Differentiable predictive control: Deep learning alternative to explicit model predictive control for unknown nonlinear systems
Not Available
97 MATHEMATICS AND COMPUTING↗
SEARCH · Engineering Papers
Results for “Predictive Modeling”
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
SU-F-J-199: Predictive Models for Cone Beam CT-Based Online Verification of Pencil Beam Scanning Proton Therapy
Purpose: To utilize online CBCT scans to develop models for predicting DVH metrics in proton therapy of head and neck tumors. Methods: Nine patients with locally advanced oropharyngeal cancer were retrospectively selected in this study. Deformable image registration was applied to the simulation CT, target volumes, and organs at risk (OARs) contours onto each weekly CBCT scan. Intensity modulated proton therapy (IMPT) treatment plans were created on the simulation CT and forward calculated onto each corrected CBCT scan. Thirty six potentially predictive metrics were extracted from each corrected CBCT. These features include minimum/maximum/mean over and under-ranges at the proximal and distal surface of PTV volumes, and geometrical and water equivalent distance between PTV and each OARs. Principal component analysis (PCA) was used to reduce the dimension of the extracted features. Three principal components were found to account for over 90% of variances in those features. Datasets from eight patients were used to train a machine learning model to fit these principal components with DVH metrics (dose to 95% and 5% of PTV, mean dose or max dose to OARs) from the forward calculated dose on each corrected CBCT. The accuracy of this model was verified on the datasets from the 9th patient. Results: The predicted changes of DVH metrics from the model were in good agreement with actual values calculated on corrected CBCT images. Median differences were within 1 Gy for most DVH metrics except for larynx and constrictor mean dose. However, a large spread of the differences was observed, indicating additional training datasets and predictive features are needed to improve the model. Conclusion: Intensity corrected CBCT scans hold the potential to be used for online verification of proton therapy and prediction of delivered dose distributions.
60 APPLIED LIFE SCIENCES↗
Model predictive combustion control of a Gasoline Compression Ignition engine
Gasoline Compression Ignition is a novel combustion concept that derives its superiority from the high compression ratio of a compression ignition engine as well as the properties of gasoline fuel, such as longer ignition delay and higher volatility compared to diesel. Here, this combustion concept was experimentally tested on a 12.4L Class 8 truck engine. Based on these experimental data, prior efforts by the authors focused on the development of an engine model for a heavy-duty engine operating on a low-reactivity fuel. This engine model was leveraged within this study to investigate a combustion control strategy at different engine conditions and injection methods and was augmented to incorporate cycle-to-cycle combustion variations. State estimation is performed by means of a Kalman filter which feeds into a model predictive controller. The model predictive controller chooses control actions based on a predefined cost function under consideration of bounds reflecting physical constraints. The engine model was utilized to establish a state-space model that serves the Kalman filter and model predictive controller for estimation and prediction. A comparative study investigating control actions and engine behavior was performed with and without limiting in-cylinder peak pressure as well as combustion noise, which is of particular interest for early pilot injection strategies. In addition, the proposed control architecture was investigated at two different levels of cycle-to-cycle variations and compared to the performance of a control structure with input disturbance rejection. For increased cycle-to-cycle variations, disturbance estimation reduces state fluctuations and control effort. In general, this investigation highlights control aspects specific to a compression-ignited combustion regime with low-reactivity fuel. The control algorithm is able to maintain the desired references for brake mean effective pressure and combustion phasing while controlling peak in-cylinder pressure and combustion noise.
42 ENGINEERING↗
A Machine Learning Based System Performance Prediction Model for Transportable FHR
A machine learning based system performance prediction model is currently created to support the development of autonomous control for small reactors, such as the Transportable Fluoride-salt-cooled High-temperature Reactor (TFHR), which has a 20 MWth compact core proposed by MIT for remote sites. The prediction model is constructed using support vector regression (SVR) with training data generated by RELAP5. A particle filtering framework is utilized to estimate and update model parameters with instrument measurements. Verifications of the prediction and filtering models have been carried out using TFHR reactivity insertion cases. Satisfactory performance in predicting the core behavior and in recognizing the inserted reactivity rate is concluded. (authors)
21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗
Enhanced Co-Expression Extrapolation (COXEN) Gene Selection Method for Building Anti-Cancer Drug Response Prediction Models
The co-expression extrapolation (COXEN) method has been successfully used in multiple studies to select genes for predicting the response of tumor cells to a specific drug treatment. Here, we enhance the COXEN method to select genes that are predictive of the efficacies of multiple drugs for building general drug response prediction models that are not specific to a particular drug. The enhanced COXEN method first ranks the genes according to their prediction power for each individual drug and then takes a union of top predictive genes of all the drugs, among which the algorithm further selects genes whose co-expression patterns are well preserved between cancer cases for building prediction models. We apply the proposed method on benchmark in vitro drug screening datasets and compare the performance of prediction models built based on the genes selected by the enhanced COXEN method to that of models built on genes selected by the original COXEN method and randomly picked genes. Models built with the enhanced COXEN method always present a statistically significantly improved prediction performance (adjusted p-value ≤ 0.05). Our results demonstrate the enhanced COXEN method can dramatically increase the power of gene expression data for predicting drug response.
