Browse Topic: Artificial intelligence (AI)
SCOPE IS UNAVAILABLE.
Deep learning (DL) models have attained state-of-the-art performance in numerous fields. Nevertheless, for certain real-world applications, existing models encounter diverse challenges, ranging from a lack of generability to new data to issues of scalability and overfitting. In this context, integrating information extracted from different modalities holds promise as a potential solution to alleviate these challenges. This paper introduces MAVEN, a multimodal deep-learning framework for long-range atmospheric visibility estimation. Using multimodal deep learning, MAVEN fuses various modalities to estimate long-range atmospheric visibility. These modalities include RGB imagery, Edge Map, Entropy Map, Depth Map, and Normal Surface Map. Results show that in contrast to single-modality RGB, which achieves only 87.92% accuracy, multimodal deep learning models achieve an accuracy of over 96%. This significant improvement highlights the potential of multimodal approaches to enhance the accuracy and reliability of atmospheric visibility estimation, which is crucial for improving safety in applications such as aviation, maritime navigation, and autonomous vehicles. By addressing challenges such as data variability, environmental factors, and the inherent complexity of atmospheric conditions, MAVEN contributes to more reliable and robust visibility estimation systems, thereby enhancing safety and operational efficiency in critical environments.
This paper presents a reinforcement learning (RL)–based outer-loop controller for quadrotor UAV trajectory tracking and its real-world experimental validation. The proposed approach integrates RL into a standard cascaded flight-control architecture by replacing the conventional PID outer loop while retaining the onboard attitude and body-rate PID controllers. This hierarchical design preserves reliable inner-loop stabilization while leveraging RL to address nonlinear dynamics, coupling effects, and modeling uncertainty in translational motion. The controller is trained entirely in a physics-based simulation using Proximal Policy Optimization (PPO) and transferred directly to a Crazyflie quadrotor without additional tuning. Performance is evaluated through real-world figure-8 trajectory tracking experiments with varying time scales to impose increasing dynamic demands. Compared to a conventional PID outer-loop controller operating under identical conditions, the RL-based controller consistently reduces effective phase delay and achieves lower position and velocity tracking errors, particularly for aggressive trajectories. The results demonstrate robust sim-to-real transfer and highlight the potential of learning-based outer-loop control as a drop-in enhancement to classical quadrotor flight controllers.
This paper presents a comprehensive evaluation of data-driven machine learning (ML) frameworks for the estimation of critical operational parameters, gross weight (GW), longitudinal center-of-gravity (CGx ), and airspeed (Ux ) for a UAM-scale Lift plus Cruise eVTOL aircraft. Artificial Neural Networks (ANN), Gaussian Process Regression (GPR), and Support Vector Machines (SVM) are compared for their ability to track these dynamic parameters across both low-speed rotor-borne and high-speed wing-borne flight regimes. The models are rigorously tested on steady-state clean data and stochastic atmospheric turbulence data sets to assess performance trade-offs between computational cost, noise robustness, and predictive accuracy. Results demonstrate that GPR consistently achieves the highest accuracy on clean data, particularly for GW and CGx estimation, though it exhibits the highest sensitivity to stochastic noise. Conversely, SVM demonstrates the greatest relative robustness under turbulent conditions and superior computational efficiency, identifying it as a practical candidate for resource-constrained onboard flight computers. Furthermore, a dynamic continuous-time analysis reveals a critical trade-off between responsiveness and accuracy. Instantaneous predictions are shown to suffer from severe transient error spikes during maneuvers, whereas a moving average filtering strategy effectively mitigates these errors at the cost of response latency. These analyses demonstrate the feasibility of ML-based parameter estimation for UAM operations and highlight the necessity of adaptive temporal filtering to balance agility with resilience in turbulent environments.
Dimensional reduction of data can be accomplished through various methods and has applications critical to machine learning and surrogate modeling. Within the rotorcraft community, leveraging these techniques allows for improved rotor parameterization and performance prediction. Machine learning models generally perform faster and better with lower input dimensions, so long as all necessary information is retained, making appropriate dimension reduction paramount. Data can also be arranged in a one-dimensional (concatenated/stacked) or two-dimensional arrays to take advantage of function correlations, and this arrangement may allow for greater reduction at lower reconstruction costs. Principal Component Analysis with a stacked input shape proves to be the most effective reduction method considered, with reconstruction accuracy being validated though a suite of mid-fidelity aerodynamic simulations. A blade geometry defined using 204 original parameters can be fully described using just 10 component parameters with the reconstructed blade maintaining performance figures within 1% of the original blade.
