Analysis of the Effects of a Swing Door Opening on Indoor Airflow Fields—An Experimental Study

Abstract

Occupant interactions with built environments significantly influence indoor airflow patterns. Among these interactions, door openings are common, which affect airflow fields and contribute to the dispersion of airborne contaminants. The wakes originating from alterations in airflow patterns contribute to the transport of pollutants and must be carefully considered for system design in critical environments to avoid cross-contamination of susceptible bodies (e.g., patients, precision manufacturing, etc.). Therefore, knowledge about the movement patterns of these wakes is crucial in the context of indoor air quality. In this study, a series of experiments were conducted in a controlled chamber under two different schemes of a swing door opening and three different flow regimes to study the turbulent vortices produced from door openings and their spatiotemporal propagation. Additionally, an enhanced event-based modeling (EBM) approach was employed to develop a data-driven prediction of the transient indoor air patterns resulting from door-opening activities. The results suggest a significant effect of a door opening on indoor airflow fields. The velocity fields demonstrate that consecutive openings under different ventilation conditions have a prolonged impact on the propagation of door-opening-induced wakes farther into the test chamber. The quantification of the change in kinetic energy from the door opening also shows that the level of ventilation governs the flow patterns resulting from human-induced perturbation of the steady-state flow fields. The EBM approach effectively approximated the airflow patterns and demonstrated its potential to predict transient airflow disturbances caused by door operations.

1. Introduction

Maintaining optimal indoor air quality (IAQ) is vital for ensuring occupant health, comfort, and productivity, especially in environments where contamination control is essential, such as healthcare facilities, cleanrooms, and laboratories [1,2,3]. The spread of airborne contaminants within indoor environments can lead to adverse outcomes, including the transmission of infectious diseases and compromised product integrity in sensitive manufacturing processes [4]. Given that people spend approximately 90% of their time indoors, understanding and controlling indoor airflow patterns is essential to develop meaningful contamination control [5]. Incorporating spatial pattern analysis can further enhance our understanding of airflow patterns and their impacts on occupant health and well-being [6]. One of the key factors influencing IAQ is the interaction between occupants and the built environment, particularly through activities such as door operations and human movement [7,8]. These occupant-induced disturbances can significantly disrupt the designed airflow patterns established by heating, ventilation, and air conditioning (HVAC) systems. This disruption causes unintended dispersion of contaminants [9,10]. Understanding and controlling these transient airflow disturbances are essential for enhancing the effectiveness of ventilation systems and minimizing the risk of contaminant spread.

Recent research has further explored these occupant-induced disturbances. The authors of Ref. [11] studied the enhancement effect of human movement on the high-risk range of viral aerosols exhaled by a sitting person. Their full-scale experiments demonstrated that human movement could significantly increase the transmission range of exhaled aerosols, doubling the forward high-risk range from 1.5 m to 3.0 m and increasing the lateral range from 1.3 m to 2.0 m. This enhancement effect was more pronounced when the moving person walked in the same direction as the exhalation flow. The finding highlights the importance of considering human movement in airborne transmission studies, confirming the findings of) [12]. Jo et al. (2024) [13] investigated contaminant leakage from airborne infection isolation rooms during medical staff entry. Through realistic walking motion incorporated in Computational Fluid Dynamics (CFD) simulations to enhance the accuracy of their findings, the results from that study indicated that incorporating realistic human models and movements significantly affected airflow patterns and contaminant distribution, with contaminant leakage ranging from 20.6% to 28.6%. This suggests that previous studies may have underestimated infection risks due to oversimplified models and highlights the importance of realistic modeling of human activities in airflow simulations. Feng et al. (2023) investigated the generation of internal waves by human movement and their diffusion in displacement-ventilated rooms [14]. They concluded that moving humans produce internal waves, which can affect air temperatures even at a distance from the moving body, which, in turn, impacts thermal stratification in displacement ventilation systems. The authors of [15] studied the effect of walking modes and temperatures on the strength of ventilation systems in controlling walking-induced disturbances. Their experimental results demonstrated that walking-induced disturbances significantly affect ventilation performance, with different walking modes and temperatures influencing the effectiveness of ventilation systems in controlling indoor pollutant fields.

