Physics-based Modeling Method Research Articles

Physics-informed ensemble learning with residual modeling for enhanced building energy prediction

Accurate modeling of building energy use is important to support a diverse spectrum of its downstream applications, such as building energy efficiency assessment, resilience analysis, smart control, etc. As mainstream approaches of building energy modeling, physics-based modeling builds on different fidelities of physics rules yet are usually compromised in modeling accuracy due to insufficiency of physics rules to capture real-world dynamics and incomplete input information. Data-driven approaches are computationally efficient, but black box (uninterpretable) in nature. For improved modeling of building energy use, this work proposes a physics-informed ensemble learning approach in building energy prediction through residual modeling. Specifically, we first analyze the components of building energy use data. Evidence suggests that the building energy use data can be decomposed into physics-driven part, occupant-driven part, and white noise. Second, high-fidelity physics-based building models (EnergyPlus) and low-fidelity ones (RC models) are developed to capture the physics-driven part while time series methods are explored as the residual modeling approach to capture the occupant-driven discrepancies between physics-based simulation and measured building energy use (i.e., residuals). Finally, the physics-informed ensemble learning is proposed to integrate physics-based and data-driven models for enhanced accuracy and robustness of building energy modeling. Results demonstrate 40–90% increase of accuracy between modeling and field observations compared to traditional physics-based modeling methods. Moreover, when the training dataset size is small, the proposed ensemble model overperforms the pure data-driven models, demonstrating its higher robustness in extrapolation scenarios. This work makes fundamental contributions to the development of convergent modeling approaches in the building modeling field.

Evaluation of finite element modeling methods for predicting compression screw failure in a custom pelvic implant.

Introduction: Three-dimensional (3D)-printed custom pelvic implants have become a clinically viable option for patients undergoing pelvic cancer surgery with resection of the hip joint. However, increased clinical utilization has also necessitated improved implant durability, especially with regard to the compression screws used to secure the implant to remaining pelvic bone. This study evaluated six different finite element (FE) screw modeling methods for predicting compression screw pullout and fatigue failure in a custom pelvic implant secured to bone using nine compression screws. Methods: Three modeling methods (tied constraints (TIE), bolt load with constant force (BL-CF), and bolt load with constant length (BL-CL)) generated screw axial forces using functionality built into Abaqus FE software; while the remaining three modeling methods (isotropic pseudo-thermal field (ISO), orthotropic pseudo-thermal field (ORT), and equal-and-opposite force field (FOR)) generated screw axial forces using iterative physics-based relationships that can be implemented in any FE software. The ability of all six modeling methods to match specified screw pretension forces and predict screw pullout and fatigue failure was evaluated using an FE model of a custom pelvic implant with total hip replacement. The applied hip contact forces in the FE model were estimated at two locations in a gait cycle. For each of the nine screws in the custom implant FE model, likelihood of screw pullout failure was predicted using maximum screw axial force, while likelihood of screw fatigue failure was predicted using maximum von Mises stress. Results: The three iterative physics-based modeling methods and the non-iterative Abaqus BL-CL method produced nearly identical predictions for likelihood of screw pullout and fatigue failure, while the other two built-in Abaqus modeling methods yielded vastly different predictions. However, the Abaqus BL-CL method required the least computation time, largely because an iterative process was not needed to induce specified screw pretension forces. Of the three iterative methods, FOR required the fewest iterations and thus the least computation time. Discussion: These findings suggest that the BL-CL screw modeling method is the best option when Abaqus is used for predicting screw pullout and fatigue failure in custom pelvis prostheses, while the iterative physics-based FOR method is the best option if FE software other than Abaqus is used.

Knee-Loading Predictions with Neural Networks Improve Finite Element Modeling Classifications of Knee Osteoarthritis: Data from the Osteoarthritis Initiative

Physics-based modeling methods have the potential to investigate the mechanical factors associated with knee osteoarthritis (OA) and predict the future radiographic condition of the joint. However, it remains unclear what level of detail is optimal in these methods to achieve accurate prediction results in cohort studies. In this work, we extended a template-based finite element (FE) method to include the lateral and medial compartments of the tibiofemoral joint and simulated the mechanical responses of 97 knees under three conditions of gait loading. Furthermore, the effects of variations in cartilage thickness and failure equation on predicted cartilage degeneration were investigated. Our results showed that using neural network-based estimations of peak knee loading provided classification performances of 0.70 (AUC, p < 0.05) in distinguishing between knees that developed severe OA or mild OA and knees that did not develop OA eight years after a healthy radiographic baseline. However, FE models incorporating subject-specific femoral and tibial cartilage thickness did not improve this classification performance, suggesting there exists an optimal point between personalized loading and geometry for discrimination purposes. In summary, we proposed a modeling framework that streamlines the rapid generation of individualized knee models achieving promising classification performance while avoiding motion capture and cartilage image segmentation.

