Simulation Step Research Articles

Peridynamic modeling of transient heat conduction in solids using a highly efficient algorithm

This article presents a novel peridynamic algorithm to model the transient heat conduction in two-dimensional (2-D) problems. The peridynamic method is used to solve the heat transfer equation, and a new algorithm is developed to simulate heat conduction efficiently. The new peridynamic algorithm can simulate the heat conduction for the whole geometry within only 2 s when using up to 5,000 particles and 5,000 simulation steps, where the mesh-free nature of peridynamics is well utilized. Moreover, the thermal results obtained from the new peridynamic algorithm demonstrate high accuracy by solving several heat conduction problems under different boundary conditions. The results are validated against exact and analytical solutions and compared with other numerical models. The developed algorithm can be easily implemented in software packages and industrial applications.

A collaborative framework for unifying typical multidimensional solar cell simulations – Part I. Ten common simulation steps and representing variables

AbstractMultidimensional simulations for diverse solar cells often encounter distinctive configurations, even when employing the same simulation software. The complexity and inefficiency of this process are further exacerbated when employing different simulators. From our extensive decade‐long experience in numerical simulations of diverse solar cells, we have identified ten common simulation steps intrinsic to typical electrical and optical simulations. Subsequently, we propose ten sets of variables that encompass all the relevant details required for these steps. To address the challenge of varying information requirements for each variable across different simulations, we assign a list, a versatile data type, to each variable. This approach, by design, enables concise, coherent, and flexible input, accommodating the unique demands of each simulation. However, to ensure unambiguous simulations, precise specifications for these variables are essential. Computer code has been successfully implemented to ensure adherence to specifications and expedite variable synchronization with Sentaurus, the de facto standard for device simulation. Within this framework, users are only tasked with editing variables in a plain text file, obviating the need for in‐depth knowledge of Sentaurus. This streamlines the prerequisites for engaging in numerical simulation significantly. Through thoughtful design considerations, we preserve the simulation capacity while simultaneously enhancing productivity considerably. This open‐source framework welcomes global collaboration within the photovoltaic community and has the potential to generate an extensive dataset for cost‐effective artificial intelligence training.

Modelling of Photovoltaic Generators Based on a Linarized Equivalnet Circuit

In this paper, a new method on modeling of PV generators, for association with electromagnetic software programs, is proposed. The application of a linearization process, on nonlinear and well-tested current-voltage equations using Taylor’s scheme, leads to uncoupling of the current and voltage quantities in each time step of a digital simulation. This uncoupling is represented by a linearized equivalent electrical circuit. Using the proposed equivalent photovoltaic generator circuit, the application of nodal analysis on equivalent resistive circuits derives a photovoltaic system model of linear algebraic equations. This remarkably simplifies photovoltaic systems modeling because one can develop a photovoltaic generator element in electromagnetic transient programs for power systems analysis.

Estimation of Activity and Molar Excess Gibbs Energy of Binary Liquid Alloys Pb-Sn, Al-Sn and In-Zn from the Partial Radial Distribution Function Simulated by Ab Initio Molecular Dynamics

For the present, it is difficult to obtain thermodynamic data for binary liquid alloys by experimental measurements. In this study, the molecular dynamics processes of the binary liquid alloys Pb50-Sn50, Al50-Sn50, and In50-Zn50 were simulated by using the ab initio molecular dynamics (AIMD) principle, and their partial radial distribution functions (PRDF) were obtained at different simulation steps. Combined with the relevant binary parameters of the Molecular Interaction Volume Model (MIVM), Regular Solution Model (RSM), Wilson Model, and Non-Random Two-Liquid (NRTL) models. The integral terms containing the PRDF were computed using the graphical integration method to obtain the parameters of these models, thus estimating their activity and molar excess Gibbs energy. The total average relative deviations (ARD) of the activity and molar excess Gibbs energy estimates of the four models for the binary liquid alloys Pb50-Sn50, Al50-Sn50, and In50-Zn50 at full concentration when the PRDF is obtained by the symmetry method are MIVM: 21.59% and 59.35%; RSM: 21.63% and 60.27%; Wilson: 24.27% and 86.7%; NRTL: 23.9% and 83.24%. When the PRDF is obtained by the asymmetric method: MIVM: 22.86% and 68.08%; RSM: 32.84% and 68.66%; Wilson: 25.14% and 82.75%; NRTL: 24.49% and 85.74%. This indicates that the estimation performance of the MIVM model is superior to the other three models, and the symmetric method performs better than the asymmetric method. The present study also derives and verifies the feasibility of Sommer’s equation for estimating the molar excess Gibbs energy and activity of binary liquid alloy systems in the Miedema model by using different equations of enthalpy of mixing versus excess entropy given by Tanaka, Ding, and Sommer. The total ARD of Tanaka, Ding, and Sommer’s relational equations in the Miedema model for estimating the activities and molar excess Gibbs energies of the binary liquid alloys Pb-Sn, Al-Sn, and In-Zn are 3.07% and 8.92%, 6.09% and 17.1%, and 4.1% and 14.77%. The results indicate that the estimation performance of the Miedema model is superior to the other four models.

