Multi-domain Model Research Articles

Multi-Domain Aspect Extraction Using Bidirectional Encoder Representations From Transformers

Deep learning and neural language models have obtained state-of-the-art results in aspects extraction tasks, in which the objective is to automatically extract characteristics of products and services that are the target of consumer opinion. However, these methods require a large amount of labeled data to achieve such results. Since data labeling is a costly task, there are no labeled data available for all domains. In this paper, we propose an approach for aspect extraction in a multi-domain transfer learning scenario, thereby leveraging labeled data from different source domains to extract aspects of a new unlabeled target domain. Our approach, called MDAE-BERT (Multi-Domain Aspect Extraction using Bidirectional Encoder Representations from Transformers), explores neural language models to deal with two major challenges in multi-domain learning: (1) inconsistency of aspects from target and source domains and (2) context-based semantic distance between ambiguous aspects. We evaluated our MDAE-BERT considering two perspectives (1) the aspect extraction performance using F1-Macro and Accuracy measures; and (2) by comparing the multi-domain aspect extraction models and single-domain models for aspect extraction. In the first perspective, our method outperforms the LSTM-based approach. In the second perspective, our approach proved to be a competitive alternative compared to the single-domain model trained in a specific domain, even in the absence of labeled data from the target domain.

A Ferroelectric-based Volatile/Non-volatile Dual-mode Buffer Memory for Deep Neural Network Accelerators

In this paper, a dual-mode buffer memory based on the CMOS compatible HfZrO2 ferroelectric material is proposed for DNN accelerators. It can operate in both volatile eDRAM mode and non-volatile ferroelectric RAM (FeRAM) mode. The functionality of the proposed dual-mode memory bit-cell design is verified using SPICE simulation with the multi-domain Preisach physical model. A data-lifetime-aware memory mode configuration protocol is proposed to optimize the buffer access energy for both DNN inference and training. Detailed circuitry and architectural support for the dual-mode memory are discussed. For DNN training with ferroelectric-field-effect-transistor (FeFET) and SRAM-based compute-in-memory (CIM) accelerator, the proposed dual-mode buffer design improves the overall energy efficiency by 92.2%~98.7%, 44.1%~47.6%, 12.6%~13.0% compared to baseline designs using SRAM buffer with the same buffer area, eDRAM and FeRAM with the same buffer capacity, respectively. For DNN inference with tensor-processing-unit (TPU)-like systolic array, the energy efficiency during computing is improved by 40.7%~45.6%, 18.4%~29.6% compared to the designs with eDRAM and FeRAM buffer, respectively. By storing the persistent data using the non-volatile mode, the energy efficiency of systolic array is improved by 2.3~5.5 over SRAM-based design when standby is frequent.

A multi-domain shear-stress dependent diffusive model of cell transport-aided dialysis: analysis and simulation.

Kidney dialysis is the most widespread treatment method for end-stage renal disease, a debilitating health condition common in industrialized societies. While ubiquitous, kidney dialysis suffers from an inability to remove larger toxins, resulting in a gradual buildup of these toxins in dialysis patients, ultimately leading to further health complications. To improve dialysis, hollow fibers incorporating a cell-monolayer with cultured kidney cells have been proposed; however, the design of such a fiber is nontrivial. In particular, the effects of fluid wall-shear stress have an important influence on the ability of the cell layer to transport toxins. In the present work, we introduce a model for cell-transport aided dialysis, incorporating the effects of the shear stress. We analyze the model mathematically and establish its well-posedness. We then present a series of numerical results, which suggest that a hollow-fiber design with a wavy profile may increase the efficiency of the dialysis treatment. We investigate numerically the shape of the wavy channel to maximize the toxin clearance. These results demonstrate the potential for the use of computational models in the study and advancement of renal therapies.

Data Farming in Production Systems - A Review on Potentials, Challenges and Exemplary Applications

The current trend in production optimization extensively relies on data analytics methods such as statistics and machine learning algorithms, aiming at the exploration of undiscovered relationships in existing production systems. When real processes are monitored and analyzed, only the actually implemented features and relationships can be observed. This limits the amount of detectable interdependencies, since prolonged execution of defective processes is often avoided due to negative impact on the running production system. To discover additional insight into production systems, their digital representations i.e. the multi-domain simulation models can be leveraged. Based on these models, Data Farming allows to gain even more insight on production systems. First, major simulation experiments are executed automatically to generate new data sets. Second, the resulting data is analyzed using Data Mining methods to gain additional insight into simulation models and as a result knowledge over the real production system itself. Thus, using Data Farming approaches avoids test scenarios on the production system, which would lead to negative effects on productivity. At the same time, the quality of the simulation model, especially in its border areas, has to be precise enough to make valid statements on newly discovered interdependencies. This paper presents the current state of Data Farming applications in the context of production systems via a review on existing theory, applications and methods. On this basis, the potentials of Data Farming in the context of production systems as well as current challenges in its implementation are pointed out. In conclusion, new use cases for further work are presented.