Characterization of rotor–rotor wake interactions and their influence on flight dynamics is an important step toward advancing control system design and evaluating the performance of next-generation Mars multirotors. In this work, a Viscous Vortex Particle Method (VVPM) is utilized to generate rotor–rotor interference data for the Chopper Mars Helicopter platform, a large-scale hexacopter concept designed to be capable of carrying payload and pursuing independent science tasks. A reduced-order model compatible with finite state dynamic inflow is derived from the database. Interpolation strategies for continuous look-up are evaluated, with Gaussian Process Regression providing up to 20% improvement in prediction accuracy over linear interpolation of the interference data, although its scalability is limited by the large number of output channels. The interference model is implemented in HeliCAT, the flight dynamics analysis framework used for the Ingenuity Mars Helicopter, to assess the impact of rotor–rotor interference on trim, open-loop response and key stability derivatives. For Chopper, the first-order impact of rotor–rotor interaction on flight dynamics in hover is small, but with an observed increased pitch and roll damping. In forward flight, on-axis bare airframe responses exhibit attenuation at the lower frequencies due to interference, and increased sensitivity of the pitch response to vertical perturbations in cruise. Finally, a real-time closed-loop simulation is performed with and without the interference model to assess the impact on rotor power during a representative science mission flight trajectory on Mars, showing that total increase in mechanical shaft power due to multirotor wake interactions does not exceed 5% throughout the flight.
This paper presents updates to The Rotorcraft Optimization Tools (RCOTools) package to streamline iterative rotorcraft comprehensive design. The work is presented in three parts. Part I. a brief introduction to our simplified API is shown, in addition to a new mission profile dashboard. Part II. demonstrates high-throughput using the embarrassingly parallel paradigm to produce large-scale datasets structured by simple design of experiments (DOE) as shown by our discussion on urban air mobility (UAM) emission minimization. Such datasets provide a necessary component for rapid database generation and supervised machine learning. Part III. the API is used to couple rotor performance and sizing optimization. A simple technique for ultra-fast hover calibration is given, as well as possible applications for neural network modeling in comprehensive design. These enhancements accelerate design workflows and enable data-driven approaches for next-generation urban air mobility and planetary rotorcraft concepts.
This paper proposes a first iteration towards a framework for enhancing the trustworthiness of machine learning in the health and usage monitoring of in-service helicopters. This bottom-up approach is based on our experience operating machine learning models for monitoring Airbus Helicopters' customer fleets. Key factors for improving trustworthy machine learning have been identified for both the development and execution phases, with specific methods defined for each enabler. These methods have been implemented in two use-cases involving machine learning models for regression tasks: monitoring the helicopter's main gearbox lubrication system, deployed in the FlyScan predictive maintenance service, and tracking the usage of the main rotor lead-lag damper loads. The results from both use cases show that confidence in machine learning model predictions can be effectively improved.
Rotorcraft continue to experience higher fatal accident rates compared to fixed-wing aircraft, primarily due to low altitude flight operations and reduced situational awareness in complex environments. A critical factor is the limited availability of accurate, up-to-date information on helipads and surrounding obstacles - such as trees, poles, and buildings - that pose significant risks during takeoff and landing. Existing resources, including the Federal Aviation Administration's heliport registry, are often outdated and incomplete, particularly for private or state-operated sites, and fail to report nearby obstacles. This lack of up-to-date data is largely due to privacy restrictions at certain locations and the high cost associated with comprehensive obstacle surveys. To address this challenge, we develop a deep learning (DL) framework that automatically detects helipads and nearby obstacles from high-resolution satellite imagery. Our approach combines Mask R-CNN for precise pixel-level helipad segmentation with Grounding DINO, a zero-shot vision-language model that identifies obstacles using flexible text prompts (e.g., "Pole", "Tree") without task-specific training. This text-guided, scalable detection method adapts to diverse and evolving operational settings. We validate our framework across helipads in the United States, and demonstrate strong performance in both helipad localization and obstacle detection. In addition, we build a web-based application that automates image processing, updates incorrect heliport coordinates, and provides obstacle reports. This work aims to enhance aviation safety, modernize infrastructure records, and deliver scalable tools to the aviation and machine learning communities.