Addressing Challenges in Dynamic Modeling of Stewart Platform using Reinforcement Learning-Based Control Approach

In this paper, we focus on enhancing the performance of the controller utilized in the Stewart platform by investigating the dynamics of the platform. Dynamic modeling is crucial for control and simulation, yet challenging for parallel robots like the Stewart platform due to closed-loop kinematics. We explore classical methods to solve its inverse dynamical model, but conventional approaches face difficulties, often resulting in simplified and inaccurate models. To overcome this limitation, we propose a novel approach by replacing the classical feedforward inverse dynamic block with a reinforcement learning (RL) agent, which, to our knowledge, has not been tried yet in the context of the Stewart platform control. Our proposed methodology utilizes a hybrid control topology that combines RL with existing classical control topologies and inverse kinematic modeling. We leverage three deep reinforcement learning (DRL) algorithms and two model-based RL algorithms to achieve improved control performance, highlighting the versatility of the proposed approach. By incorporating the learned feedforward control topology into the existing PID controller, we demonstrate enhancements in the overall control performance of the Stewart platform. Notably, our approach eliminates the need for explicit derivation and solving of the inverse dynamic model, overcoming the drawbacks associated with inaccurate and simplified models. Through several simulations and experiments, we validate the effectiveness of our reinforcement learning-based control approach for the dynamic modeling of the Stewart platform. The results highlight the potential of RL techniques in overcoming the challenges associated with dynamic modeling in parallel robot systems, promising improved control performance. This enhances accuracy and reduces the development time of control algorithms in real-world applications. Nonetheless, it requires a simulation step before practical implementations.

Artificial neural networks in the modeling of the catalytic activity of a biosensor composed of conjugated polymers and urease.

Thin films of conjugated polymer and enzyme can be used to unravel the interaction between components in a biosensor. Using artificial neural networks (ANNs) improves data interpretability and helps construct models with great capacity for classifying and processing information. The present work used kinetic data from the catalytic activity of urease immobilized in different conjugated polymers to create ANN models using time, substrate concentration, and absorbance as input variables since the models had absorbance in a posterior instant as output value to explore the predictivity of the ANNs. The performance of the models was evaluated by Pearson's correlation coefficient (ρ) and mean squared error (MSE) values. After the learning process, a series of new experiments were performed to verify the generality of the models. As the main results, the best ANN model presented 0.9980 and 3.0736 × 10-5 for ρ and MSE, respectively. For the simulation step, intermediary values of substrate concentration were used. The mean absolute percentage error (MAPE) values were 3.34, 3.07, and 3.78 for 12mM, 22mM, and 32mM concentrations, respectively. Overall, with the simulations, it was possible to ascertain the interpolatory capacity of the model, which has a learning mechanism based on absorbance and time as variables. Thus, the potential of ANNs would be in their use in pre-evaluations, helping to determine the substrate concentration at which there is higher catalytic activity or in determining the linear range of the sensor.

Development and Effectiveness of a Rapid Cycle Deliberate Practice Neonatal Resuscitation Simulation Program: A Quasi-Experimental Study.

The Rapid Cycle Deliberate Practice (RCDP) simulation during neonatal resuscitation program (NRP) training provides in-event feedback for each simulation step, repeats the simulation from the beginning, and undergoes a continuous improvement process. It also offers after-event debriefing that involves follow-up discussion and reflection after completing simulations. These two methods differ in the timing and frequency of feedback application, and there may be differences in the effectiveness of neonatal resuscitation training. A quasi-experimental simulation study with a pre- and post-test design was used; the experimental group received RCDP simulation NRP training, based on the self-determination theory, while the control group received an after-event debriefing, following the NRP scenario. The experimental group displayed significantly improved clinical decision-making skills compared with the control group. When responding to emergencies involving high-risk newborns, we found that RCDP simulation during NRP training and better preparation for neonatal resuscitation among nursing students improved outcomes for newborns.