60 APPLIED LIFE SCIENCES↗
Wake Vortex Prediction Models for Decay and Transport Within Stratified Environments
This paper proposes two simple models to predict vortex transport and decay. The models are determined empirically from results of three-dimensional large eddy simulations, and are applicable to wake vortices out of ground effect and not subjected to environmental winds. The results, from the large eddy simulations assume a range of ambient turbulence and stratification levels. The models and the results from the large eddy simulations support the hypothesis that the decay of the vortex hazard is decoupled from its change in descent rate.
Switzer, George F.↗
Thermal barrier coating life prediction model development, phase 2
The objective of this program was to generate a life prediction model for electron-beam-physical vapor deposited (EB-PVD) zirconia thermal barrier coating (TBC) on gas turbine engine components. Specific activities involved in development of the EB-PVD life prediction model included measurement of EB-PVD ceramic physical and mechanical properties and adherence strength, measurement of the thermally grown oxide (TGO) growth kinetics, generation of quantitative cyclic thermal spallation life data, and development of a spallation life prediction model. Life data useful for model development was obtained by exposing instrumented, EB-PVD ceramic coated cylindrical specimens in a jet fueled burner rig. Monotonic compression and tensile mechanical tests and physical property tests were conducted to obtain the EB-PVD ceramic behavior required for burner rig specimen analysis. As part of that effort, a nonlinear constitutive model was developed for the EB-PVD ceramic. Spallation failure of the EB-PVD TBC system consistently occurred at the TGO-metal interface. Calculated out-of-plane stresses were a small fraction of that required to statically fail the TGO. Thus, EB-PVD spallation was attributed to the interfacial cracking caused by in-plane TGO strains. Since TGO mechanical properties were not measured in this program, calculation of the burner rig specimen TGO in-plane strains was performed by using alumina properties. A life model based on maximum in-plane TGO tensile mechanical strain and TGO thickness correlated the burner rig specimen EB-PVD ceramic spallation lives within a factor of about plus or minus 2X.
Meier, Susan Manning↗
Verification by Remote Sensing of an Oil Slick Movement Prediction Model
The author has identified the following significant results. LANDSAT, aircraft, ships, and air-dropped current drogues were deployed to determine current circulation and to track oil slick movement on four different dates in Delaware Bay. Results were used to verify a predictive model for oil slick movement and dispersion. The model predicts the behavior of oil slicks given their size, location, tidal stage (current), weather (wind), and nature of crude. Both LANDSAT satellites provided valuable data on gross circulation patterns and convergent coastal fronts which by capturing oil slicks significantly influence their movement and dispersion.
Klemas, V.↗
Closing the Contrast Gap Between Testbed and Model Prediction with WFIRST-CGI Shaped Pupil Coronagraph
JPL has recently passed an important milestone in its technology development for a proposed NASA WFIRST mission: demonstration of a better than 1x10-8 contrast over broad bandwidth (10%) on both shaped pupil coronagraph (SPC) and hybrid Lyot coronagraph (HLC) testbeds with WFIRST telescope optics. Challenges remain, however, in the technology readiness for the proposed mission. One such is the discrepancy between the achieved contrast on testbeds and their corresponding model predictions. A series of testbed diagnosis and modeling activity were planned and (some) carried out on SPC testbed in order to close the gap. A very useful tool we developed was to analyze the measured testbed Jacobian and compare with its model version that was used to control image plane speckle pattern. The difference between these two is the error in the control Jacobian. When the error, which includes both amplitude and phase, was inserted into a separate prediction model with nominal Jacobian, the model prediction became closely matching the SPC testbed behavior in both contrast floor and contrast convergence speed. This offers a new perspective and is a step closer toward model validation for high contrast coronagraph. Further Jacobian analysis and modeling provided clues to the possible sources for the mismatch: the DM misregistration and the underrepresented testbed optics wavefront error (WFE) and the deformable mirror (DM) setting for correcting this WFE. These analyses suggested that high contrast coronagraph has a tight tolerance in the accuracy of its control Jacobian. Modifications to both testbed control model as well as prediction model are being implemented and future works discussed.
