Model Inversion Approach Research Articles

Neural activity shaping in visual prostheses with deep learning

Objective. The visual perception provided by retinal prostheses is limited by the overlapping current spread of adjacent electrodes. This reduces the spatial resolution attainable with unipolar stimulation. Conversely, simultaneous multipolar stimulation guided by the measured neural responses—neural activity shaping (NAS)—can attenuate excessive spread of excitation allowing for more precise control over the pattern of neural activation. However, defining effective multipolar stimulus patterns is a challenging task. Previous attempts focused on analytical solutions based on an assumed linear nonlinear model of retinal response; an analytical model inversion (AMI) approach. Here, we propose a model-free solution for NAS, using artificial neural networks (ANNs) that could be trained with data acquired from the implant. Approach. Our method consists of two ANNs trained sequentially. The measurement predictor network (MPN) is trained on data from the implant and is used to predict how the retina responds to multipolar stimulation. The stimulus generator network is trained on a large dataset of natural images and uses the trained MPN to determine efficient multipolar stimulus patterns by learning its inverse model. We validate our method in silico using a realistic model of retinal response to multipolar stimulation. Main results. We show that our ANN-based NAS approach produces sharper retinal activations than the conventional unipolar stimulation strategy. As a theoretical bench-mark of optimal NAS results, we implemented AMI stimulation by inverting the model used to simulate the retina. Our ANN strategy produced equivalent results to AMI, while not being restricted to any specific type of retina model and being three orders of magnitude more computationally efficient. Significance. Our novel protocol provides a method for efficient and personalized retinal stimulation, which may improve the visual experience and quality of life of retinal prosthesis users.

Intra‐Specific Variability in Plant Hydraulic Parameters Inferred From Model Inversion of Sap Flux Data

AbstractUnderstanding plant hydraulic regulation is critical for predicting plant and ecosystem responses to projected increases in drought stress. Plant hydraulic regulation is controlled by observable, diverse plant hydraulic traits that can vary as much across individuals of the same species as they do across different species. Direct measurements of plant hydraulic traits from a range of ecosystems remain limited in comparison to other, more readily measured traits (e.g., specific leaf area). Furthermore, plant hydraulic trait measurements, often made at leaf or branch levels, are not easily scaled to whole‐plant values that are typically used to predict plant and ecosystem fluxes. In this study, multiple whole‐plant hydraulic parameters are inferred from observations of plant water use (i.e., sap flow), soil properties, and meteorological data. We use a Markov Chain Monte Carlo model inversion approach to obtain the best estimates and uncertainty of plant hydraulic parameters that capture whole‐plant effective embolism resistance and stomatal sensitivity to decreasing plant water potential. We then use the inferred values in the model to estimate whole‐tree water use and isohydricity. This approach reliably infers whole‐plant parameter values with enough specificity to resolve inter‐ and intra‐specific differences, and thus supplements time‐ and labor‐intensive direct measurements of traits.

Satellite-derived bathymetry in optically complex waters using a model inversion approach and Sentinel-2 data

This study presents an assessment of a model inversion approach to derive shallow water bathymetry in optically complex waters, with the aim of both understanding localised capability and contributing to the global evaluation of Sentinel-2 for coastal monitoring. A dataset of 12 Sentinel-2 MSI images, in three different study areas along the Irish coast, has been analysed. Before the application of the bathymetric model two atmospheric correction procedures were tested: Deep Water Correction (DWC) and Case 2 Regional Coastal Color (C2RCC) processor. DWC outperformed C2RCC in the majority of the satellite images showing more consistent results. Using DWC for atmospheric correction before the application of the bathymetric model, the lowest average RMSE was found in Dublin Bay (RMSE = 1.60, bias = −0.51), followed by Mulroy Bay (RMSE = 1.66, bias = 1.30) while Brandon Bay showed the highest average error (RMSE = 2.43, bias = 1.86). However, when the optimal imagery selection was considered, depth estimations with a bias less than 0.1 m and a spread of ±1.40 m were achieved up to 10 m. These results were comparable to those achieved by empirical tuning methods, despite not relying on any in situ depth data. This conclusion is of particular relevance as model inversion approaches might allow future modifications in crucial parts of the processing chain leading to improved results. Atmospheric correction, the selection of optimal images (e.g. low turbidity), the definition of suitably limited ranges for the per-pixel occurrence of optical constituents (phytoplankton, CDOM, backscatter) and seabed reflectances, in combination with the understanding of the specifics characteristics at each particular site, were critical steps in the derivation of satellite bathymetry.

