The occurrence of combustion instabilities poses a high risk to the safe operation of liquid propellant rocket engines. Forecasting such instabilities in real-time is an active field of research, as it would allow the engine operator to initiate emergency procedures during testing and operation before the combustion chamber sustains potentially irreversible damage. In recent years, machine learning has emerged as a promising approach to time-series forecasting. However, like most data-driven methods, they require large amounts of relevant data to be collected beforehand, which is often not available or cost intensive to produce. This study investigates the feasibility of using Domain Adversarial Neural Networks to combine datasets of three combustion chambers, representing a wide range of operating conditions and experimental setups. Using sliding windows of multimodal sensor data, the models forecast the future evolution of pressure fluctuation statistics 200 ms ahead, from which impending instabilities can be inferred. We show that Domain Adversarial Neural Networks are capable of extracting domain-invariant features from the multimodal sensor inputs, leading to improved cross-domain prediction performance, particularly under limited target data availability.

A combustion chamber pressure- and mixture ratio-controller based on deep reinforcement learning methods is under development at German Aerospace Center - DLR. Training and implementation of the controller is described. The performance of this controller is evaluated in simulation and experimental tests. Application use case is a 22N nitrous oxide/ethane thruster that can be operated in a wide range of operation points. Steady state and trajectory-following performance is measured via mean squared error, standard deviation, overshot and settling time evaluation in simulation and experiment. For steady state set-point control the controller shows good performance within the trained operation points in simulation and experiment. Root mean squared error of chamber pressure and mixture ratio is below 0.5 bar and 1 respectively for all physically achievable set-points. Control performance in simulation is better than at the real system. Trajectory-following performance in simulation is convincing, showing nearly no offset to the target values, whereas in the experimental evaluation deviations and fluctuations occur, indicating the need for further optimization when transferring the controller from simulation to the real system.

At DLR neural networks, as potential future controller for rocket engines, are studied. A neural network-based chamber pressure controller for a simplified cold gas thruster is presented and analyzed in simulation and experiment. The goal of the controller is twofold: It can track a trajectory with different changes of setpoints and it allows to set and control a wide variety of steady state chamber pressures. The neural network gets feeding line pressure measurement data as input and calculates valve positions as output values. The training phase of the controller is done with a reinforcement learning algorithm in an EcosimPro/ESPSS simulation, that is validated with data from the corresponding experimental set up. To increase the robustness and to allow a transfer from the simulation directly to the test facility domain randomization is applied. The controller is evaluated in simulations and experiment. It was found that - in the range of physically possible operation points - the controller achieves a constantly high reward which corresponds to a low error and a good control performance. In the simulation the controller was able to adjust all required set points with a steady state error of less than 0.1bar while retaining a small overshoot and an optimal settling time. It is found that the controller is also able to regulate all desired set points in the real experiment. A reference trajectory with different steps, linear and sinus changes in target pressure is tested in simulation and experiment. The controller was in both cases able to successfully follow the given trajectory.

Traditional experimental design methods often face challenges in handling complex aerospace systems due to the high dimensionality and nonlinear behavior of such systems, resulting in nonoptimal experimental designs. To address these challenges, machine learning techniques can be used to further increase the application areas of modern Bayesian Optimal Experimental Design (BOED) approaches, enhancing their efficiency and accuracy. The proposed method leverages neural networks as surrogate models to approximate the underlying physical processes, thereby reducing computational costs and allowing for full differentiability. Additionally, the use of reinforcement learning enables the optimization of sequential designs and essential real-time capability. Our framework is validated by optimizing experimental designs that are used for the efficient characterization of turbopumps for liquid propellant rocket engines. The reinforcement learning approach yields superior results in terms of the expected information gain related to a sequence of 15 experiments, exhibiting mean performance increases of 9.07% compared to random designs and 6.47% compared to state-of-the-art approaches. Therefore, the results demonstrate significant improvements in experimental efficiency and accuracy compared to conventional methods. This work provides a robust framework for the application of advanced BOED methods in aerospace testing, with implications for broader engineering applications.