From a systems design perspective, Li et al. (2023) [16] reviewed engineering controls for IAQ and emphasized the importance of integrating traditional IAQ controls with information technology. The review highlighted the potential of sensors and information technology in supporting IAQ management. It suggests that flexible and resilient IAQ systems are better equipped to address dynamic indoor environments influenced by occupant activities. The influence of ventilation system design and operation on contaminant dispersion has also been a focus of recent research. The authors of [17] investigated the combined effects of ventilation rate and filtration schemes on cleanroom air quality. Their study highlighted that occupant movements and interactions can influence particle dispersion, disrupt airflow, and decrease ventilation efficacy. Additionally, Feng et al. (2023) [18] conducted a comprehensive review of the dynamic characteristics of particulate matter resuspension due to human activities in indoor environments. They highlighted that human movement significantly influences particulate matter resuspension, which affects indoor air quality. The review called for more precise simulation models to capture the complex interactions between human activities and particulate matter dynamics.

Considering the impact of door operation on indoor airflow distribution, Wu et al., (2022) [11] conducted a numerical study on transient airflows and air exchange induced by door motion in thermally stratified environments. Their findings highlighted that door motion significantly affects air exchange and thermal stratification, with the amount of air exchange depending on factors such as door motion speed, opening direction, and indoor–outdoor temperature differences. Notably, about 40% of air exchange occurred when the door angle was less than 30°, even though this phase consumed only 15% of the total door motion time. Such advanced techniques are essential for quantifying environmental interactions, as demonstrated in studies on emissions and their spatial dynamics [6]. Liu et al. (2024) examined the interfacial exchange of airflow and bacteria-carrying particles induced by door opening and foot traffic in an operating room [19]. Using numerical simulations, they found that walking across the door interface caused a sharp increase in airflow exchange, and higher pressure differentials led to increased bacteria-carrying particle intrusion ratios. Their study emphasizes the impact of door operations combined with human movement on contaminant dispersion in critical environments. Previous studies have highlighted the complexity of airflow disturbances caused by door movements and occupant activities. The authors of [9] investigated the impact of door opening on airflow patterns in hospital isolation rooms. They found that door operation significantly affects pressure differentials and airflow velocities, which can result in potential contaminant spread. Similarly, Tang et al. (2006) [20] demonstrated that door openings could compromise isolation conditions in hospital settings by altering airflow patterns and facilitating airborne contaminant transmission.

Recently, more studies on the impact of door operation and human movement on indoor air quality have been taken into account [14,17]. The authors of [21], in a numerical study, investigated the influence of human activity and door opening and closing on the airflow and particle distribution. The results show that the wake of an opening of a swinging door delays aerosol escape, but when the person is walking out, it entrains aerosol out of the room. In another study, Shao et al. (2020) [22] demonstrated that when a person passes through the doorway between an airlock and a cleanroom, the particle concentration in the cleanroom rises significantly. Their experiments revealed that this increase is particularly pronounced when the doorway airflow rate is low, that is, when the amount of clean air moving through the doorway is reduced, the system is less effective at counteracting the contaminant influx caused by human movement. This is because when the door remains closed, the cleanroom experiences much lower particle concentrations compared to the period during which the door is open and occupied by a moving person. In the context of hospital isolation zones, Z. Li et al. (2024) [23] focused on airflow and aerosol transport based on dynamic scenarios. Their transient CFD simulations revealed that healthcare worker movement and door operations significantly affect airflow exchange and bioaerosol transport between negative-pressure isolation wards and buffer rooms. The study emphasized the crucial role of ventilation layout and healthcare worker positions in mitigating infection risks. It suggests that dynamic events must be considered in the design and operation of areas with higher risks, such as isolation areas. Understanding the mechanisms of contaminant dispersion due to occupant activities is essential for developing effective control strategies. Furthermore, Li et al. (2024) [24] conducted a numerical study on airflow patterns and bioaerosol dispersion during door motion and foot traffic in a negative-pressure isolation ward. Their results showed that temperature differences between wards and buffer rooms significantly impact airflow exchange and contaminant migration. Adjusting door operation modes and reducing temperature differences can help mitigate infection risks.