Observability of the ambient conditions in model‐based estimation for wind farm control: A focus on static models

AbstractWind farm control (WFC) algorithms rely on an estimate of the ambient wind speed, wind direction, and turbulence intensity in the determination of the optimal control setpoints. However, the measurements available in a commercial wind farm do not always carry sufficient information to estimate these atmospheric quantities. In this paper, a novel measure (“observability”) is introduced that quantifies how well the ambient conditions can be estimated with the measurements at hand through a model inversion approach. The usefulness of this measure is shown through several case studies. While the turbine power signals and the inter‐turbine wake interactions provide information on the wind direction, the case studies presented in this article show that there is a strong need for wind direction measurements for WFC to sufficiently cover observability for any ambient condition. Further, generally, more wake interaction leads to a higher observability. Also, the mathematical framework presented in this article supports the straightforward notion that turbine power measurements provide no additional information compared with local wind speed measurements, implying that power measurements are superfluous. Irregular farm layouts result in a higher observability due to the increase in unique wake interaction. The findings in this paper may be used in WFC to predict which ambient quantities can (theoretically) be estimated. The authors envision that this will assist in the estimation of the ambient conditions in WFC algorithms and can lead to an improvement in the performance of WFC algorithms over the complete envelope of wind farm operation.

Control Allocation for Hybrid Braking Considering Dynamic Battery Behaviour

This work deals with the problem of maximizing energy recovery while optimizing braking capability of high-performance electric vehicles under given actuator constraints. A scalable, real-time capable concept using a complementary filter with an additional daisy chain for the control allocation between hydraulic brakes and electric motors is proposed. Furthermore, a model inversion approach that derives necessary battery limits while trying to decouple the wheel speed control loop from the protection loop of a modern battery management system is described. The hybrid braking control architecture is studied in simulation and validated in a prototype electric vehicle by highly dynamical driving manoeuvres.

Modeling the Vertical Backscattering Distribution in the Percolation Zone of the Greenland Ice Sheet With SAR Tomography

The penetration of microwave signals into snow and ice, especially in dry conditions, introduces a bias in digital elevation models generated by means of synthetic aperture radar (SAR) interferometry. This bias depends directly on the vertical backscattering distribution in the subsurface. At the same time, the sensitivity of interferometric SAR measurements on the vertical backscattering distribution provides the potential to derive information about the subsurface of glaciers and ice sheets from SAR data, which could support the assessment of their dynamics. The aim of this article is to improve the interferometric modeling of the vertical backscattering distribution in order to support subsurface structure retrieval and penetration bias estimation. Vertical backscattering distributions are investigated at different frequencies and polarizations on two test sites in the percolation zone of Greenland using fully polarimetric X-, C-, L-, and P-band SAR data. The vertical backscattering distributions were reconstructed by means of SAR tomography and compared to different vertical structure models. The tomographic assessment indicated that the subsurface in the upper percolation zone is dominated by scattering layers at specific depths, whereas a more homogeneous scattering structure appears in the lower percolation zone. The performance of the evaluated structure models, namely an exponential function with a vertical shift, a Gaussian function, and a Weibull function, was evaluated. The proposed models improve the representation of the data compared with existing models while the complexity is still low to enable potential model inversion approaches. The tomographic analysis and the model assessment is therefore a step forward toward subsurface structure information and penetration bias estimation from SAR data.

