent the detailed and expensive product model. these simplified predictive models are also known as surrogate models. this approach employs experimental data to train neural networks and assesses its predictive accuracy against the response surface method (rsm). findings: the results demonstrated that the use of deep learning significantly enhances prediction accuracy and provides the capability to estimate performance under various conditions. consequently, at the comparison point for the total mass of the launch vehicle, the simulation code produced a value of 108,500 kg using the deep learning method, while the total mass with the rsm technique was 108,556.6 kg, indicating that the accuracy of the deep learning model is superior.practical implications: the deep learning model can identify complex nonlinear relationships among input variables, leading to more robust predictions. this advantage arises from the flexibility of deep neural networks, which adeptly learn intricate patterns within the training data. this model significantly decreases the time needed for predictions compared to traditional modeling techniques, which often require extensive iterative processes and parameter tuning. this reduction in computational load is vital for real-time aerospace applications, where swift decision-making is essential. originality/value: this paper analyzes the effectiveness of deep learning models in flight trajectory modeling, specifically concerning the two-stage launch vehicle kosmos 3m, aiming to reduce total mass while improving computational efficiency and accuracy in predictive modeling for optimal flight path design.