Helicopter load monitoring and other health and usage monitoring system applications regularly involve large datasets and machine learning models. The amount of effort that could be devoted to training an optimal solution could be unlimited and quickly becomes prohibitive. However, there are important analyses and tools that could be implemented upfront to expedite finding well performing models as well as providing insight into the black-box machine learning models. In this work, we explore the use of the Gamma test and feature selection techniques to optimize input parameter sets and reduce dataset size, applying this approach to a helicopter load estimation problem. We demonstrate that we can remove half of the features from the input set and reduce the dataset by over 95% while still maintaining a similar level of accuracy and performance of the load estimation model. By reducing the number of features, we can produce simpler models, which are easier to train and explain.
This paper presents a comprehensive evaluation of machine learning approaches for real-time operational/ flight parameter estimation in large electric vertical takeoff and landing (eVTOL) vehicles, addressing the challenges of time-varying payloads and atmospheric disturbances in Advanced Air Mobility (AAM) missions. Artificial Neural Networks (ANN), Gaussian Process Regression (GPR), and Support Vector Machines (SVM), are compared for their ability to estimate gross weight (GW), longitudinal center of gravity position (CGx), and airspeed (Ux) using readily available flight control inputs and aircraft attitudes. The models are tested on clean data, turbulence-affected data, and reduced training data to assess performance trade-offs between computational cost and prediction accuracy. Results demonstrate that GPR consistently achieves the highest accuracy across all prediction tasks with maximum errors below 0.3% of nominal values, though at significantly higher computational cost compared to ANN and SVM. Under turbulent conditions, ANN and GPR exhibit notable reductions in accuracy, resulting in all three models (ANN, GPR, and SVM) achieving similar levels of prediction performance. Data reduction analysis reveals that using the Multipoint Maximal Variance Retention (MMVR) algorithm allows an 85–90% reduction in training data while keeping errors below 3.5%, striking an optimal balance between accuracy and efficiency. These analyses demonstrate the feasibility of ML-based operational/flight parameter estimation for AAM operations where direct measurement systems are impractical or cost-prohibitive.
Research into the feasibility of a scaled rim-drive propulsion product to enable ultra-heavy vertical lift (UHVL) is ongoing at the University of South Carolina in partnership with KRyanCreative, LLC, a start-up aerospace small business. The research team is advancing a superconductive design concept for a rotor system that delivers significant performance gains and flight envelope expansion disruptive to the vertical lift transportation sector. The team has conceived a novel electric tip-driven ducted propulsor to guide architectural and engineering investigations that improve hover and acoustic performance over current practice without penalty to weight and cost. This paper summarizes the data and assumptions that emerge from the systems engineering process of requirements decomposition for product realization. Requirements are categorized as to whether they are explicit (programs of record) or implied (comparable business case or as an alternative to a program of record). Risk reduction enroute to technical feasibility is addressed with a methodology that applies predictive analytics aided by artificial intelligence that will accelerate prototype fabrication by 2030 and fast track market incentives for multiple aviation technologies.
The work performed for the Adaptive Resilient Engineered Structures (ARES) program sponsored by the U.S. Army constitutes a trade study and resulting proposal for a structural demonstrator platform. The trade study was conducted using the Quality Function Deployment (QFD) process and a subsequent Artificial Intelligence (AI) exercise to find clusters of technologies for structural efficiency and resilience from Boeing's internal research activities. From a selection of approximately 150 technologies at different TRLs, Boeing subject matter experts (SMEs) for structural technologies identified several characteristics that could potentially determine the development of ARES structural demonstrator. Through the QFD process, the list of technologies was down selected about 50 unique technologies for consideration. The next stage of the QFD process entailed in identifying 37 different attributes or criteria long which each of these technologies would be assessed. They were grouped under two different categories: vehicle performance criteria and program performance criteria. Importance scores were provided by the SMEs independently and then a statistical approach for AI was used to distill them to 9 significant ones (labeled as 'Pillars') and a further distillation to 3 significant features (labeled as 'Super Metrics'). Clustering algorithms were then employed to group the set of technologies that could provide the resiliency targets sought for the demonstrator platform. The clusters were compared a hypothetical ideal platform to determine suitability and finally, 12 technologies merited attention toward the stated goals of the demonstrator platform.