Despite these advancements, previous studies have largely focused on assessing volumetric air exchange across doors or modeling airflow patterns under specific conditions, often neglecting the detailed analysis of transient airflow behaviors within the specific space into which a door opens. Modeling these transient events is computationally intensive and time-consuming, and few studies have explored alternative ways to model and integrate transient phenomena for determining indoor ventilation needs [25]. To address this gap, Mousavi and Bhattacharya (2022) proposed an event-based modeling (EBM) approach to simulate airflow patterns resulting from transient events like door openings and human movements [25]. Their method simplifies the complex modeling of transient airflow disturbances by treating recurrent events as additive. This approach enables efficient approximation of airflow characteristics under various initial conditions. However, most existing models, including the original EBM, do not account for the overlapping or simultaneous occurrence of events, limiting their applicability in real-world scenarios where multiple transient events can occur concurrently.

Existing EBM approaches primarily consider only sequential, isolated events and can become computationally intensive when capturing multiple overlapping disturbances. To address this limitation and the lack of detailed spatiotemporal analysis of transient airflow dynamics, this study enhances the EBM framework to approximate airflow patterns from recurrent door-opening events with minimal computational and experimental overhead. This approach posits that recurrent door-opening events can be modeled once and applied multiple times, thereby eliminating redundant modeling efforts. Notably, the events can occur simultaneously or overlap and do not need to be sequential. Using air velocity data collected from experiments, we demonstrate how data-driven predictive modeling can approximate airflow patterns resulting from transient door-opening activities. This approach highlights the efficiency of the enhanced EBM in reducing computational demands for transient event analysis. Unlike previous research, our investigation focuses on the spatiotemporal dynamics of airflow patterns inside the room where the door opens. The novelty of this study lies in its comprehensive approach, which not only quantifies changes in flow properties introduced by different door-opening scenarios but also examines the penetration and persistence of wakes generated by door movements. By emulating common door-opening practices, this study provides new insights into the transient behaviors that influence indoor air quality and contaminant dispersion. This detailed experimental analysis offers valuable implications for the design and management of environments such as cleanrooms, operating theaters, and healthcare facilities, where precise control of airflow is critical. Therefore, building upon the identified research gap, this study addresses the following research questions:

• How does the movement of a swing door under different ventilation conditions affect the spatiotemporal dynamics of indoor airflow fields?
• What is the impact of consecutive door openings on the propagation and persistence of wakes generated by door movements in a controlled environment?
• Can an enhanced event-based modeling approach effectively approximate transient indoor air patterns resulting from door-opening activities, and how does it compare with experimental data?

2. Methodology

2.1. Experiment Facility and Setups

The controlled environment chamber in UC Berkeley’s Center for Built Environment was used to conduct experiments. This 5.48 m × 5.44 m × 2.5 m experiment facility, with a door of 1.98 m × 0.98 m at one corner, had capabilities to supply air at altering rates from floor mounted grilles, overhead supply diffusers, or wall-mounted grilles. For the experiments, the air was supplied through the 0.3 m × 0.3 m louvre type grille mounted on the wall at a height of 0.3 m from the ceiling (Figure 1), with the capability to distribute air in a single adjustable direction with the exfiltration arrangement through the gap around the door, which created positive pressure inside the chamber compared to the outside corridor when supply fans were on.

The fan and the air handling unit (AHU) for the mechanical ventilation system was outside of the experiment chamber, while a supply air duct connected the AHU and the supply grille. The experiment team only had control over the fan drive connected to the AHU.

A series of experiments was conducted to analyze the effects of swing door motion on steady-state indoor airflow. Due to limited availability, eight sensors were used instead of the planned 32, which demanded data collection in sets. Figure 1b shows the arrangement of the sensors along four rows (L1 to L4). Row L1 was parallel to the closed door with a distance of 0.1 m from the wall, whereas row L4 was perpendicular to the closed door, with row L2 and L3 at angles of 30 degrees and 60 degrees with row L1. Each row consisted of four distinct locations where tripod-mounted sensors were placed at locations designated as 1 through 4 at two separate elevations. The first set of sensing systems was closest to the door tip, being 0.1 m away from it. The following sets of sensors were placed at 1 m subsequent distances. At every location, one sensor was at the lower level, at 0.66 m above the ground, covering 1/3rd of the door height, whereas another sensor was at 0.66 m height from the lower-level sensor elevation, covering a total 2/3rd of the door height in Figure 1c. For ease of nomenclature, the sensing stations are designated according to their locations. For example, at row L1, the lower-level sensor at the first position is identified as LL11 (the first ‘L’ is to indicate the lower elevation), and at row L3, the sensor at the higher elevation at position 2 is called UL32 (the letter ‘U’ to indicate upper elevation).