diffraction modeling↗
Accurate Holdup Calculations with Predictive Modeling & Data Integration
In facilities that process special nuclear material (SNM) it is important to account accurately for the fissile material that enters and leaves the plant. Although there are many stages and processes through which materials must be traced and measured, the focus of this project is material that is “held-up” in equipment, pipes, and ducts during normal operation and that can accumulate over time into significant quantities. Accurately estimating the holdup is essential for proper SNM accounting (vis-à-vis nuclear non-proliferation), criticality and radiation safety, waste management, and efficient plant operation. Usually it is not possible to directly measure the holdup quantity and location, so these must be inferred from measured radiation fields, primarily gamma and less frequently neutrons. Current methods to quantify holdup, i.e. Generalized Geometry Holdup (GGH), primarily rely on simple source configurations and crude radiation transport models aided by ad hoc correction factors. This project seeks an alternate method of performing measurement-based holdup calculations using a predictive model that employs state-of-the-art radiation transport codes capable of accurately simulating such situations. Inverse and data assimilation methods use the forward transport model to search for a source configuration that best matches the measured data and simultaneously provide an estimate of the level of confidence in the correctness of such configuration. In this work the holdup problem is re-interpreted as an inverse problem that is under-determined, hence may permit multiple solutions. A probabilistic approach is applied to solving the resulting inverse problem. This approach rates possible solutions according to their plausibility given the measurements and initial information. This is accomplished through the use of Bayes’ Theorem that resolves the issue of multiple solutions by giving an estimate of the probability of observing each possible solution. To use Bayes’ Theorem, one must have a model y(x) that maps the state variables x (the solution in this case) to the measurements y. In this case, the unknown state variables are the configuration and composition of the heldup SNM. The measurements are the detector readings. Thus, the natural model is neutral-particle radiation transport where a wealth of computational tools exists for performing these simulations accurately and efficiently. The combination of predictive model and Bayesian inference forms the Data Integration with Modeled Predictions (DIMP) method that serves as foundation for this project. The cost functional describing the model-to-data misfit is computed via a norm created by the inverse of the covariance matrix of the model parameters and responses. Since the model y(x) for the holdup problem is nonlinear, a nonlinear optimization on Q is conducted via Newton-type iterative methods to find the optimal values of the model parameters x. This project comprised a collaboration between NC State University (NCSU), the University of South Carolina (USC), and Oak Ridge National Laboratory (ORNL). The project was originally proposed in seven main tasks with an eighth contingency task to be performed if time and funding permitted; in fact time did not permit commencement of the contingency task and it was not performed. The remaining tasks involved holdup analysis with gamma detection strategies and separately with neutrons based on coincidence counting. Early in the project, and upon consultation with experts in coincidence counting it became evident that this approach is not viable for holdup applications and this task was replaced with an alternative, but valuable investigation that was carried out by the USC partner. Nevertheless, the experimental 4 measurements at ORNL of both gamma and neutron sources for the purpose of constructing Detector Response Functions (DRFs) with the associated uncertainties were indeed completed.
11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
Freeze Prediction Model
Measurements of wind speed, net irradiation, and of air, soil, and dew point temperatures in an orchard at the Rock Springs Agricultural Research Center, as well as topographical and climatological data and a description of the major apple growing regions of Pennsylvania were supplied to the University of Florida for use in running the P-model, freeze prediction program. Results show that the P-model appears to have considerable applicability to conditions in Pennsylvania. Even though modifications may have to be made for use in the fruit growing regions, there are advantages for fruit growers with the model in its present form.