This study investigates the stress concentration and damage tolerance of lug structures, with an application example using the horizontal tail plane lug of a light utility helicopter. Using Finite Element Analysis (FEA), stress distributions around the lug hole were simulated under varying load conditions to understand how different loading angles and magnitudes affect stress concentrations. A machine learning approach was employed to predict stress distributions based on a dataset generated from FEA simulations. Several regression models were tested and Random Forest Regression model yields the best predictive accuracy among the others. The study also incorporates a flaw tolerance analysis using the NASGRO® software to calculate the crack growth characteristics of the horizontal tail plane lug structure under service loading. The results highlight the importance of stress distribution variability and identifying the most critical point of the lug structure under service loading with changing angle and amplitude. This research provides insights into the structural integrity of lug components under dynamic loading, contributing to more reliable flaw tolerance assessments for aerospace applications using machine learning techniques.
The advent of electric propulsion technology has led to a paradigm shift in aircraft design over the past few decades. This shift has expanded the possibilities for design and optimization processes more than at any previous time. To support these advancements, efficient flight dynamics simulation models that can be employed in iterative optimization and design processes are essential. Among the modules of a typical flight dynamics framework—namely, control, flight dynamics, and aerodynamics—the aerodynamics module, which includes the rotor performance model, generally demands the most computational effort, thereby limiting simulation efficiency. In this study, a novel machine learning (ML)-assisted flight dynamics framework is developed, incorporating a Neural Network Blade Element Theory (NN-BET) model as the rotor performance module. The results show a 7- to 8-fold reduction in computational time compared to fast, physics-based frameworks utilizing efficient Blade Element Momentum Theory (BEMT) models, without compromising predictive accuracy. Furthermore, the modular architecture of the framework allows for easy adaptation to a wide range of practical applications by replacing modules with functionally equivalent alternatives. The demonstrated accuracy and computational efficiency of the proposed flight dynamics framework make it a highly promising candidate for optimization and design applications.
Simulation data consisting of multiple fidelity levels were generated using Graphical Processing Unit (GPU) resources on the NASA supercomputers. First, two large aerodynamic simulation databases were generated for geometric perturbations over a range of flight conditions for a hex-rotor bi-plane tailsitter aircraft. Results were visualized using the NASA Advanced Supercomputing Division's Hyperwall to improve the geometric design constraints. More than 3,000 full aircraft aerodynamic simulations were run using GPU enabled OVERFLOW with an actuator disk model to generate the airframe aerodynamic database. These simulations were completed in roughly 1.5 weeks on 32 GPU nodes using 128 NVIDIA V100 GPUs. Surrogate modeling techniques including Gaussian Process Regression (GPR), sparse GPRs, and a variety of Neural Networks (NNs) were used to create surrogate models to predict airfoil aerodynamic performance as well as airframe aerodynamics as a function of flight condition, airframe geometry, and rotor control input. These surrogate models were combined with additional Python modules predicting aircraft mass and inertia to generate another 3,000 aircraft simulations in CAMRAD-II in less than six hours using 240 Central Processing Unit (CPU) cores. Stability and control derivative matrices were obtained from the output to evaluate the open-loop characteristics. Lastly, the optimization framework was setup to allow simultaneous optimization of the aircraft and flight condition. The framework can now be used to optimize the aircraft for various objectives such as maximum range, endurance, or payload while satisfying constraints on controllability. This work brings higher-fidelity simulation data into the earlier stages of conceptual design, improving accuracy of the results and reducing the risk of missing critical design issues.
This study investigates the application of neural network architectures to predict control inputs required to replicate rotorcraft responses under vertical gust disturbances. Two modeling approaches are developed: the Control Equivalent Gust Input (CEGI) model, using body-axis inputs and the Rotor Control Equivalent Gust Input (RCEGI) model using rotor-specific inputs. Initial models employed single-input single-output (SISO) LSTM networks, which demonstrated limitations in capturing transient behavior and exhibited delay in predicted control inputs. By incorporating multiple vehicle response features and increasing the number of hidden neurons, multiple-input single-output (MISO) architectures significantly improved accuracy and reduced Root Mean Square Error (RMSE). Further enhancement was achieved by implementing bidirectional LSTM (BiLSTM) layers, which reduced both delay and transient error. Comparisons with inverted linear time-invariant (LTI) approximations showed that neural networks provided superior performance, particularly in modeling nonlinear dynamics. The results highlight the potential of deep learning approaches to improve the accuracy of control input mapping and inform real-time control strategies in unsteady flight environments.