A Hybrid Ensemble Word Embedding based Classification Model for Multi-document Summarization Process on Large Multi-domain Document Sets

Contextual text feature extraction and classification play a vital role in the multi-document summarization process. Natural language processing (NLP) is one of the essential text mining tools which is used to preprocess and analyze the large document sets. Most of the conventional single document feature extraction measures are independent of contextual relationships among the different contextual feature sets for the document categorization process. Also, these conventional word embedding models such as TF-ID, ITF-ID and Glove are difficult to integrate into the multi-domain feature extraction and classification process due to a high misclassification rate and large candidate sets. To address these concerns, an advanced multi-document summarization framework was developed and tested on number of large training datasets. In this work, a hybrid multi-domain glove word embedding model, multi-document clustering and classification model were implemented to improve the multi-document summarization process for multi-domain document sets. Experimental results prove that the proposed multi-document summarization approach has improved efficiency in terms of accuracy, precision, recall, F-score and run time (ms) than the existing models.

Development of virtual lab module for wind and solar energy systems using 3DEXPERIENCE platform of Dassault Systemes

Solar and wind energy are the renewable energy sources that have the great potential to fulfill the electricity needs of the world. They are available in abundance and also while generating electricity, they produce little or no global warming emissions. A lot of work has been carried out in the literature in the field of solar and wind energy systems. This has been taught as courses in various universities. 3DEXPERIENCE platform of Dassault Systemes has an integrated mathematical Modelling application known as Dymola. This integrated Dymola software allows solving complex multi-domain system modeling and analysis by using modelica and simulation technology. In this work, a virtual lab module for wind and solar PV systems have been designed and developed in 3DEXPERIENCE platform. The characteristic Plots of solar PV system for various configurations are plotted and effect of partial shading is also analyzed. The variation of mechanical power of wind system for various pitch angles are plotted.

Knowledge-based Design Support for Model-based Development with Multi-domain Modeling Language

Knowledge-based Design Support for Model-based Development with Multi-domain Modeling Language

Modeling dynamic mechanical system behavior using sequence modeling of embodiment function relations: case study on a hammer mechanism

In engineering design, an issue for using complex simulation models in system analysis are unknown causes for dynamic system behavior, which make parameterization difficult. This paper presents a case study in which a structured system analysis is used for the parameterization of complex dynamic multi-domain models. The dynamic system behavior of an impact wrench during fastening of a bolt is analyzed and modeled using the Contact and Channel Approach for structured parameterization of a multibody simulation model. This qualitative model building serves as a basis for a simulation model that quantifies the relations of design parameters and system behavior. Comparison with experimental test results is done as a validation. With this approach, the behavior identified in the simulation model could be traced back in a structured way to its cause in the system embodiment. The simulation model represented the real dynamic system behavior with an initial sufficient precision, but showed a lack of precision in detail. On this basis, the Contact and Channel Model was extended by adding additional statistical behavior of the system. Parameters of the system embodiment were identified qualitatively to improve the simulation model. A limitation in qualitative modeling of dynamic changes in the system has been identified that needs to be addressed in further research.

Dynamics Simulation Analysis and Optimal Design of the Boom System of a Concrete Pump Truck

In order to improve the smoothness of the movement process of the arm system, this paper adopts Modelica multi-domain modeling simulation technology based on MWorks platform to analyze and improve the boom system of a concrete pump. Firstly, the mechanical rigid-flexible coupling model and hydraulic model of the arm system of a concrete pump truck are built; in order to study the influencing factors of the arm movement smoothness, stiffness and damping of hydraulic system under the joint arm linkage condition in the arm system are simulated and analyzed; finally, the corresponding optimization and improvement scheme is put forward according to the simulation results. The experimental results show that compared with the original system, the proposed improvement scheme significantly improves the smoothness, which provides theoretical basis and technical support for product improvement.