A comparative study on existing and new methods to design internal model controllers for non-square systems

This paper focuses on a challenging problem in the internal model control (IMC) strategy: the model inversion to design the IMC controller for non-square systems. Several existing approaches for the synthesis of a specific inversion of the identified model will be presented in this paper to deal with the differences between the system’s inputs and outputs numbers. The non-square effective relative gain is firstly presented. It consists of the measurement of interactions between the loops of the system in order to square the system and make it invertible. The equivalent transfer function method is presented as well. It is based on tuning the pseudo-inverse of the process to design the internal model controller. These methods are then compared with a novel proposed model inversion approach based on virtual outputs method. Virtual adding is considered in order to obtain an invertible square transfer matrix to design the internal model controller. This simple yet effective method ensures robust control performance. Its efficiency and availability, as compared with other presented methods, is illustrated through simulations on an overactuated system with three inputs/two outputs.

Partial volume mapping using magnetic resonance fingerprinting.

Magnetic resonance fingerprinting (MRF) is a quantitative imaging technique that maps multiple tissue properties through pseudorandom signal excitation and dictionary-based reconstruction. The aim of this study is to estimate and validate partial volumes from MRF signal evolutions (PV-MRF), and to characterize possible sources of error. Partial volume model inversion (pseudoinverse) and dictionary-matching approaches to calculate brain tissue fractions (cerebrospinal fluid, gray matter, white matter) were compared in a numerical phantom and seven healthy subjects scanned at 3T. Results were validated by comparison with ground truth in simulations and ROI analysis in vivo. Simulations investigated tissue fraction errors arising from noise, undersampling artifacts, and model errors. An expanded partial volume model was investigated in a brain tumor patient. PV-MRF with dictionary matching is robust to noise, and estimated tissue fractions are sensitive to model errors. A 6% error in pure tissue T1 resulted in average absolute tissue fraction error of 4% or less. A partial volume model within these accuracy limits could be semi-automatically constructed in vivo using k-means clustering of MRF-mapped relaxation times. Dictionary-based PV-MRF robustly identifies pure white matter, gray matter and cerebrospinal fluid, and partial volumes in subcortical structures. PV-MRF could also estimate partial volumes of solid tumor and peritumoral edema. We conclude that PV-MRF can attribute subtle changes in relaxation times to altered tissue composition, allowing for quantification of specific tissues which occupy a fraction of a voxel.

Control of MIMO nonlinear systems: A data-driven model inversion approach

A data-driven control design approach for Multiple Input Multiple Output nonlinear systems is presented in this paper. The approach, called Nonlinear Inversion Control (NIC), is based on the identification of a polynomial prediction model of the system to control and the on-line inversion of this model. The main features of the NIC approach can be summarized as follows: it does not require a physical model of the plant to control which, in many real-world situations, may be difficult to derive; it can guarantee a priori properties such as closed-loop stability and tracking error accuracy; it is general, numerically efficient and relatively simple. Extensive simulations are carried out to test the numerical efficiency of the NIC approach. A simulated example of industrial interest is also presented, concerned with control of a robotic manipulator.

Application of Sentinel-2A data for pasture biomass monitoring using a physically based radiative transfer model