Efficient Deep Spiking Multilayer Perceptrons With Multiplication-Free Inference.

Advancements in adapting deep convolution architectures for spiking neural networks (SNNs) have significantly enhanced image classification performance and reduced computational burdens. However, the inability of multiplication-free inference (MFI) to align with attention and transformer mechanisms, which are critical to superior performance on high-resolution vision tasks, imposes limitations on these gains. To address this, our research explores a new pathway, drawing inspiration from the progress made in multilayer perceptrons (MLPs). We propose an innovative spiking MLP architecture that uses batch normalization (BN) to retain MFI compatibility and introduce a spiking patch encoding (SPE) layer to enhance local feature extraction capabilities. As a result, we establish an efficient multistage spiking MLP network that blends effectively global receptive fields with local feature extraction for comprehensive spike-based computation. Without relying on pretraining or sophisticated SNN training techniques, our network secures a top-one accuracy of 66.39% on the ImageNet-1K dataset, surpassing the directly trained spiking ResNet-34 by 2.67%. Furthermore, we curtail computational costs, model parameters, and simulation steps. An expanded version of our network compares with the performance of the spiking VGG-16 network with a 71.64% top-one accuracy, all while operating with a model capacity 2.1 times smaller. Our findings highlight the potential of our deep SNN architecture in effectively integrating global and local learning abilities. Interestingly, the trained receptive field in our network mirrors the activity patterns of cortical cells.

Computer simulation of operation plant effective modes for water disinfection by electrical discharges in gas bubbles

Purpose. Determination by means of computer simulation of the most efficient modes of operation of the installation for water disinfection using discharges in gas bubbles, in which (modes) the amplitude of voltage pulses at the processing unit and on the layer of treated water is not less than the voltage amplitude immediately after the switching discharger. Methodology. To achieve this goal, we used computer simulation using Micro-Cap 10. We used two different electrical circuits that simulate the operation of the experimental setup in two different modes: in a mode with a restoring electrical strength of the discharge gap in the gas bubble between two adjacent voltage pulses on the discharge node and in the mode without restoring this dielectric strength. In computer simulation, we varied the following factors: the maximum simulation step, inductances, capacitances, active resistances, wave resistance of a long line, and the delay time for the operation of a spark gap simulating a discharge gap in a gas bubble. Results. Computer modeling has shown that in order to increase the voltage amplitude at the treatment unit and on the layer of treated water, it is necessary to reduce the load capacitance – the capacitance of the water layer in the treatment unit to 10 pF or less, to increase the active resistance of the water layer to 500 W or more. An important factor for increasing the voltage and electric field strength in the discharge unit and, consequently, for increasing the efficiency of treated water disinfection is the discharge delay time in gas bubbles. The most rational delay time for the operation of the arrester, which is the gap in the gas bubble inside the water, under the conditions considered by us is 4–5 ns. It is with this delay time that the amplitude of voltage pulses at the node of disinfecting water treatment and on the layer of treated water is maximum, all other things being equal. Furthermore, with such a delay time this amplitude of voltage pulses significantly exceeds the voltage amplitude directly after the main high-voltage discharger, switching energy from the high-voltage capacitive storage to the processing unit through a long line filled with water. Originality. Using computer simulation, we have shown the possibility of increasing the voltage at the discharge unit of the experimental setup by 35 % without increasing the voltage of the power source. This provides a higher efficiency of microbiological disinfection of water by nanosecond discharges in gas bubbles and lower specific energy consumption. Practical value. The obtained results of computer simulation confirm the prospect of industrial application of installations using nanosecond discharges for disinfection and purification of wastewater, swimming pools and post-treatment of tap water.

Predictive Modeling of Light–Matter Interaction in One Dimension: A Dynamic Deep Learning Approach

The mathematical modeling and the associated numerical simulation of the light–matter interaction (LMI) process are well-known to be quite complicated, particularly for media where several electronic transitions take place under electromagnetic excitation. As a result, numerical simulations of typical LMI processes usually require a high computational cost due to the involvement of a large number of coupled differential equations modeling electron and photon behavior. In this paper, we model the general LMI process involving an electromagnetic interaction medium and optical (light) excitation in one dimension (1D) via the use of a dynamic deep learning algorithm where the neural network coefficients can precisely adapt themselves based on the past values of the coefficients of adjacent layers even under the availability of very limited data. Due to the high computational cost of LMI simulations, simulation data are usually only available for short durations. Our aim here is to implement an adaptive deep learning-based model of the LMI process in 1D based on available temporal data so that the electromagnetic features of LMI simulations can be quickly decrypted by the evolving network coefficients, facilitating self-learning. This enables accurate prediction and acceleration of LMI simulations that can run for much longer durations via the reduction in the cost of computation through the elimination of the requirement for the simultaneous computation and discretization of a large set of coupled differential equations at each simulation step. Our analyses show that the LMI process can be efficiently decrypted using dynamic deep learning with less than 1% relative error (RE), enabling the extension of LMI simulations using simple artificial neural networks.