Physics-Dominated Neural Network for Spatiotemporal Modeling of Battery Thermal Process

Modeling the temperature distribution of a battery is critical to its safe operation. Data-based modeling methods are computationally efficient, but require a large number of sensors; while physics-based modeling methods have better generalization, but the unknown dynamics of the actual scene are ignored. A physics-dominated neural network is presented to integrate electric-thermal mechanism of the battery and data information through a weight adaptive function. The electric-thermal coupling equation of the battery under complex conditions is taken as the prior knowledge to update parameters of the network; while the characteristic data obtained by the unique sensor is used to compensate the unknown disturbance in the actual scene. A well-trained model can predict the temperature distribution of the battery over entire space with a single sensor, and can also provide reasonable predictions for longer periods of time under extreme conditions. Experiments show that the proposed method outperforms traditional methods that rely only on pure data or pure physics.

Physics-Informed Deep Learning for Muscle Force Prediction With Unlabeled sEMG Signals.

Computational biomechanical analysis plays a pivotal role in understanding and improving human movements and physical functions. Although physics-based modeling methods can interpret the dynamic interaction between the neural drive to muscle dynamics and joint kinematics, they suffer from high computational latency. In recent years, data-driven methods have emerged as a promising alternative due to their fast execution speed, but label information is still required during training, which is not easy to acquire in practice. To tackle these issues, this paper presents a novel physics-informed deep learning method to predict muscle forces without any label information during model training. In addition, the proposed method could also identify personalized muscle-tendon parameters. To achieve this, the Hill muscle model-based forward dynamics is embedded into the deep neural network as the additional loss to further regulate the behavior of the deep neural network. Experimental validations on the wrist joint from six healthy subjects are performed, and a fully connected neural network (FNN) is selected to implement the proposed method. The predicted results of muscle forces show comparable or even lower root mean square error (RMSE) and higher coefficient of determination compared with baseline methods, which have to use the labeled surface electromyography (sEMG) signals, and it can also identify muscle-tendon parameters accurately, demonstrating the effectiveness of the proposed physics-informed deep learning method.

A Novel Physics-Informed Hybrid Modeling Method for Dynamic Vibration Response Simulation of Rotor–Bearing System

For rotor–bearing systems, their dynamic vibration models must be built to simulate the vibration responses that affect the safe and reliable operation of rotating machinery under different operating conditions. Single physics-based modeling methods can be used to produce sufficient but inaccurate vibration samples at the cost of computational complexity. Moreover, single data-driven modeling methods may be more accurate, employing larger numbers of measured samples and reducing computational complexity, but these methods are affected by the insufficient and imbalanced samples in engineering applications. This paper proposes a physics-informed hybrid modeling method for simulating the dynamic responses of rotor–bearing systems to vibration under different rotor speeds and bearing health statuses. Firstly, a three-dimensional model of a rolling bearing and its supporting force are introduced, and a physics-based dynamic vibration model that couples flexible rotors and rigid bearings is constructed using multibody dynamics simulation. Secondly, combining the simulation vibration data obtained using the physics-based model with measured vibration data, algorithms are designed to learn vibration generation and data mapping networks in series connection to form a physics-informed hybrid model, which can quickly and accurately output the vibration responses of a rotor–bearing system. Finally, a case study on the single-span rotor platform is provided. By comparing the signal output by the proposed physics-informed hybrid modeling method with the measured signal in the time and frequency domains, the effectiveness of proposed method under both constant- and variable-speed operating conditions are illustrated.

A physics-based modeling method of THz Schottky diode for circuit simulation

A modeling method for terahertz Schottky diode is proposed based on the special physical effects of Schottky diode in the terahertz frequency band. In this method, the device is split into a nonlinear Schottky junction part and a linear parasitic part, which are modeled separately. The nonlinear part uses the physical based modeling method by analyzing the special nonlinear effects generated by the device in the terahertz band, focusing on the modeling of the current saturation effect, and then extracting the intrinsic parameters according to the established model; for the linear part of the diode, its 3D electromagnetic (EM) model is built in the EM field simulation software to extract the parasitic parameters of the diode. Finally, the model is implemented using the circuit simulator HSpice. The model is verified by comparing the parameters extracted from the model with the actual device electrical parameters. This method ensures accuracy and convenience at the same time.