This setup facilitated obtaining three-dimensional airflow fields generated by door movement, with sensor positioning providing airflow data at two altitudes. Despite the limited sensor count, the tiered arrangement allowed for a detailed spatial velocity profile, capturing the wake evolution and its effects on indoor airflow patterns.

2.2. Initial Conditions and Door-Opening Events

The experiments were conducted under six different setups, combining three initial conditions and two-door opening schemes, as summarized in

Table 1. Summary of Experimental Conditions and Door Opening Scenarios.

Test Number	Supply Fan Operation	Airflow Rate (m3/s)	Differential Pressure (Pa)	Door Opening Scheme	Total Cycle Time (s)	Description
Test 1	Off	0	0	Door Opening and Closing Once	~5.38 (σ = 0.21)	The door opens in 2 s, remains open for 1 s, and closes in 2 s. Represents typical entry or exit.
Test 2	70% Capacity	0.063	10	Door Opening and Closing Once	~5.52 (σ = 0.82)	Same as Test 1 but with a medium airflow supply.
Test 3	100% Fan	0.09	20	Door Opening and Closing Once	~5.42 (σ = 0.39)	Same as Test 1 but with a high airflow supply.
Test 4	Off	0	0	Door Opening and Closing Twice	~12.33 (σ = 1.14)	Two cycles of door operation with a 2 s pause between them. Represents entering to deliver or retrieve items and getting out.
Test 5	70% Capacity	0.063	10	Door Opening and Closing Twice	~12.49 (σ = 0.24)	Same as Test 4 but with a medium airflow supply.
Test 6	100% Fan	0.09	20	Door Opening and Closing Twice	~12.48 (σ = 0.19)	Same as Test 4 but with a high airflow supply

.

The experiment facility was equipped with a drive controlling the fan speed, which was accessible to the experiment team. Due to a no. of reasons, including the limited availability of the test facility and the minimum allowable operable fan speed being at least 60% of the maximum fan speed, three different flow regimes were defined, as described below:

• Still air—The initial steady-state condition inside the experiment chamber was quiescent as the supply fan and the Air Handling Unit (AHU) were not operating, and the supply diffuser was kept off.
• 70% of fan—This condition corresponded to a moderate airflow at around 0.063 m³/s (63 L/s). After obtaining steady-state conditions, the differential pressure between the chamber and the outside corridor was 10 Pa, with the chamber being on the positive side.
• 100% fan—With the fan and AHU operating at full capacity, this initial condition was set for supplying air at 0.09 m³/s and generated a 20 Pa pressure differential with respect to the corridor outside of the test chamber.

Two different door operations were used for conducting the experiments, which emulated common door opening practices in indoor built environments.

• Door opening and closing once—The door opens in 2 s, remains open for 1 s, and closes in another 2 s, simulating a typical entry or exit. Each cycle lasted approximately 5 s.
• Door opening and closing twice—The door followed the same initial cycle, then remained closed for 2 s before opening and closing again, representing scenarios like item delivery or retrieval. This scheme lasted around 12 s per cycle.

The flow rates and the pressure differential studied in this paper represents typical office buildings, classrooms, or general patient rooms in healthcare facilities [26]. A person manually operated the door for all tests, standing across the hallway and using a handheld device to minimize any impact from body movement. Timekeeping was also manual, with door operation cycles lasting slightly longer than 5 s for single openings and 12 s for double openings. Table 1 also presents the average door opening-closing cycle time for each of the six experiment arrangements with standard deviation noted in parentheses. Each experiment was repeated 30 times, with velocity data recorded for 60 s per cycle, allowing a substantial number of observations to confirm statistical consistency.

These experiments aimed to examine the effects of swing door movements on airflow fields in a positively pressurized chamber, specifically isolating the influence of door movement without additional occupant interference. This study’s objective was to capture and analyze variations in airflow patterns to support predictive modeling under various initial conditions. While the broader project aimed to model these transient flow variations from door openings, this manuscript focuses exclusively on the experimental results and the event-based predictive model.