Morrow, C. T.↗
SU-F-T-119: Development of Heart Prediction Model to Increase Accuracy of Dose Reconstruction for Radiotherapy Patients
Purpose: To assess individual variation in heart volume and location in order to develop a prediction model of the heart. This heart prediction model will be used to calculate individualized heart doses for radiotherapy patients in epidemiological studies. Methods: Chest CT images for 30 adult male and 30 adult female patients were obtained from NIH Clinical Center. Image-analysis computer programs were used to segment the whole heart and 8 sub-regions and to measure the volume of each sub- region and the dimension of the whole heart. An analytical dosimetry method was used for the 30 adult female patients to estimate mean heart dose during conventional left breast radiotherapy. Results: The average volumes of the whole heart were 803.37 cm{sup 3} (COV 18.8%) and 570.19 cm{sup 3} (COV 18.8%) for adult male and female patients, respectively, which are comparable with the international reference volumes of 807.69 cm{sup 3} for males and 596.15 cm{sup 3} for females. Some patient characteristics were strongly correlated (R{sup 2}>0.5) with heart volume and heart dimensions (e.g., Body Mass Index vs. heart depth in males: R{sup 2}=0.54; weight vs. heart width in the adult females: R{sup 2}=0.63). We found that the mean heart dose 3.805 Gy (assuming prescribed dose of 50 Gy) in the breast radiotherapy simulations of the 30 adult females could be an underestimate (up to 1.6-fold) or overestimate (up to 1.8-fold) of the patient-specific heart dose. Conclusion: The study showed the significant variation in patient heart volumes and dimensions, resulting in substantial dose errors when a single average heart model is used for retrospective dose reconstruction. We are completing a multivariate analysis to develop a prediction model of the heart. This model will increase accuracy in dose reconstruction for radiotherapy patients and allow us to individualize heart dose calculations for patients whose CT images are not available.
60 APPLIED LIFE SCIENCES↗
Rate-Based Model Predictive Control of Turbofan Engine Clearance
An innovative model predictive control strategy is developed for control of nonlinear aircraft propulsion systems and sub-systems. At the heart of the controller is a rate-based linear parameter-varying model that propagates the state derivatives across the prediction horizon, extending prediction fidelity to transient regimes where conventional models begin to lose validity. The new control law is applied to a demanding active clearance control application, where the objectives are to tightly regulate blade tip clearances and also anticipate and avoid detrimental blade-shroud rub occurrences by optimally maintaining a predefined minimum clearance. Simulation results verify that the rate-based controller is capable of satisfying the objectives during realistic flight scenarios where both a conventional Jacobian-based model predictive control law and an unconstrained linear-quadratic optimal controller are incapable of doing so. The controller is evaluated using a variety of different actuators, illustrating the efficacy and versatility of the control approach. It is concluded that the new strategy has promise for this and other nonlinear aerospace applications that place high importance on the attainment of control objectives during transient regimes.
DeCastro, Jonathan A.↗
Empirical testing of an analytical model predicting electrical isolation of photovoltaic models
A major design requirement for photovoltaic modules is that the encapsulation system be capable of withstanding large DC potentials without electrical breakdown. Presented is a simple analytical model which can be used to estimate material thickness to meet this requirement for a candidate encapsulation system or to predict the breakdown voltage of an existing module design. A series of electrical tests to verify the model are described in detail. The results of these verification tests confirmed the utility of the analytical model for preliminary design of photovoltaic modules.
Garcia, A., III↗
The Effect of Nondeterministic Parameters on Shock-Associated Noise Prediction Modeling
Engineering applications for aircraft noise prediction contain models for physical phenomenon that enable solutions to be computed quickly. These models contain parameters that have an uncertainty not accounted for in the solution. To include uncertainty in the solution, nondeterministic computational methods are applied. Using prediction models for supersonic jet broadband shock-associated noise, fixed model parameters are replaced by probability distributions to illustrate one of these methods. The results show the impact of using nondeterministic parameters both on estimating the model output uncertainty and on the model spectral level prediction. In addition, a global sensitivity analysis is used to determine the influence of the model parameters on the output, and to identify the parameters with the least influence on model output.