A method for melt pool state monitoring in laser-based direct energy deposition based on DenseNet

Detecting and classifying the melt pool states in laser-based direct energy deposition (L-DED) is crucial for reducing defects and enhancing the mechanical properties of L-DED metal parts. Although physics-based modeling methods and traditional machine learning algorithms such as convolutional neural network have been introduced to monitor the melt pool states, improving the low accuracy of these methods remains to be challenging. To address this issue, we developed a DenseNet-39 model to classify the melt pool states. 80 single-track samples were fabricated using a linear scan strategy by L-DED and using a coaxial high-speed camera to capture the melt pool images in-process. Experimental results have demonstrated the superior performance of DenseNet-39 in classifying the melt pool states with 99.3% accuracy, achieved a lower computation burden, and less processing time. We used CAM to explain the mechanism of classification by DenseNet-39. DenseNet-39 provides the potential applications of online process monitoring in L-DED.

Physics-Based Neural Network Methods for Solving Parameterized Singular Perturbation Problem

This work is devoted to the description and comparative study of some methods of mathematical modeling. We consider methods that can be applied for building cyber-physical systems and digital twins. These application areas add to the usual accuracy requirements for a model the need to be adaptable to new data and the small computational complexity allows it to be used in embedded systems. First, we regard the finite element method as one of the “pure” physics-based modeling methods and the general neural network approach as a variant of machine learning modeling with physics-based regularization (or physics-informed neural networks) and their combination. A physics-based network architecture model class has been developed by us on the basis of a modification of classical numerical methods for solving ordinary differential equations. The model problem has a parameter at some values for which the phenomenon of stiffness is observed. We consider a fixed parameter value problem statement and a case when a parameter is one of the input variables. Thus, we obtain a solution for a set of parameter values. The resulting model allows predicting the behavior of an object when its parameters change and identifying its parameters based on observational data.

Convective Drying Modeling Approaches: a Review for Herbs, Vegetables, and Fruits

This paper presents advances in drying systems made through research and development. The advances have borne novel technologies that have improved energy use and optimized on food quality. Using a scale approach to modelling as macroscale and microscale models, the paper underscores the importance of modelling in research and development of drying systems. Gaps are identified in development of material properties in modelling, development of comprehensive Multiphysics models and user-friendly software for solution computation solution of the models. Perspectives into the future of drying modelling should be concerted on the development of material properties, more complete multiphysics models, and development of software that is more user-friendly and is accurate. This would facilitate the convergence of research and industry in terms of developing advanced processes. The physics-based modeling methods have shown to be promising and will presumably be an imperative tool in research and development.

An improved hybrid modeling method based on extreme learning machine for gas turbine engine

Engine model plays a critical role in many applications of gas turbine engine, especially for fault tolerant control and gas-path prognostic and health management. However, traditional pure physics-based modeling method has difficulty in dealing with engine-to-model mismatch, sensor measurement uncertainty, etc., leading to a big limitation in real applications. In this paper, a novel improved hybrid modeling (IHM) method is proposed to accurately model a dual spool turbofan gas turbine engine. The proposed IHM method is innovatively formulated by combining a nonlinear physics-based model (PBM), a series of residual learning models (RLMs) and an input selection module for each RLM. The PBM is constructed based on component level method while the RLMs are constructed based on extreme learning machine (ELM). Particularly, the parameters applied to train the RLMs are selected independently in a customized way based on the input selection module. The performance of the proposed method is illustrated through a series of simulations considering engine-to-model mismatch, sensor measurement uncertainty and verified through a group of ground test data. The results demonstrate that the proposed IHM method can improve modeling accuracy effectively compared with PBM. In addition, through incorporating input selection module, the RLMs in IHM have better performance than using traditional input selection way. Meanwhile, when compared to other networks, the ELM-based RLM guarantees faster training speed while achieving better residual learning effect, and shows more effectiveness and superiority.

Model-Based Scale-Up Methodologies for Pharmaceutical Granulation.

In the pharmaceutical industry, it is a major challenge to maintain consistent quality of drug products when the batch scale of a process is changed from a laboratory scale to a pilot or commercial scale. Generally, a pharmaceutical manufacturing process involves various unit operations, such as blending, granulation, milling, tableting and coating and the process parameters of a unit operation have significant effects on the quality of the drug product. Depending on the change in batch scale, various process parameters should be strategically controlled to ensure consistent quality attributes of a drug product. In particular, the granulation may be significantly influenced by scale variation as a result of changes in various process parameters and equipment geometry. In this study, model-based scale-up methodologies for pharmaceutical granulation are presented, along with data from various related reports. The first is an engineering-based modeling method that uses dimensionless numbers based on process similarity. The second is a process analytical technology-based modeling method that maintains the desired quality attributes through flexible adjustment of process parameters by monitoring the quality attributes of process products in real time. The third is a physics-based modeling method that involves a process simulation that understands and predicts drug quality through calculation of the behavior of the process using physics related to the process. The applications of these three scale-up methods are summarized according to granulation mechanisms, such as wet granulation and dry granulation. This review shows that these model-based scale-up methodologies provide a systematic process strategy that can ensure the quality of drug products in the pharmaceutical industry.