Architecture Modeling and Test of Tractor Power Shift Transmission

Architecture modeling, simulation, and test verification are performed to meet the requirements of multi-domain modeling and simulation of tractor power shift transmission (PST). The principles of architecture modeling and high level architecture (HLA) are analyzed, and the PST architecture modeling connotation and simulation system structure are studied. Then the dynamic principles of PST mechanical, hydraulic and control subsystems are analyzed in the preparation phase, sliding friction phase and holding phase of clutch engagement. To build the PST system simulation model, the simulation models and components of the three subsystems are established, and the object and interaction classes of the simulation object model (SOM) for each component are defined. After that, the parameter mapping relationship between the components is analyzed, the interfaces between the simulation components and HLA are encapsulated, and the simulation component running sequence is analyzed. Finally, the PST clutch engagement law under tractor shift conditions is simulated and tested. The relative error and correlation coefficient between the simulated and test data of shift solenoid valve drive current are used as indexes to evaluate the PST system simulation model. The results show that the simulated drive current curves are consistent with the test ones. The maximum error between the simulated and test shift time appears at the third gear downshift point, with a value of 7.89%, and the minimum correlation coefficient value between the simulated and the test drive current at all shift points is 0.92, which indicate that the PST system simulation model is effective and accurate.

Distributed Controllers for Human-Robot Locomotion: A Scalable Approach Based on Decomposition and Hybrid Zero Dynamics

This letter presents a formal foundation, based on decomposition, hybrid zero dynamics (HZD), and a scalable optimization, to develop distributed control algorithms for hybrid models of collaborative human-robot locomotion. The proposed approach considers a centralized controller and then decomposes the dynamics and feedback laws with a parameterization to synthesize local controllers. The Jacobian matrix of the Poincaré map with local controllers is studied and compared to that with centralized ones. An optimization problem is then set up to tune the parameters of the local controllers for asymptotic stability. The proposed approach can significantly reduce the number of controller parameters to be optimized for the synthesis of distributed controllers. The analytical results are numerically evaluated with simulations of a multi-domain hybrid model with 19 degrees of freedom for stable amputee locomotion with a powered knee-ankle prosthetic leg.

CrossWOZ: A Large-Scale Chinese Cross-Domain Task-Oriented Dialogue Dataset

To advance multi-domain (cross-domain) dialogue modeling as well as alleviate the shortage of Chinese task-oriented datasets, we propose CrossWOZ, the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. Moreover, the corpus contains rich annotation of dialogue states and dialogue acts on both user and system sides. About 60% of the dialogues have cross-domain user goals that favor inter-domain dependency and encourage natural transition across domains in conversation. We also provide a user simulator and several benchmark models for pipelined task-oriented dialogue systems, which will facilitate researchers to compare and evaluate their models on this corpus. The large size and rich annotation of CrossWOZ make it suitable to investigate a variety of tasks in cross-domain dialogue modeling, such as dialogue state tracking, policy learning, user simulation, etc.

Abstract 12568: Impact of Aortic Stenosis on Myofiber Stress: Translational Application of Left Ventricle-Aortic Coupling Simulations

Introduction: Grading aortic stenosis (AS) has traditionally relied on measuring hemodynamic parameters of transvalvular pressure gradient, ejection jet velocity, or estimating valve orifice area. Recent research has highlighted limitations of these criteria at effectively grading AS in presence of left ventricle (LV) dysfunction. We hypothesize that simulations coupling the aorta and LV will provide meaningful insight into myocardial biomechanical derangements that accompany AS. Reference data from the normal ventricle should first be obtained. Methods: A multi-domain cardiac model with representative anatomy and material properties was used to create AS simulations. Finite element analysis was performed with ABAQUS FEA®. An anisotropic hyperelastic model was assigned to the aorta and LV passive properties, while time-varying elastance function governed LV active response. Mild and severe AS were created by restricting the aortic valve orifice area. Results: Global LV myofiber end systolic (ES) stress (mean±SD) was 9.31±10.33 kPa at baseline (no AS), 13.13±10.29 kPa for mild AS, and 16.18±10.59 kPa with severe AS. Mean LV myofiber ES strains were -22.4±8.7%, -22.2±8.9%, and -21.9±9.2%, respectively. Mild and severe AS had significant stress elevation compared to baseline (mild AS vs base; p<0.01, severe AS vs base; p<0.001) and when compared to each other (p<0.01). See Figure 1 . Ventricular regions that experienced greatest magnitude ES stress were (severe AS vs baseline) basal inferior (39.87±14.73 vs 30.02±12.08 kPa; p<0.01), mid-anteroseptal (32.29±11.56 vs 24.79±12.00 kPa; p<0.001), and apex (27.99±8.44 vs 23.52±10.19 kPa; p<0.001). Conclusions: Isolated AS in a normal heart was simulated, and significantly elevated LV myofiber stress was quantified. This data serves as a comparison to future studies that will incorporate patient-specific ventricular geometries and material parameters, aiming to correlate LV biomechanics to AS severity.