(Invited) Molecular Dynamics Simulation Insights into Plasma Treatment of Emerging Pollutants in Water

Eliminating organic pollutant molecules in water is a world challenge, that non-thermal plasmas at atmospheric pressure are aiming to address. Non thermal plasmas are used for treating contaminated water due to their ability to produce reactive oxygen and nitrogen species, the so-called RONS, which diffuse into water, react with pollutant molecules, and degrade them. Many experimental works have been designed, using different plasma sources [1, 2], but in all cases HO● radicals are acknowledged to play the main role in the degradation process.Classical and ab-initio reactive molecular dynamics (MD) simulations are relevant techniques for describing interactions of HO● with pollutant molecules in water, leading to the knowledge of degradation products, chemical pathways and rates [3]. Basically, MD simulations are solving the Newton equations of motion of a set of species, say atoms, molecules, clusters, etc. It is able to describe system size up to billion of species. It only needs the knowledge of interactions (force fields) between all species as well as initial conditions, preferably consistent with experiments. So it is suitable for addressing chemical reactivity. Classical reactive MD simulations are using available semi-empirical force fields, such as reaxFF, while ab-initio MD are required when no such force field is available. In this case, force fields are calculated quantum mechanically at relevant time steps of MD simulations.In fact, MD simulations act as a predictive "virtual microscope" provided the force fields are enough precise.In this work, MD simulations of the interaction of HO● radicals with molecules representative of various emerging pollutant families, as pesticide, antibiotics, PFAS in water, are studied. For mimicking HO● radicals release into water by non-thermal plasma, HO● radicals are periodically injected into a simulation box containing a pollutant molecule surrounded by several tens of water molecules. For obtaining statistically significant results, at least 10 simulations, with different initial conditions are run. The products that are formed will be compared with experiments, especially HPLC, GCMS and MS/MS measurements of plasma treated water.Another approach, consists in introducing a temperature ramp during the simulation for determining at which temperature products are appearing and in which order. Such an approach is revealing the reaction barriers up to full decomposition of the pollutant molecule. It can thus, by back-analysis, help to define which plasma or other advanced oxidation processing will be relevant [3]. Acknowledgements Part of this work has been supported by Conseil Régional Centre-Val de Loire with grant #2021-00144786 for projectPerturb’Eau. Monica Magureanu, Corina Bradu, Florin Bilea, Olivier Aubry, Dunpin Hong, Manoj Panyamthatta-Tayaroth are gratefully acknowledged for stimulating discussions and for providing experimental results.

Computational study of the balloon dilation steps on transcatheter aortic valve replacement.

Balloon dilation is a commonly used assistant method in transcatheter aortic valve replacement (TAVR) and plays an important role during valve implantation procedure. The balloon dilation steps need to be fully considered in TAVR numerical simulations. This study aims to establish a TAVR simulation procedure with two different balloon dilation steps to analyze the impact of balloon dilation on the results of TAVR implantation. Two cases of aortic stenosis were constructed based on medical images. An implantation simulation procedure with self-expandable valve was established, and multiple models including different simulation steps such as balloon pre-dilation and balloon post-dilation were constructed to compare the different effects on vascular stress, stent morphology and paravalvular leakage. Results show that balloon pre-dilation of TAVR makes less impact on post-operative outcomes, while post-dilation can effectively improve the implantation morphology of the stent, which is beneficial to the function and durability of the valve. It can effectively improve the adhesion of the stent and reduce the paravalvular leakage volume more than 30% after implantation. However, balloon post-dilation may also lead to about 20% or more increased stress on the aorta and increase the risk of damage. The balloon dilation makes an important impact on the TAVR outcomes. Balloon dilation needs to be fully considered during pre-operative analysis to obtain a better clinical result.