Reusability of LREs in Europe is increasingly attracting the attention of scientific community and industry with leading projects such as THEMIS, CALLISTO (reusable demonstrators for vertical take-off and landing (VTVL)) and Ariane Next - all powered by the reusable cryogenic Oxygen/Methane (LOX/LCH4) engine "Prometheus". To enable further expansion and cost-effectiveness of the reusability technology for future liquid rocket engines (LREs), research on critical engine components such as turbopumps is crucial. Therefore, within our research we focus on the turbine blade investigation for reusable LRE applications including high cycle fatigue (HCF) and low cycle fatigue (LCF). Validation of defined applied analytical and numerical techniques is established through the Liquid Upper stage deMonstrator ENgine (LUMEN)'s, developed at DLR Lamplodshausen for enhanced expertise in the complete cycle operation for various engine applications, as well as to empower validation studies of the operational conditions to which turbopump components, such as turbine blades, are subjected. Turbine blades are exposed to large thermo-mechanical cyclic strains emerging from an increased temperature driving gas combined with a fast start-up sequence as well as a large rotational speed - essential for acquiring high performance and structural mass efficiency for LREs. Therefore, in addition to bending & torsion as well as thermal gradient and centrifugal forces, it is critical to consider creep effects in durability studies. To forecast the turbine blade fatigue life, analytical (0-D) and numerical approaches for a selected test case are studied. Within the proposed method, a BLISK is assessed for the most severe loading condition considering HCF load by a modified Goodman method, along with a Coffin-Manson based approach for LCF contribution. Each operational cycle under constant maximum loading condition is applied to study the creep effect. As a result, an enhanced fatigue life prediction method including both creep and fatigue conditions for a turbine blade is obtained.

Advanced control algorithms will change the way we operate liquid propellant rocket engines in the future. One approach is to use neural networks as the control policy, which are trained in simulation with deep reinforcement learning. Neural network controllers can easily deal with multiple high-level control goals such as maximum efficiency, long fatigue-life, or mechanical and thermodynamic constraints. Within its technological road map towards cost-efficient and reusable engines, the German Aerospace Center (DLR) is developing an advanced engine controller for an expander-bleed engine. In this work, we design a neural network controller for the LUMEN turbopump and test the control performance at our test bench, tackling the challenges of real-world systems.

The German Aerospace Center (DLR) Institute of Space Propulsion has unique expertise in operating test facilities for rocket engine testing and development in Europe since 1959. However, essential elements of the test site were designed up to half a century ago. In order to ensure a futureproof and intelligent digital test infrastructure, the potential of test automation, advanced control, and monitoring systems is investigated based on machine learning. Such intelligent control systems are expected to reduce engine development and test preparation times, thereby lowering the associated costs. Additionally, advanced monitoring systems are anticipated to increase the safety and reliability of the test infrastructure. This paper presents the results of two pilot projects: the first project uses reinforcement learning to automatically generate test sequences based on test requirements, while the second project develops a feed-forward forecasting model to predict deviations from expected behavior in the feed-line of a rocket engine test facility.

Accurate calculations of the heat transfer and the resulting maximum wall temperature are essential for the optimal design of reliable and efficient regenerative cooling systems. However, predicting the heat transfer of supercritical methane flowing in cooling channels of a regeneratively cooled rocket combustor presents a significant challenge. High-fidelity CFD calculations provide sufficient accuracy but are computationally too expensive to be used within elaborate design optimization routines. In a previous work it has been shown that a surrogate model based on neural networks is able to predict the maximum wall temperature along straight cooling channels with convincing precision when trained with data from CFD simulations for simple cooling channel segments. In this paper, the methodology is extended to cooling channels with curvature. The predictions of the extended model are tested against CFD simulations with different boundary conditions for the representative LUMEN combustor contour with varying geometries and heat flux densities. The high accuracy of the extended model's predictions, suggests that it will be a valuable tool for designing and analyzing regenerative cooling systems with greater efficiency and effectiveness.