Traditional safe-life methodologies for rotorcraft structural components often result in overly conservative life estimates, increasing maintenance costs and reducing aircraft availability. This study explores the integration of digital twin concepts with probabilistic modeling and machine learning to enhance structural life assessment, demonstrated through a practical case involving the Royal Canadian Air Force CH-146 Griffon helicopter. A probabilistic fatigue model determines a fatigue life distribution by incorporating material variability and uncertain operational loads inferred directly from flight data. Unlike conventional approaches, this method dynamically estimates load spectra, including uncertainty instead of relying on conservative assumptions. Monte Carlo simulations are used to quantify structural risk and assess the impact of load and material uncertainties. Sensitivity analyses highlight these uncertainties’ contributions to failure probability. The proposed approach provides probabilistic life predictions, supporting risk-based maintenance strategies to potentially optimize operational efficiency. The long-term goal is to develop an adaptive digital twin model that continuously updates with new operational flight data, enhancing predictive accuracy for helicopter fleet management.
ABSTRACT Automatic guided vehicles (AGV) have made big inroads in the automation of assembly plants and warehouse operations. There are thousands of AGV units in operation at OEM supplier and service facilities worldwide in virtually every major manufacturing and distribution sector. Although today’s AGV systems can be reconfigured and adapted to meet changes in operation and need, their adaptability is often limited because of inadequacies in current systems. This paper describes a wireless navigated (WN) omni-directional (OD) autonomous guided vehicle (AGV) that incorporates three technical innovations that address the shortfalls. The AGV features consist of: 1) A newly developed integrated wireless navigation technology to allow rapid rerouting of navigation pathways; 2) Omnidirectional wheels to move independently in different directions; 3) Modular space frame construction to conveniently resize and reshape the AGV platform. It includes an overview of the AGVs technical features and how the flexibility and agility can be adapted to fit military and commercial application. The AGV is being evaluated as a mobile work station platform and a precise material handling robot.
As part of maintenance improvement on helicopters, Airbus Helicopters has made available a proactive analysis service based on Health and Usage Monitoring System data generated during the flight. The present paper describes the new approach used to detect and classify any changes in time series behavior thanks to A.I. (Artificial Intelligence) especially computer vision. This new approach is more efficient and relevant than the classical approach based one statistical law [Ref 1]; in fact, it is acting, as the human eye, which is able to identify easily any abrupt change on the time series, and classifies it, whether Machine learning or Deep Neural Networks both have shown excellent results in term of classification accuracy. First part of this paper highlights how the learning data were prepared, then the second and the third parts give more details about how the time series are transformed into image presentation and how the different Artificial Intelligence models were selected and feed, ultimately the obtained results.
This paper presents a new approach, variant of the Direct Load Recognition (DLR) methodology, to estimate the main rotor (MR) pitch-link load on customer flights. The original DLR methodology is based on the combination of a harmonic decomposition and the use of Machine Learning algorithms. The DLR variant replaces the harmonic decomposition by a wavelet decomposition. The application of this paper consists in two parts. First, the comparison between the original DLR and DLR variant on prototype flight test data. Two results are highlighted in this part. The capacity of representation of the pitch-link load is better for the wavelet decomposition. The modelling of its coefficients enables to slightly improve the pitch-link load estimation, especially on the high load values having more impact on the fatigue computation. This first part allows to study the feasibility of the DLR variant to estimate the pitch-link load. The second part of this paper focuses on the application of the pitch-link load estimator built with DLR variant on the H175 fleet. From the estimated MR pitch-link load, the MR pitch horn damage is derived and compared to the Design Usage Spectrum (DUS), used today for the certification. The damage of all the studied customer aircraft is well below the DUS, showing a potential gain in component life time. The results of this paper manifest the advantage of the DLR methodology, and more precisely DLR variant methodology, for predictive maintenance on the MR pitch horn, that is to adapt the maintenance to the aircraft usage.