CFD based design gas rotameters: Dynamic mesh transient simulation

Rotameters are variable-area flowmeters widely used in various industries. In this paper, a comprehensive discussion has been presented on simulation of transient behavior of a rotameter using Computational Fluid Dynamics (CFD) technique. Various methods of dynamic meshing and the reasons behind choosing the diffusion-based model have been discussed. 2D and 3D simulations were compared, and in the 2D simulation, triangular and quadratic meshes were compared, and their orthogonal quality changes due to dynamic mesh methods are shown. It was illustrated that triangular mesh cells preserve their orthogonal quality far more effectively. Moreover, the CFD simulation results were validated by comparing them with experimental data obtained from a real rotameter and it was found that the 3D simulation results are more consistent with experimental data than the 2D simulation results. A difference of less than 4% has been observed between the simulation and experimental results, indicating a reasonable agreement. The CFD solver configuration and simulation steps presented in this paper can be used to optimize rotameter designs and improve their performance. Additionally, the simulation results can provide valuable insights into the behavior of the rotameter, allowing designers to identify the potential areas for improvement and optimization of the design, accordingly.

Monte Carlo Sensitivities Using the Absolute Measure-Valued Derivative Method

Measure-valued differentiation (MVD) is a relatively new method for computing Monte Carlo sensitivities, relying on a decomposition of the derivative of transition densities of the underlying process into a linear combination of probability measures. In computing the sensitivities, additional paths are generated for each constituent distribution and the payoffs from these paths are combined to produce sample estimates. The method generally produces sensitivity estimates with lower variance than the finite difference and likelihood ratio methods, and can be applied to discontinuous payoffs in contrast to the pathwise differentiation method. However, these benefits come at the expense of an additional computational burden. In this paper, we propose an alternative approach, called the absolute measure-valued differentiation (AMVD) method, which expresses the derivative of the transition density at each simulation step as a single density rather than a linear combination. It is computationally more efficient than the MVD method and can result in sensitivity estimates with lower variance. Analytic and numerical examples are provided to compare the variance in the sensitivity estimates of the AMVD method against alternative methods.

An immersed selective discontinuous Galerkin method in particle-in-cell simulation with adaptive Cartesian mesh and polynomial preserving recovery

In this paper, a selective discontinuous Galerkin (SDG) method and a polynomial preserving recovery (PPR) method are developed and integrated with the immersed-finite-element particle-in-cell (IFE-PIC) method, in order to carry out the plasma-material interaction simulation on adaptive Cartesian meshes (ACM). To significantly save the computational cost in practice, the PIC simulation often needs to use Cartesian meshes for the whole domain and the ACM for some focused regions in the plasma-material interaction problems. Therefore, we utilize the SDG method with immersed finite elements for the field solver of the IFE-PIC method, as the key technique to allow the local Cartesian mesh refinement which will generate hanging nodes. To minimize the degree of freedom of the SDG on the ACM and reduce the computational cost, a new selective technique of the discontinuous global basis functions is proposed for the SDG. Various types of nodes in the ACM are discussed for the implementation of this selective technique, including the hanging nodes. Meanwhile, the gathering and scattering steps of the PIC simulation also need to be appropriately performed on the ACM. The PPR method is a generic gradient recovery technique to be incorporated into the IFE-PIC method for more accurate force deposition on the ACM. In addition, two other algorithmic structures of the traditional IFE-PIC method have been modified to accommodate the ACM in the simulation, including a new mapping array structure for the particle positioning and a new charge deposition with some special particle positions in the ACM. As a result, the integrated immersed-selective-discontinuous-Galerkin particle-in-cell (ISDG-PIC) method can efficiently perform the plasma-material interactions simulation on the Cartesian meshes with local mesh refinement independent of the interface. Several numerical experiments are performed to demonstrate the convergence, efficiency, and applicability of the proposed ISDG-PIC method.

Experiment on evacuation behavior: Applying the sector grid model to a clear area

Researching personnel evacuation is crucial for preventing and reducing emergencies. In this study, the movement and retention laws relating to personnel in building bottlenecks are derived from research on safe evacuation from building bottlenecks and personnel evacuation tests. Accordingly, a fan-grid partitioning method is proposed for building bottlenecks. Because the personnel distribution in the area is fan-shaped, a dynamic grid-division method is proposed and applied to the safe evacuation simulation model. Additionally, a dislocation arrangement is utilized for the front and rear grids, and the experimental analysis reveals that the time step in the safety evacuation simulation. Moreover, a grid partition model is proposed, and the evacuation parameters of the evacuation area bottleneck are characterized via the human flow rate sequence. The Visual Basic language can be utilized in the safe evacuation simulation algorithm for building bottleneck, which has great significance in guiding research on the safe evacuation of personnel during fire hazards.