To increase liquid rocket engines (LREs) lifetime capability and allow for reusability applications, the efficient evaluation of the most critical subcomponents' remaining useful life plays a vital role. Regeneratively cooled combustion chamber (CC) wall must withstand extremely high loads emerging from a massive temperature gradient between the hot gas and the low temperature of the coolant. The combined loading and unloading operations, together with high temperature and rate dependent inelastic strain, significantly lessen the combustion chamber inner liner life. Within the presented research, the post-processing model was developed for low cycle fatigue (LCF) evaluation of the reusable LRE’s combustion chamber walls. The proposed damage accumulation model is based on the amalgamation of Bonora-Gentile-Pirondi (2004) and Dufailly-Lemaitre (1995) methods, and it incorporates ductile and brittle damage components which are embedded in the post-processing method. Moreover, the required numerical calculation time is further decreased on account of the proposed routine which allows for analysis of only two initial numerically acquired FE cycles. The obtained results based on the developed method combined with coupled thermal-structural quasi 2D Finite Element Analysis (FEA) of the nozzle throat cross-section, were confirmed to be in good agreement with the validation data acquired from the M51 thermo-mechanical laboratory site at DLR Lampoldshausen. The proposed model can be successfully applied for a quick evaluation of the remaining useful life of the CC wall for various rocket engine architectures.

We present a method that combines multiple sensory modalities in a rocket thrust chamber to predict impending thermoacoustic instabilities with uncertainties. This is accomplished by training an autoregressive Bayesian neural network model that forecasts the future amplitude of the dynamic pressure time series, using multiple sensor measurements (injector pressure/ temperature measurements, static chamber pressure, high-frequency dynamic pressure measurements, high-frequency OH* chemiluminescence measurements) and future flow rate control signals as input. The method is validated using experimental data from a representative cryogenic research thrust chamber. The Bayesian nature of our algorithms allows us to work with a dataset whose size is restricted by the expense of each experimental run, without making overconfident extrapolations. We find that the networks are able to accurately forecast the evolution of the pressure amplitude and anticipate instability events on unseen experimental runs 500 milliseconds in advance. We compare the predictive accuracy of multiple models using different combinations of sensor inputs. We find that the high-frequency dynamic pressure signal is particularly informative. We also use the technique of integrated gradients to interpret the influence of different sensor inputs on the model prediction. The negative log-likelihood of data points in the test dataset indicates that prediction uncertainties are well-characterized by our model and simulating a sensor failure event results in a dramatic increase in the epistemic component of the uncertainty, as would be expected when a Bayesian method encounters unfamiliar, out-of-distribution inputs.

The last years have witnessed an enormous interest in the use of artificial intelligence methods, especially machine learning algorithms. This also has a major impact on aerospace engineering in general, and the design and operation of liquid rocket engines in particular, and research in this area is growing rapidly. The paper describes current machine learning applications at the DLR Institute of Space Propulsion. Not only applications in the field of modeling are presented, but also convincing results that prove the capabilities of machine learning methods for control and condition monitoring are described in detail. Furthermore, the advantages and disadvantages of the presented methods as well as current and future research directions are discussed.

We present a data-driven method for the early detection of thermoacoustic instabilities. Recurrence quantification analysis is used to calculate characteristic combustion features from short-length time series of dynamic pressure sensor data. Features like recurrence rate are used to train support vector machines to detect the onset of instability a few hundred milliseconds in advance. The performance of the proposed method is investigated on experimental data from a representative LOX/H 2 research thrust chamber. In most cases, the method is able to timely predict two types of thermoacoustic instabilities on test data not used for training. The results are compared with state-of-the-art early warning indicators.