A large proportion of the global land surface is covered by pasture. The advent of the Sentinel satellites program provides free datasets with good spatiotemporal resolution that can be a valuable source of information for monitoring pasture resources. We combined optical remote sensing data (proximal hyperspectral and Sentinel 2A) with a radiative transfer model (PROSAIL) to estimate leaf Area Index (LAI), and biomass, in a dairy farming context. Three sites in Southern England were used: two pasture farms that differed in pasture type and management, and a set of small agronomy trial plots with different mixtures of grasses, legumes and herbs, as well as pure perennial ryegrass. The proximal and satellite spectral data were used to retrieve LAI via PROSAIL model inversion, which were compared against field observations of LAI. The potential of bands of Sentinel 2A that corresponded with a 10 m resolution was studied by convolving narrow spectral bands (from a handheld hyperspectral sensor) into Sentinel 2A bands (10 m). Retrieved LAI, using these spectrally resampled S2A data, compared well with measured LAI, for all sites, even for those with mixed species cover (although retrieved LAI was somewhat overestimated for pasture mixtures with high LAI). This proved the suitability of 10 m Sentinel 2A spectral bands for capturing LAI dynamics for different types of pastures. We also found that inclusion of 20 m bands in the inversion scheme did not lead to any further improvement in retrieved LAI. Sentinel 2A image based retrieval yielded good agreement with LAI measurements obtained for a typical perennial ryegrass based pasture farm. LAI retrieved in this way was used to create biomass maps (that correspond to indirect biomass measurements by Rising Plate Meter (RPM)), for mixed-species paddocks for a farm for which limited field data were available. These maps compared moderately well with farmer-collected RPM measurements for this farm. We propose that estimates of paddock-averaged and within-paddock variability of biomass are more reliably obtained from a combined Sentinel 2A-PROSAIL approach, rather than by manual RPM measurements. The physically based radiative transfer model inversion approach outperformed the Normalized Difference Vegetation Index based retrieval method, and does not require site specific calibrations of the inversion scheme.

Remote Sensing of Leaf and Canopy Nitrogen Status in Winter Wheat (Triticum aestivum L.) Based on N-PROSAIL Model

Plant nitrogen (N) information has widely been estimated through empirical techniques using hyperspectral data. However, the physical model inversion approach on N spectral response has seldom developed and remains a challenge. In this study, an N-PROSAIL model based on the N-based PROSPECT model and the SAIL model canopy model was constructed and used for retrieving crop N status both at leaf and canopy scales. The results show that the third parameter (3rd-par) retrieving strategy (leaf area index (LAI) and leaf N density (LND) optimized where other parameters in the N-PROSAIL model are set at different values at each growth stage) exhibited the highest accuracy for LAI and LND estimation, which resulted in R2 and RMSE values of 0.80 and 0.69, and 0.46 and 21.18 µg·cm−2, respectively. It also showed good results with R2 and RMSE values of 0.75 and 0.38% for leaf N concentration (LNC) and 0.82 and 0.95 g·m−2 for canopy N density (CND), respectively. The N-PROSAIL model retrieving method performed better than the vegetation index regression model (LNC: RMSE = 0.48 − 0.64%; CND: RMSE = 1.26 − 1.78 g·m−2). This study indicates the potential of using the N-PROSAIL model for crop N diagnosis on leaf and canopy scales in wheat.

Deterministic and statistical parameter characterization in resonant fluid-structure interaction problems

This research focuses on developing computational methods to estimate model parameters in resonant fluid-structure interaction problems over a wide frequency range by means of model inversion approaches. The considered problems are widely known to be subjected to local minima, which represent a major challenge in the field of parameter identification. In the proposed method, the frequency spectrum is divided into successive substeps allowing to efficiently guide the estimation towards the global minimum, i.e., the true model parameters. The estimation is performed through two frameworks, namely, the deterministic using gradient-based optimization and Bayesian using Markov chain Monte Carlo method. Proposed numerical examples illustrate the effectiveness and potential of the proposed stepwise scheme to find the global minimum and reduce the overall computational burden.

Calcium imaging and dynamic causal modelling reveal brain-wide changes in effective connectivity and synaptic dynamics during epileptic seizures.

Pathophysiological explanations of epilepsy typically focus on either the micro/mesoscale (e.g. excitation-inhibition imbalance), or on the macroscale (e.g. network architecture). Linking abnormalities across spatial scales remains difficult, partly because of technical limitations in measuring neuronal signatures concurrently at the scales involved. Here we use light sheet imaging of the larval zebrafish brain during acute epileptic seizure induced with pentylenetetrazole. Spectral changes of spontaneous neuronal activity during the seizure are then modelled using neural mass models, allowing Bayesian inference on changes in effective network connectivity and their underlying synaptic dynamics. This dynamic causal modelling of seizures in the zebrafish brain reveals concurrent changes in synaptic coupling at macro- and mesoscale. Fluctuations of both synaptic connection strength and their temporal dynamics are required to explain observed seizure patterns. These findings highlight distinct changes in local (intrinsic) and long-range (extrinsic) synaptic transmission dynamics as a possible seizure pathomechanism and illustrate how our Bayesian model inversion approach can be used to link existing neural mass models of seizure activity and novel experimental methods.