Model-Based Deterioration Estimation with Cyber Physical System

A key aspect of life cycle management for pursuing sustainability is machine condition prognosis, which requires a condition monitoring system that estimates machine system deterioration to assist engineers in determining which maintenance actions to take. Conventional data-driven methods such as machine learning, have two issues. One is data dependency. The accuracy of a data-driven method depends on the data volume because a data-driven method builds a classification model on the basis of historical data as training data. However, it is difficult to acquire enough data on all deterioration modes, which requires a long time, because deterioration modes are diverse, and some of them rarely happen. The other issue is interpretability. When a condition monitoring system using a data-driven method sends the degree of deterioration (DoD) of the machine system to maintenance engineers, they have difficulty in understanding the results because the method is a black box. The objective of this paper is to address these two issues. We propose a model-based method that simulates machine system deterioration with a cyber physical system (CPS). Model-based methods address these issues in the following manner. First, the methods can simulate the progress of deterioration from an initial condition to failure to estimate the DoD. Second, the methods employ mathematical models that represent machine systems. Engineers create such mathematical models (which we call “physical models”) by referring to various kinds of knowledge like design information and the result of failure mode and effects analysis. A physical model allows us to reason about a machine system to address interpretability. For dealing with machinery that has multiple operation modes, we introduce a state space to clarify the relationship among input, observable state variables, and DoD in a physical model. The CPS estimates DoD by comparing observed data with simulated data in the state space. In our case study, we evaluated our proposed method with a hydraulic pump of a mining machine. First we created a physical model with Modelica, which is a multi-domain modeling language. Then, the method constructed the state space by simulating deterioration with the physical model given all combinations of inputs and DoD. After that, we showed that the estimated DoD tended to increase until the hydraulic pump was replaced, using the observed data from an actual mining machine. As a result, the experimental results revealed that the proposed method succeeded in identifying the DoD with observed data of the hydraulic pump of a mining machine.

Random polarization distribution of multi-domain model for polycrystalline ferroelectric HfZrO2

Device dimension scaling down to be comparable to the domain size of polycrystalline ferroelectric HfZrO2 (HZO) is evaluated for subthreshold swing (SS) and drain-induced barrier lowering (DIBL) by numerical simulation. The proposed multi-domain modeling involves polarization random location in HZO and probability with Gaussian distribution, as well as being integrated with the Landau–Khalatnikov equation. A small device with a few domains exhibits steep SS compared with large dimension with many domains. The N-DIBL (negative-DIBL) is also estimated by using this model, and the negative capacitance effect retards the short-channel effects significantly. The trend of the experimental data and simulation results of fin field-effect transistors and planar field-effect transistors is consistent with nano-scale and micro-scale devices, respectively.

Influence of Shape and Size of Foundation on Vibration Reduction of High-Speed Punching Press Based on Multidomain Modeling

With the aid of multidomain modeling method, the vibration of a high-speed punching press was modeled and predicted, and the influence of the installation of foundation, as well as the shape and size parameters of the foundation on the vibration of the punching press, is discussed. The rectangular foundation size with the best effect of vibration reduction under the premise of saving the installation area was obtained. The validity of the multidomain model of the high-speed punching press with the foundation for vibration reduction is verified, by comparing the simulation results with the experimental results.

Towards a Personalized Multi-Domain Digital Neurophenotyping Model for the Detection and Treatment of Mood Trajectories.

The commercial availability of many real-life smart sensors, wearables, and mobile apps provides a valuable source of information about a wide range of human behavioral, physiological, and social markers that can be used to infer the user’s mental state and mood. However, there are currently no commercial digital products that integrate these psychosocial metrics with the real-time measurement of neural activity. In particular, electroencephalography (EEG) is a well-validated and highly sensitive neuroimaging method that yields robust markers of mood and affective processing, and has been widely used in mental health research for decades. The integration of wearable neuro-sensors into existing multimodal sensor arrays could hold great promise for deep digital neurophenotyping in the detection and personalized treatment of mood disorders. In this paper, we propose a multi-domain digital neurophenotyping model based on the socioecological model of health. The proposed model presents a holistic approach to digital mental health, leveraging recent neuroscientific advances, and could deliver highly personalized diagnoses and treatments. The technological and ethical challenges of this model are discussed.