NASA's 4th New Frontiers Mission is the Titan Dragonfly relocatable lander. This coaxial quadrotor vehicle will be launched on a rocket to Titan in 2028. Following a gravity assisted Earth flyby and an approximate 6-year transit, Dragonfly will enter the Titan atmosphere around 2034 with the goal of exploring Titan's pre-biotic chemistry and habitability. The multirotor design for this unique application has continually evolved since 2016 with constraints such as Titan's cryogenic atmosphere at 95 Kelvin (-288 F), gravity 14% that of Earth's, atmospheric density 440% of standard sea-level air, and the inability to test the entire system together under all these conditions until the first flight on Titan. This paper focuses on rotor design aspects of the Dragonfly lander and introduces a novel framework for multirotor design optimization considering multiple flight conditions. The methodology leverages machine learning methods and is demonstrated in the context of Dragonfly. A new OVERFLOW Machine Learning Airfoil Performance (PALMO) database is first presented. PALMO is then wrapped inside a Bayesian optimization framework and applied to a 4-rotor system (one side of the Dragonfly lander). Training data is generated on each iteration of the optimization using the CAMRAD-II comprehensive analysis software to evaluate successive rotor designs in multiple relevant flight conditions. An optimal design for the 4-rotor system was found with approximately 900 rotor designs analyzed in CAMRAD-II, which required 9 million queries of the PALMO surrogate models. This demonstration case evaluated 10,000,000 potential candidate rotor designs in 5.5 hours on 114 CPU cores using uniform inflow, and in 27.8 hours using the prescribed wake model. This work thus enables mid-fidelity rotor design optimization without requiring access to high-performance computing.
This paper presents an original method that takes advantage of existing large in-service flight data, damper load Machine Learning models as well as the inventory of degraded dampers (elastomeric part), to link the estimated loads and operational conditions to damper degradation cases. The Machine Learning models are trained on flight test campaigns data, and then applied on in-service helicopter data to estimate damper loads as a function of flight parameters. The estimated load history is then used as an input to generate engineering load indicators. These latter, jointly with operational and usage data, are correlated with the reported dampers' degradation observations. Finally, an explainability mechanism is investigated to better understand the Machine Learning models inferences, opening perspectives towards precise damper degradation root causes identification. The obtained results are promising, showing that the occurrence of damper degradation correlates with load history and helicopter operations.
The airframe digital twin analysis framework developed at the National Research of Canada is being transposed to safe life applications for rotorcraft components. A probabilistic safe life prediction approach, consisting of uncertain material property data and uncertain load spectra is used to calculate risk assessment metrics, such as the cumulative probability of failure, the hazard rate, and the average hazard rate as a function of time. A demonstration of this approach is presented for a CH-146 Griffon component, for which the uncertain loads are estimated from a model developed through machine learning. This preliminary assessment shows the feasibility of using digital twin concepts as a viable alternative to traditional deterministic life predictions, with the potential to reduce maintenance costs and increase aircraft availability.
This paper reports on the initial implementation of Machine Learning (ML) for predicting the workload experienced by a pilot when performing a recovery to a naval ship. Pilots classify their workload for each landing by providing a subjective rating, which is used to determine the ship-helicopter operating limit (SHOL). Different workload metrics have been trialled to bridge the gap between pilot subjective ratings and objective flight data. With hundreds of different helicopter, ship and airwake parameters available to examine, ML provides an approach to understanding the complex interactions between these variables. This paper looks at the initial results obtained by applying ML techniques to train a classification algorithm with pilot control input data. Preliminary results showed 77.14% accuracy when training a Linear Discriminant algorithm to predict pilot workload from cyclic, collective, and pedal input data.
For the last few decades, Canada's National Research Council (NRC) has been at the forefront in analyzing dynamic systems and developing tools to construct aircraft models based on flight test data. With a fixed and rotary-wing aircraft fleet available, NRC has the capability to perform leading edge R&D System Identification (SI); this worldleading SI technology has been developed and has assisted industry partners, Department of National Defense (DND), and various universities in aircraft simulation and development. As a result, NRC has gained extensive experience in modeling aircraft using SI techniques. In collaboration with CAE, this paper demonstrates the acceleration of the NRC's current flight modeling techniques, highlighting recent advances in Artificial Intelligence (AI) and Machine Learning (ML). A new Bayesian ML software is being developed to identify a 6 degrees of freedom (6-DoF) quasisteady model using simulated flight test data. To achieve this, data from the CAE Sample electric Vertical Take-Off and Landing (eVTOL) simulation platform vehicle during hover maneuvers is utilized. Additionally, this paper presents results on extending the model to include rotor dynamics using the classical SI approach for comparison purposes. In summary, all methods provide a high-fidelity model; with the higher model structure, the vertical acceleration match was noticeably better.