Estimation of leaf traits from reflectance measurements: comparison between methods based on vegetation indices and several versions of the PROSPECT model

BackgroundLeaf biochemical composition corresponds to traits related to the plant state and its functioning. This study puts the emphasis on the main leaf absorbers: chlorophyll a and b (C_{ab}), carotenoids (C_{c}), water (C_{w}) and dry mater (C_{m}) contents. Two main approaches were used to estimate [C_{ab} ,C_{c}, C_{w}, C_{m}] in a non-destructive way using spectral measurements. The first one consists in building empirical relationships from experimental datasets using either the raw reflectances or their combination into vegetation indices (VI). The second one relies on the inversion of physically based models of leaf optical properties. Although the first approach is commonly used, the calibration of the empirical relationships is generally conducted over a limited dataset. Consequently, poor predictions may be observed when applying them on cases that are not represented in the training dataset, i.e. when dealing with different species, genotypes or under contrasted environmental conditions. The retrieval performances of the selected VIs were thus compared to the ones of four PROSPECT model versions based on reflectance data acquired at two phenological stages, over six wheat genotypes grown under three different nitrogen fertilizations and two sowing density modalities. Leaf reflectance was measured in the lab with a spectrophotometer equipped with an integrating sphere, the leaf being placed in front of a white Teflon background to increase the sensitivity to leaf biochemical composition. Destructive measurements of [C_{ab} ,C_{c}, C_{w}, C_{m}] were performed concurrently.ResultsThe destructive measurements demonstrated that the carotenoid, C_{c}, and chlorophyll, C_{ab}, contents were strongly correlated (r2 = 0.91). The sum of C_{ab} and C_{c}, i.e. the total chlorophyllian pigment content, C_{abc}, was therefore used in this study. When inverting the PROSPECT model, accounting for the brown pigment content, C_{bp}, was necessary when leaves started to senesce. The values of C_{abc} and C_{w} were well estimated (r2 = 0.81 and r2 = 0.88 respectively) while the dry matter content, C_{m}, was poorly estimated (r2 = 0.00). Retrieval of C_{w} from PROSPECT versions was only slightly biased, while substantial overestimation of C_{abc } was observed. The ranking between estimated values of C_{abc} and C_{w} from the several PROSPECT versions and that derived using the VIs were similar to the ranking observed over the destructively measured values of C_{abc} and C_{w}.ConclusionsPROSPECT model inversion and empirical VI approach provide similar retrieval performances and are useful methods to estimate leaf biochemical composition from spectral measurements. However, the PROSPECT model inversion gives potential access to additional traits on surface reflectivity and leaf internal structure. This study suggests that non-destructive estimation of leaf chlorophyll and water contents is a relevant method to provide leaf traits with relatively high throughput.

Retrieval of Specific Leaf Area From Landsat-8 Surface Reflectance Data Using Statistical and Physical Models