A multi-domain prognostic model of disorder of consciousness using resting-state fMRI and laboratory parameters.

Although laboratory parameters have long been recognized as indicators of outcome of traumatic brain injury (TBI), it remains a challenge to predict the recovery of disorder of consciousness (DOC) in severe brain injury including TBI. Recent advances have shown an association between alterations in brain connectivity and recovery from DOC. In the present study, we developed a prognostic model of DOC recovery via a combination of laboratory parameters and resting-state functional magnetic resonance imaging (fMRI). Fifty-one patients with DOC (age = 52.3 ± 15.2 y, male/female = 31/20) were recruited from Hangzhou Hospital of Zhejiang CAPR and were sub-grouped into conscious (n = 34) and unconscious (n = 17) groups based upon their Glasgow Outcome Scale-Extended (GOS-E) scores at 12-month follow-ups after injury. Resting-state functional connectivity, network nodal measures (centrality), and laboratory parameters were obtained from each patient and served as features for support vector machine (SVM) classifications. We found that functional connectivity was the most accurate single-domain model (ACC: 70.1% ± 4.5%, P = 0.038, 1000 permutations), followed by degree centrality, betweenness centrality, and laboratory parameters. The stacked multi-domain prognostic model(ACC: 73.4% ± 3.1%, P = 0.005, 1000 permutations) combining all single-domain models yielded a significantly higher accuracy compared to that of the best-performing single-domain model (P = 0.002). Our results suggest that laboratory parameters only contribute to the outcome prediction of DOC patients, whereas combining information from neuroimaging and clinical parameters may represent a strategy to achieve a more accurate prognostic model, which may further provide better guidance for clinical management of DOC patients.

Multidomain modeling of moving coil speakers including multiple nonlinearities

Accurate modeling of moving coil speakers must include nonlinearities in the magnetic force factor, mechanical mounting stiffness, and magnetic circuits. The same model must also include complicated linear effects such as eddy currents in the magnetic structure and thermoviscous losses in enclosures and small transmission paths. Including all of these in a single model is beyond the capability of conventional analog circuit models. Multidomain modeling with Simscape or Modelica offers a method of solution for all the features mentioned. A Simscape model with several nonlinearities will be demonstrated and compared with measurements from prototype speakers.

Digital Luminaire Design Using LED Digital Twins—Accuracy and Reduced Computation Time: A Delphi4LED Methodology

Light-emitting diode (LED) digital twins enable the implementation of fast digital design flows for LED-based products as the lighting industry moves towards Industry 4.0. The LED digital twin developed in the European project Delphi4LED mimics the thermal-electrical-optical behavior of a physical LED. It consists of two parts: a package-level LED compact thermal model (CTM), coupled to a chip-level multi-domain model. In this paper, the accuracy and computation time reductions achieved by using LED CTMs, compared to LED detailed thermal models, in 3D system-level models with a large number of LEDs are investigated. This is done up to luminaire-level, where all heat transfer mechanisms are accounted for, and up to 60 LEDs. First, we characterize a physical phosphor-converted white high-power LED and apply LED-level modelling to produce an LED detailed model and an LED CTM following the Delphi4LED methodology. It is shown that the steady-state junction temperature errors of the LED CTM, compared to the detailed model, are smaller than 2% on LED-level. To assess the accuracy and the reduction of computation time that can be realized in a 3D system-level model with a large number of LEDs, two use cases are considered: (1) an LED module-level model, and (2) an LED luminaire-level model. In the LED module-level model, the LED CTMs predict junction temperatures within about 6% of the LED detailed models, and reduce the calculation time by up to nearly a factor 13. In the LED luminaire-level model, the LED CTMs predict junctions temperatures within about 1% of LED detailed models and reduce the calculation time by about a factor of 4. This shows that the achievable computation time reduction depends on the complexity of the 3D model environment. Nevertheless, the results demonstrate that using LED CTMs has the potential to significantly decrease computation times in 3D system-level models with large numbers of LEDs, while maintaining junction temperature accuracy.