The regenerative cooling still is one of the most crucial parts in designing a liquid rocket engine. The goal of an effective design is a trade-off between a sufficient cooling of the structure and a low pressure drop in the cooling channels. For an expander type cycle a third requirement arises: A sufficient enthalpy increase of the cooling fluid for the turbopumps to operate effectively. To prevent the combustion chamber from melting, the regenerative cooling has to be designed with a sufficient margin to respect all uncertainties. A neural network based surrogate model is used to study the robustness of the cooling channel design for other load points of the engine. In this paper, the design of the regenerative cooling system and the manufacturing of a subscale combustion chamber will be discussed.

Intelligent engine control could be one of the most important innovations in the development of future reusable engines, facilitating a safer and more economical engine operation. In this work, we investigate the closed-loop control of the LUMEN expander-bleed engine by combining machine learning with a transient simulation environment. The controller can dynamically change the set-point of the engine between a chamber pressure of 40 bar to 80 bar by adjusting up to six flow control valves while maintaining several boundary conditions at any given time.

Advanced engine control is an important requirement for the efficient operation of future reusable engines, facilitating a safer and more economical engine operation. For this reason, modern control strategies are extensively studied in recent years, mainly through the use of simulation environments. An important development step is to test the performance and robustness of the control algorithm at real test benches. The present paper describes the first steps towards the use of an reinforcement learning based controller on a N2O C2H6 22N green propellant thruster. The control objectives are given by regulating the mixture ratio and combustion pressure. The existing test bench is modelled in EcosimPro\,/\,ESPSS. Based on the simulation model deep reinforcement learning is used to train the controller and domain randomization is used to increase the robustness. The overall goal is to transfer the controller from the simulation model to the real test bench. Finally, preliminary experiments demonstrate the basic functionality of reinforcement learning based controllers for real rocket propulsion systems.

Nowadays, liquid rocket engines use closed-loop control at most near-steady operating conditions. The control of the transient phases is traditionally performed in open loop due to highly nonlinear system dynamics. This situation is unsatisfactory, in particular for reusable engines. The open-loop control system cannot provide optimal engine performance due to external disturbances or the degeneration of engine components over time. In this article, we study a deep reinforcement learning approach for optimal control of a generic gas-generator engine's continuous startup phase. It is shown that the learned policy can reach different steady-state operating points and convincingly adapt to changing system parameters. Compared to carefully tuned open-loop sequences and proportional-integral-derivative (PID) controllers, the deep reinforcement learning controller achieves the highest performance. In addition, it requires only minimal computational effort to calculate the control action, which is a big advantage over approaches that require online optimization, such as model predictive control.

The 100 MW cryogenic liquid oxygen/hydrogen multi-injector combustor BKD operated by the DLR Institute of Space Propulsion is a research platform that allows the study of thermoacoustic instabilities under realistic conditions, representative of small upper stage rocket engines. We use data from BKD experimental campaigns in which the static chamber pressure and fuel-oxidizer ratio are varied such that the first tangential mode of the combustor is excited under some conditions. We train an autoregressive Bayesian neural network model to forecast the amplitude of the dynamic pressure time series, inputting multiple sensor measurements (injector pressure/ temperature measurements, static chamber pressure, high-frequency dynamic pressure measurements, high-frequency OH* chemiluminescence measurements) and future flow rate control signals. The Bayesian nature of our algorithms allows us to work with a dataset whose size is restricted by the expense of each experimental run, without making overconfident extrapolations. We find that the networks are able to accurately forecast the evolution of the pressure amplitude and anticipate instability events on unseen experimental runs 500 milliseconds in advance. We compare the predictive accuracy of multiple models using different combinations of sensor inputs. We find that the high-frequency dynamic pressure signal is particularly informative. We also use the technique of integrated gradients to interpret the influence of different sensor inputs on the model prediction. The negative log-likelihood of data points in the test dataset indicates that predictive uncertainties are well-characterized by our Bayesian model and simulating a sensor failure event results as expected in a dramatic increase in the epistemic component of the uncertainty.