One of the key traits in the assessment of ecosystem functions is a specific leaf area (SLA). The main aim of this study was to examine the potential of new generation satellite images, such as Landsat-8 imagery, for the retrieval of SLA at regional and global scales. Therefore, both statistical and radiative transfer model (RTM) inversion approaches for estimating SLA from the new Landsat-8 product were evaluated. Field data were collected for 33 sample plots during a field campaign in summer 2013 in the Bavarian Forest National Park, Germany, while Landsat-8 image data concurrent with the time of field campaign were acquired. Estimates of SLA were examined using different Landsat-8 spectral bands, vegetation indices calculated from these bands, and the inversion of a canopy RTM. The RTM inversion was performed utilizing continuous wavelet analysis and a look-up table (LUT) approach. The results were validated using R 2 and the root-mean-square error (RMSE) between the estimated and measured SLA. In general, SLA was estimated accurately by both statistical and RTM inversion approaches. The relationships between measured and estimated SLA using the enhanced vegetation index were strong ( R 2 = 0.77 and RMSE = 4.44%). Furthermore, the predictive model developed from combination of the wavelet features at 654.5 nm (scale 9) and 2200.5 nm (scale 2) correlated strongly with SLA ( R 2 = 0.79 and RMSE = 7.52%). The inversion of LUT using a spectral subset consisting of bands 5, 6, and 7 of Landsat-8 ( R 2 = 0.73 and RMSE = 5.33%) yielded a higher accuracy and precision than any other spectral subset. The findings of this study provide insights into the potential of the new generation of multispectral medium-resolution satellite imagery, such as Landsat-8 and Sentinel-2, for accurate retrieval and mapping of SLA using either statistical or RTM inversion methods.

Soil physical property estimation from soil strength and apparent electrical conductivity sensor data

Proximal soil sensing is an attractive approach for quantifying soil properties, but many currently available sensors do not respond to a single soil property. For example, soil strength and apparent electrical conductivity (ECa) sensor measurements are significantly affected by soil texture, bulk density (BD), and water content (WC). The objective of this study was to explore the potential for estimating soil texture, BD, and WC using combinations of sensor-based soil strength and ECa data obtained from sites with varying soil physical properties. Data collected from three research sites in Missouri included on-the-go horizontal soil strength at five depths up to 0.5 m on a 0.1-m interval, cone index measurements at the same depths, ECa measured by a Veris 3100, and depth-dependent, laboratory-determined soil properties. An ECa model inversion approach was used to generate layer EC values corresponding to the depth increments of the other variables. Fits of models using EC to estimate WC were variable (R2 = 0.31–0.79). Best fitting BD estimation models (R2 = 0.11–0.55) generally included EC, but soil strength was included in fewer than half of the models. BD model fits were improved considerably by adding lab-measured WC to the model (R2 = 0.30–0.86), suggesting the need for a WC sensor. Soil clay texture fraction models based on EC and WC fit well (R2 = 0.80–0.93). This study showed the potential of combining data from multiple mobile proximal sensors to estimate important soil physical properties.

A physics-based method for the remote sensing of seagrasses

Abstract Seagrass meadows are important environments for the blue carbon budget and are potential early indicators for environmental change. Remote sensing is a viable monitoring tool for spatially extensive meadows but most current approaches are limited by the requirement for in situ calibration data or provide categorical level maps rather than quantitative estimates of direct physiological significance. In this paper we present a method for mapping water depth and the leaf area index (LAI, ratio of leaf area to substrate area) of Thalassia testudinum meadows, based on radiative transfer model inversion using an embedded three-dimensional aquatic canopy model. Variations in reflectance due to leaf length, leaf position, sediment coverage on leaves, water depth and solar zenith angle were included in the model to parameterise uncertainty propagation. The model revealed canopy reflectance as a function of LAI decreases exponentially at all wavelengths up to an LAI around four, beyond which increasing canopy density cannot be determined from reflectance. In addition, sediment coverage on leaves has surprisingly little effect on the reflectance of sparse canopies because shading is also a contributor to darkening. The capability of the method for image based mapping was assessed through sensitivity analyses and by application to hyperspectral airborne imagery of Florida Bay collected by the Portable Remote Imaging Spectrometer (PRISM), with the uncertainty propagation providing per-pixel confidence intervals on all the estimated parameters. Results were consistent across the sensitivity and image analyses and the agreement with field data was good, given the challenges in validation of submerged pixels at metre scale. Uncertainties were high for LAIs greater than two in water of depth 8 m, but lower for sparse canopies and shallower water. For water depths approaching 10 m the pixel-to-pixel variation arising from processes at the water surface upwards was greater than the uncertainties arising from the canopy or water column. The physics-based model inversion approach is readily adaptable to any sensor configuration and to different seagrass species and canopy morphologies. No site-specific in situ data is required and the uncertainty estimates can provide an objective basis for interpreting apparent changes in the distribution of seagrasses over time and space, as revealed by remote sensing techniques.