Dahl, Milo D.↗
A neural-network-enhanced parameter-varying framework for multi-objective model predictive control applied to buildings
Management of the electrical grid is becoming more complex due to the increased penetration of alternative energy generation technologies and a broadening diversity of electric loads. This complexity creates challenges in balancing demand and generation that can increase the potential for grid instabilities. One effective way to address this issue is to leverage previously unexploited demand flexibility through advanced control strategies. In this work, we propose an advanced control method, called adaptive neural parameter-varying model predictive control (ANPV-MPC), to control the temperature and energy consumption of a building via its Heating, Ventilation, and Air Conditioning system. ANPV-MPC combines key ideas in parameter-varying control, adaptive control, and online learning strategies to bridge the gap between computationally efficient linear model predictive control and more accurate nonlinear model predictive control. The novelty in ANPV-MPC is the use of a physics-inspired Bayesian neural network to estimate the coefficients of the parameter-varying linear control model. The Bayesian neural network additionally provides uncertainty estimates, triggering online training to capture evolving building system conditions. We show that ANPV-MPC can approximate the building system dynamics with a 28.39% higher accuracy than traditional linear model predictive control, resulting in 36.23% better control performance without increasing complexity of the optimal control problem. ANPV-MPC also adapts in real time to previously unseen conditions using online learning, further improving its performance.
24 POWER TRANSMISSION AND DISTRIBUTION↗
SU-E-T-479: Development and Validation of Analytical Models Predicting Secondary Neutron Radiation in Proton Therapy Applications
Purpose: Test and validation of analytical models predicting leakage neutron exposure in passively scattered proton therapy. Methods: Taking inspiration from the literature, this work attempts to build an analytical model predicting neutron ambient dose equivalents, H*(10), within the local 75 MeV ocular proton therapy facility. MC simulations were first used to model H*(10) in the beam axis plane while considering a closed final collimator and pristine Bragg peak delivery. Next, MC-based analytical model was tested against simulation results and experimental measurements. The model was also expended in the vertical direction to enable a full 3D mapping of H*(10) inside the treatment room. Finally, the work focused on upgrading the literature model to clinically relevant configurations considering modulated beams, open collimators, patient-induced neutron fluctuations, etc. Results: The MC-based analytical model efficiently reproduced simulated H*(10) values with a maximum difference below 10%. In addition, it succeeded in predicting measured H*(10) values with differences <40%. The highest differences were registered at the closest and farthest positions from isocenter where the analytical model failed to faithfully reproduce the high neutron fluence and energy variations. The differences remains however acceptable taking into account the high measurement/simulation uncertainties and the end use of this model, i.e. radiation protection. Moreover, the model was successfully (differences < 20% on simulations and < 45% on measurements) extended to predict neutrons in the vertical direction with respect to the beam line as patients are in the upright seated position during ocular treatments. Accounting for the impact of beam modulation, collimation and the present of a patient in the beam path is far more challenging and conversion coefficients are currently being defined to predict stray neutrons in clinically representative treatment configurations. Conclusion: Analytical models predicting secondary neutrons in proton therapy represent a promising solution that substitute for time-consuming MC calculations.
60 APPLIED LIFE SCIENCES↗
TH-A-9A-01: Active Optical Flow Model: Predicting Voxel-Level Dose Prediction in Spine SBRT
Purpose: To predict voxel-level dose distribution and enable effective evaluation of cord dose sparing in spine SBRT. Methods: We present an active optical flow model (AOFM) to statistically describe cord dose variations and train a predictive model to represent correlations between AOFM and PTV contours. Thirty clinically accepted spine SBRT plans are evenly divided into training and testing datasets. The development of predictive model consists of 1) collecting a sequence of dose maps including PTV and OAR (spinal cord) as well as a set of associated PTV contours adjacent to OAR from the training dataset, 2) classifying data into five groups based on PTV's locations relative to OAR, two “Top”s, “Left”, “Right”, and “Bottom”, 3) randomly selecting a dose map as the reference in each group and applying rigid registration and optical flow deformation to match all other maps to the reference, 4) building AOFM by importing optical flow vectors and dose values into the principal component analysis (PCA), 5) applying another PCA to features of PTV and OAR contours to generate an active shape model (ASM), and 6) computing a linear regression model of correlations between AOFM and ASM.When predicting dose distribution of a new case in the testing dataset, the PTV is first assigned to a group based on its contour characteristics. Contour features are then transformed into ASM's principal coordinates of the selected group. Finally, voxel-level dose distribution is determined by mapping from the ASM space to the AOFM space using the predictive model. Results: The DVHs predicted by the AOFM-based model and those in clinical plans are comparable in training and testing datasets. At 2% volume the dose difference between predicted and clinical plans is 4.2±4.4% and 3.3±3.5% in the training and testing datasets, respectively. Conclusion: The AOFM is effective in predicting voxel-level dose distribution for spine SBRT. Partially supported by NIH/NCI under grant #R21CA161389 and a master research grant by Varian Medical System.
62 RADIOLOGY AND NUCLEAR MEDICINE↗