Whole building energy model for HVAC optimal control: A practical framework based on deep reinforcement learning

Whole building energy model (BEM) is a physics-based modeling method for building energy simulation. It has been widely used in the building industry for code compliance, building design optimization, retrofit analysis, and other uses. Recent research also indicates its strong potential for the control of heating, ventilation and air-conditioning (HVAC) systems. However, its high-order nature and slow computational speed limit its practical application in real-time HVAC optimal control. Therefore, this study proposes a practical control framework (named BEM-DRL) that is based on deep reinforcement learning. The framework is implemented and assessed in a novel radiant heating system in an existing office building as a case study. The complete implementation process is presented in this study, including: building energy modeling for the novel heating system, multi-objective BEM calibration using the Bayesian method and the Genetic Algorithm, deep reinforcement learning training and simulation results evaluation, and control deployment. By analyzing the real-life control deployment data, it is found that BEM-DRL achieves 16.7% heating demand reduction with more than 95% probability compared to the old rule-based control. However, the framework still faces the practical challenges including building energy modeling of novel HVAC systems and multi-objective model calibration. Systematic study is also needed for the design of deep reinforcement learning training to provide a guideline for practitioners.

Modeling and Simulation of Planes With Decoupling Capacitors

Based on the resonant cavity model, this paper presents a modeling approach for power/ground planes with decoupling capacitors. A bare plane pair is first modeled by the resonant cavity model. Then, decoupling capacitors are divided into segments according to port coefficients, capacitance, the number of decoupling capacitors, and the frequency response of each mode of the resonant cavity model. To get the model of the plane pair with decoupling capacitors, divided segments are incorporated into the resonant cavity model. By this way, not only the influence of decoupling capacitors on each single mode, but also the frequency responses of capacitors in each mode can be reflected. By being incorporated with the segmentation method and the physics-based modeling method, this modeling method can be extended to handle the irregular-shaped and multilayered power/ground plane with capacitors, respectively. Based on the impedance distribution throughout the whole power/ground plane pair, the effective decoupling radius of a capacitor and its variation trends with frequency, capacitance, and parasitic parameters are extracted and analyzed.

Failure analysis of power electronic devices and their applications under extreme conditions

Power electronic devices are the core components of modern power converters, not only for normal applications, but also for extreme conditions. Current design of power electronic devices require large redundancies for reliability. This results in huge volume and weight for a large-capacity power converter, especially for some extreme applications. Therefore, to optimize the power density, the reliability of power devices needs to be investigated first in order to obtain the accurate operational margin of a power device. Although much research on device failure analysis has been reported, there still lacks efficient failure evaluation methods. This paper first summarizes the current failure research. Then, a three-step failure analysis method of power electronic devices is proposed as: failure information collection, failure identification and mechanism, and failure evaluation. The physics-based modeling method is emphasized since it has a strong relationship with the device fundamentals. After that, power electronic device applications under extreme conditions are introduced and a design method of device under extreme conditions is proposed based on the thermal equilibrium idea. Finally, the challenges and prospects to improve the power device reliability under extreme conditions are concluded.

Modeling the Effects of Minimum Quantity Lubrication on Machining Force, Temperature, and Residual Stress

□ Residual stress is one of the critical characteristics for assessing the qualities and functionalities of machined products in light of its direct effect on endurance limit, distortion, and corrosion resistance. Primary factors responsible for residual stresses distribution include mechanical effects, thermal effects, microstructure evolutions, and a combination of these mechanisms. This study investigates the effects of minimum quantity lubrication (MQL) on machining force, temperature and residual stress through a physics-based modeling method. Both the lubrication and cooling effects caused by MQL air-oil mixture contribute to changes in friction due to boundary lubrication as well as variations in the thermal stress due to heat loss. The modified Oxley's model is employed to predict the cutting force and temperature directly from cutting conditions. The predicted cutting force and temperature are then coupled into a thermal-mechanical model which incorporates the kinematic hardening and strain compatibility to predict the machining-induced residual stress under lubricated conditions. The proposed analytical method is experimentally verified by orthogonal cutting tests for AISI 4130 alloy steel in the context of forces, temperatures, and residual stresses.