Ice-sheet configuration in the CMIP5/PMIP3 Last Glacial Maximum experiments

Abstract. We describe the creation of a data set describing changes related to the presence of ice sheets, including ice-sheet extent and height, ice-shelf extent, and the distribution and elevation of ice-free land at the Last Glacial Maximum (LGM), which were used in LGM experiments conducted as part of the fifth phase of the Coupled Modelling Intercomparison Project (CMIP5) and the third phase of the Palaeoclimate Modelling Intercomparison Project (PMIP3). The CMIP5/PMIP3 data sets were created from reconstructions made by three different groups, which were all obtained using a model-inversion approach but differ in the assumptions used in the modelling and in the type of data used as constraints. The ice-sheet extent in the Northern Hemisphere (NH) does not vary substantially between the three individual data sources. The difference in the topography of the NH ice sheets is also moderate, and smaller than the differences between these reconstructions (and the resultant composite reconstruction) and ice-sheet reconstructions used in previous generations of PMIP. Only two of the individual reconstructions provide information for Antarctica. The discrepancy between these two reconstructions is larger than the difference for the NH ice sheets, although still less than the difference between the composite reconstruction and previous PMIP ice-sheet reconstructions. Although largely confined to the ice-covered regions, differences between the climate response to the individual LGM reconstructions extend over the North Atlantic Ocean and Northern Hemisphere continents, partly through atmospheric stationary waves. Differences between the climate response to the CMIP5/PMIP3 composite and any individual ice-sheet reconstruction are smaller than those between the CMIP5/PMIP3 composite and the ice sheet used in the last phase of PMIP (PMIP2).

승용디젤엔진의 EGR, VGT 시스템을 위한 비선형 정적 모델 기반 피드포워드 제어 알고리즘 설계

This paper presents a feedforward control algorithm for the EGR and VGT systems of passenger car diesel engines. The air-to-fuel ratio and boost pressure are selected as control indicators and the positions of EGR valve and VGT vane are used as control inputs of the EGR and VGT controller. In order to compensate the non-linearity and coupled dynamics of the EGR and VGT systems, we have proposed a non-linear model-based feedforward control algorithm which is obtained from static model inversion approach. It is observed that the average modeling errors of the feedforward algorithm is about 2% using stationary engine experiment data of 225 operating conditions. Using a feedback controller including proportional-integral, the modeling error is compensated. Furthermore, it is validated that the proposed feedforward algorithm generates physically acceptable trajectories of the actuator and successfully tracks the desired values through engine experiments.

Transfer of a nanoparticle product between different mixers using latent variable model inversion

An experimental nanoparticle preparation process by solvent displacement in passive mixers is considered. The problem under investigation is to estimate the operating conditions in a target device (Mixer B) in order to obtain a product of assigned properties that has already been manufactured in a source device of different geometry (Mixer A). A large historical database is available for Mixer A, whereas a limited historical database is available for Mixer B. The difference in device geometries causes a different mixing performance within the devices, which is very difficult to capture using mechanistic models. The problem is further complicated by the fact that Mixer B can only be run under an experimental setup that is different from the one under which the available historical dataset was obtained. A joint‐Y projection to latent structures (JY‐PLS) model inversion approach is used to transfer the nanoparticle product from Mixer A to Mixer B. The Mixer B operating conditions estimated by the model are tested experimentally and confirm the model predictions within the experimental uncertainty. Since the inversion of the JY‐PLS model generates an infinite number of solutions that all lie in the so‐called null space, experiments are carried out to provide (to the authors' knowledge) the first experimental validation of the theoretical concept of null space. Finally, by interpreting the JY‐PLS model parameters from first principles, the understanding of the system physics is improved. © 2013 American Institute of Chemical Engineers AIChE J, 60: 123–135, 2014