Speed Optimization in DEVS-Based Simulations: A Memoization Approach

The DEVS model, designed for general discrete event simulation, explores the event status and time advance of all DEVS atomic models deployed at the time of the simulation, and then performs the scheduled simulation step. Each simulation step is accompanied by a re-exploration the event status and time advance, which is needed for maintaining the casual order of the entire model. It is time consuming to simulate a large-scale DEVS model. In a similar vein, attempts to perform an HDL simulation in a DEVS space increase simulation costs by incurring repeated search costs for model transitions. In this study, we performed a statistical analysis of engine behavior to improve simulation speed and we proposed a DP-based memoization technique for the coupled model. Through our method, we can expect significant performance improvements that range statistically from 7.4 to 11.7 times.

Simultaneous optimal site selection and sizing of a grid-independent hybrid wind/hydrogen system using a hybrid optimization method based on ELECTRE: A case study in Iran

This study presents a hybrid optimization approach to determine the optimal location and size of wind turbines and hydrogen storage systems in rural areas with high wind energy potential in three Iranian provinces. The study employs the ELECTRE decision-making method to select the optimal site, revealing Assadabad Nehbandan as the optimal site for wind farm implementation. The next step of simulation, compares a hybrid system with wind turbines and batteries to one with hydrogen storage, focusing on cost minimization and reliability maximization. The results show that the cost of energy in the battery-based hybrid system is 0.373 ($/kWh), while in the hydrogen storage-based system, it is 0.609 ($/kWh). Sensitivity analysis indicates that higher wind speeds decrease total net present cost and cost of energy, while increasing interest rates have the opposite effect. Additionally, the economic justification for battery-based hybrid system, with an initial cost of 5954 ($), surpasses the hydrogen storage-based system, costing 13,697 ($).

A novel objective for improving the sustainability of water supply system regarding hydrological response.

In general, the sustainability of the water supply system is indicative of an adaptive operational approach, wherein the decision-maker adjusts the system's performance based on the availability of water resources in a given time frame. In light of this, a novel framework is proposed in this study to evaluate sustainability, including three key indicators: resilience, reliability, and vulnerability. To address stressors that may lead to system failure, a multisectoral water allocation optimization is undertaken. In order to account for the future implications of climate change on the hydrological cycle, a simulation step, is incorporated, utilizing the Soil and Water Assessment Tool (SWAT) under various emission scenarios (RCP4.5 and RCP8.5), prior to integrating the streamflow data into our proposed optimal framework. To calibrate and validate historical data (2014-2019) and simulate future streamflow patterns (2025-2085), the Sistan Basin, located in an arid region of Iran, is analyzed as a case study. In light of the anticipated adverse impacts on the water supply system, certain adaptation measures, such as demand shrinkage scenarios, are considered to further appraise the proposed framework. Based on the final output, it is evident that the agricultural and industrial sectors, being the primary water consumers, are more susceptible to negative impacts resulting from the reduction in system sustainability. This susceptibility is primarily attributed to their highest vulnerability and comparatively lower reliability.

Guiding-centre orbit-following simulations of charge exchange loss of NBI ions with the finite Larmor radius effect

Guiding-centre orbit-following simulations of the charge exchange (CX) loss of neutral beam injection (NBI) ions are presented. The finite Larmor radius (FLR) effect in the fast ion–neutral collision can be included in guiding-centre orbit-following simulations by using the gyroaverage method. It is proved that the neutralization probability of fast ions computed by using the gyroaverage method in the guiding-centre orbit simulation is roughly the same as that computed in the full-orbit simulation when the time step in the guiding-centre simulation is the order of the gyroperiod. The CX losses of NBI fast ions for two NBIs in the EAST tokamak have been simulated by the guiding-centre orbit-following code GYCAVA, and the FLR effect in the fast ion–neutral collision on CX losses has been numerically studied. The CX effect of the fast ion–neutral collision can significantly enhance NBI ion losses on EAST. The FLR effect in the fast ion–neutral collision can enhance the CX loss. Vertical asymmetry of localized heat loads induced by CX losses is found, which is related to the FLR effect of fast ions and the strong radial gradient of the neutral density near the plasma edge. Heat loads induced by CX losses are localized in the regions near the poloidal angle $\theta =-60^\circ$ , because the likelihood of exchanging charge is the largest at gyrophase $\xi ={\rm \pi}$ , and this leads to fast downwards moving neutrals. Fast ion loss fractions induced by CX increase with the neutral density increasing.