The rotorcraft community faces significantly higher accident rates compared to fixed-wing commercial aircraft, underscoring the critical need for enhanced safety measures. While Helicopter Flight Data Monitoring programs hold promise in improving safety, their widespread adoption remains limited, partly due to challenges associated with the acquisition and analysis of flight data. This paper proposes a Deep Learning (DL) solution to address safety concerns within the rotorcraft community by efficiently acquiring and analyzing flight data for a more automated and comprehensive safety assessment. Specifically, we leverage data obtained with cost-effective off-the-shelf cameras, and process it through Convolutional Neural Networks for automated detection and classification of gauges from several helicopters' cockpits. Our DL pipeline integrates a classifier for helicopter identification, an object detector for cockpit gauges detection and classification, and a network to infer the reading of each detected gauge. The contribution of this work is two-fold: (1) enhance rotorcraft safety by developing a DL framework capable of detecting, classifying, and inferring gauge readings for different helicopter types, and (2) boost research in the field by constructing a curated dataset valuable for aviation and machine learning communities.
Safety professionals receive data from internal and external sources, then manually determine whether the issue constitutes a safety hazard. Many reports are received, and each report is reviewed, then investigated further, using a tedious, labor intensive, and possibly error prone process. In the course of reaching a decision, human bias is inevitable - any two humans could reach different conclusions, and the same individual human could draw different conclusions on different days. As technology has advanced, numerous approaches have been pursued, attempting to reduce human bias and improve both efficiency and effectiveness of the process. In recent years, moderate success was achieved, which provided accuracy rates near 85% but continued refinement did not achieve acceptable results. In early 2023, the challenge was given to a new team, and within a few months, state-of-the-art Artificial Intelligence/Machine Learning data analytics techniques were utilized to aid in safety data analysis efforts, which resulted in high accuracy and efficiency, with reduced human bias.
This study investigates the use of machine learning (ML) models to estimate the gross weight (GW), the longitudinal position of the center of gravity (CGx), and 1/rev cyclic flapping angles (Δ1c and Δ1s) of a compound helicopter with three redundant controls - main rotor RPM, collective propeller thrust, and stabilator angle. Neural Network (NN), Gaussian Process for Regression (GPR), and Support Vector Machine (SVM) algorithms are employed to develop estimation models using supervised training. The airspeed, redundant controls, main rotor controls, aircraft attitudes, and main rotor torque are selected as input variables (predictors) to the models due to their accessibility through the aircraft Health and Usage Monitoring System (HUMS). The dataset is split into low-speed and high-speed regimes to compare the prediction accuracy and training cost of separate regime models against a combined full-regime model. Separate airspeed regime GPR models showed superior performance in GW estimation, with higher accuracy and cost-effectiveness compared to a single full-regime model. For CG estimation, GPR again outperformed NN and SVM, although the maximum outlier errors increase significantly if a 95% confidence interval is considered. Finally, for 1/rev cyclic flapping angle predictions, SVM estimations, though not superior to GPR or NN, were acceptable and had a significantly lower computational cost. The study also examined the importance of predictors, highlighting that, on average, certain predictors like rotor RPM and rotor torque are less influential, but their removal degraded performance and had no cost benefit.
This study presents a statistical approach for detecting and estimating damage to multicopter propellers through a comprehensive probabilistic model. The methodology is derived from model-based analysis and applied within the time series statistical techniques. This research accounts for uncertainties in the estimation process and offers confidence intervals for assessing the extent of damage to the propellers. The framework employs functionally pooled (FP) models characterized by parameters that depend on damage sizes, proper statistical estimation, and decision-making schemes. The validation and assessment are assessed via a hexacopter flying in circles with a constant velocity and altitude under turbulence. The damage size ranges from healthy to 10 mm. The method achieves fast damage detection and precise magnitude estimation based on a segment of a single measured signal obtained from aircraft sensors during flight.
This paper investigates the feasibility of using machine learning to predict whirl flutter bifurcation diagrams. The machine learning techniques selected for the study are XGBoost and the long short-term memory neural network. These techniques are selected for their suitability for sequential and nonlinear data. The techniques are investigated for a propeller-nacelle test case with polynomial structural nonlinearities resulting in supercritical or subcritical whirl limit-cycle oscillations. The techniques are trained to learn the bifurcation diagram for the amplitude variation of pitch angle limit-cycle oscillations of the propeller-nacelle system as a function of the forward speed for various levels of cubic structural nonlinearity. Bifurcation diagram learning and testing data are generated using the bifurcation forecasting method. XGBoost is computationally faster to train but less accurate for low amounts of learning data, especially for the most weakly and strongly nonlinear cases. The long short-term memory neural network is more computationally expensive to train but shows a less scattered error pattern for sparse learning data. However, it is sensitive to the amplitude resolution of the bifurcation diagrams. The approach to sample the cubic nonlinearity range does not significantly impact the results once the techniques have a sufficient amount of data to learn from. The data requirements observed in the study suggest that, for these techniques, direct learning of bifurcation diagrams may not scale beyond a handful of input parameters.
An essential component for the advancement of autonomous flight lies in the development of an intelligent routing system designed to facilitate the maintenance and troubleshooting of electrical wiring. Utilizing software with the capability to present routed paths in a computer-aided design (CAD) format allows for a detailed representation of the rules governing the layout of wiring around structural supports and distribution channels. Despite this, three-dimensional (3D) methodologies have yet to fully incorporate critical data related to the characterization of individual wiring signals, hindering automatic routing. This paper underscores a competitive edge that can be achieved by expanding 3D capabilities to accurately depict the current state of wiring signals in terms of temperature, humidity, electromagnetic frequency, amperage, and other relevant factors. Achieving this involves integrating a non-intrusive smart sensing technology with the intelligent routing system to monitor and diagnose the health and integrity of the wiring system. With this integration, a more robust artificial intelligence (AI) system can leverage the obtained data to make more precise decisions, enhancing overall system performance.
In the field of aerodynamics, there is a growing need for rapid load prediction in engineering applications. Surrogate modeling offers a promising solution, providing faster results compared to high-fidelity computational models. This study focuses on a Machine Learning (ML) framework tailored for surrogate modeling, specifically for integrated aerodynamic load predictions in aircraft design. Central to this framework is a Deep Neural Network (DNN) component capable of handling both steady-state and fluctuating aerodynamics. A key challenge for surrogate models lies in maintaining prediction accuracy, especially in scenarios involving nonlinear flow phenomena like flow separation and transonic shifts. To address these challenges, we introduce a two-step physics-state predictor that integrates an intermediate Convolutional Neural Network (CNN) component. This approach enhances the surrogate model's capability to accurately represent dynamic separated flows and other nonlinear patterns without relying on unrealistic user inputs. Results are presented for NACA0015 dynamic stall predictions for two different physics-state inputs.
Fusion Artificial Intelligence Link Synchronization Array for eVTOL Systems (FAILSAFES™) is a resilient and redundant timing and positioning architecture based on low Size, Weight, Power, and Cost (SWaP-C) RF Ranging links for eVTOL systems navigating with Global Navigation Satellite System (GNSS) in degraded or denied environments. This paper describes the overall FAILSAFES™ concept and discusses the underlying Complementary Positioning, Navigation, and Timing (CPNT) capabilities based on ENSCO's PicoRangerTM Array technology (PRAT). PRAT provides an array of low-cost RF ranging links between FAILSAFES™ ground stations and aircrafts to support navigation and timing distribution in GNSS degraded or denied environments. This paper will explore components of FAILSAFES™ and discuss initial PRAT based fusion results with respect to frequency and time stability.
Along with unique and challenging development concerns, target hardware deployment concerns exist for artificial intelligence (AI) and machine learning (ML) applications. Those deployment concerns should be addressed in the planning phase and consist of the issues surrounding the target hardware selection and the certifiability/qualifiable of the target hardware for the AI/ML model deployment. These concerns center around certification issues identified for multi-core processors (MCP), where those MCP issues are amplified for graphics processor units (GPUs) when they are used for general computing. While the use of complex graphics processors for general computing is being reconciled for flight critical applications, the reduction of these concerns is possible through design specific target hardware choices, e.g., selection of Field Programmable Gate Array (FPGA) devices or other certifiable approaches. This paper explores these concerns and proposes design specific target hardware choice strategies to mitigate those concerns.
ABSTRACT
ABSTRACT
ABSTRACT
ABSTRACT
Items per page:
50
1 – 50 of 993