2.3. Sensing Instruments

During the experiments, omnidirectional anemometer air speed sensing systems (AirDistSys5000 by Sensor Electronic, Gliwice, Poland) were used to measure airflow velocity. Each anemometer featured a 2 mm spherical probe with hotwire sensors (SensoAnemo5100LSF) capable of measuring speeds between 0.05 and 5 m/s with an accuracy of 0.02 m/s or 1.5% and a directional sensitivity error of 2.5% for velocities over 2 m/s. Additionally, these sensors could measure temperature in the range of −10 to 50 °C with an accuracy of 0.2 °C. The data logging frequency was set to one reading every two seconds, ensuring high sensitivity and accuracy for low-speed measurements typical of indoor environments. The SensoDACon 5400 converter, a part of the AirDistSys5000, transformed digital signals into analog velocity outputs recorded via a wireless SensoBee system. Figure 2 presents a schematic of this setup, and further technical details are available in the AirDistSys5000 user manual. The sensors were calibrated per manufacturer specifications aligned with ISO 7726 [27], ISO 7730 [28], and ASHRAE 55 standards [29]. Their reliability in indoor airflow measurements is supported by prior studies [30,31,32,33,34].

This study focused on examining airflow patterns resulting from swing door movements within a positively pressurized chamber. Since temperature variations can affect buoyancy-driven airflow, temperature was monitored to control for any unwanted buoyancy effects. Recorded data showed minimal temperature fluctuation (mean temperature of 23.7 °C, standard deviation 0.025), allowing for the assumption of an isothermal condition. Consequently, any velocity changes due to temperature variation were considered negligible and excluded from the analysis.

2.4. Statistical Consistency

The time-averaged outputs of the unidirectional sensors for each test case were collected for 60 s, which generated 30 data points for each of the 30 repetitions. For all the sensing systems, those 30 data points for each test were averaged over the repetitions to obtain a transient velocity profile. The data indicated fair consistency at each point in time, for all the measuring units.

To assess the consistency of measurements, all the spatial and temporal data points for each test setting were combined in one array ($\mathit{V}$). The Relative Standard Errors (RSE) were defined as the data standard error (SE) of $V$ divided by its average (${\displaystyle \frac{SE\left(\mathit{V}\right)}{\overline{V}}}$). Thus, the index RSE was normalized by the average velocity magnitude. Test-wise RSEs are shown in

Table 2. Data consistency.

Experiment	Average RSE
Test 1	11.56%
Test 2	10.09%
Test 3	7.67%
Test 4	11.43%
Test 5	10.23%
Test 6	8.19%

. RSE was the largest for quiescent air, perhaps due to the low average value of data points.

2.5. Kinetic Energy Calculation

The movement of the door imparted kinetic energy to the moving air, which is proportional to the square of the velocity (Equation (1)). Since we measured the velocity magnitudes using omnidirectional sensors, we could calculate the magnitude of kinetic energy using only the velocity magnitudes, as demonstrated in Equations (1)–(3). It is important to note that for these calculations, the mass of the air element was considered to be unity. Additionally, the time-averaged kinetic energy during door opening ($K_{do}$) and background kinetic energy ($K_{bg}$) are defined as follows:

$$K=\int {\displaystyle \frac{1}{2}}{v}^{2}dt$$

(1)

$$K_{do}={\displaystyle \frac{{\int }_{t=3\mathrm{s}}^{23\mathrm{s}}{\frac{1}{2}v}^{2}dt}{(23-3)\mathrm{s}}}$$

(2)

$$K_{bg}={\displaystyle \frac{{\int }_{t=40\mathrm{s}}^{60}{\frac{1}{2}v}^{2}dt}{(60-40)\mathrm{s}}}$$

(3)

The sensors employed for the experiments logged data every two seconds, and hence, the first data point after the start of door movement was at second 3. As observed from all the data collected through all six experiments, the most prominent impacts of the door openings on the resultant velocity fields were noticed until the second 23 before the velocity values started to decline. Thus, the values between second 3 and second 23 were used to calculate the kinetic energy associated with the door opening. As negligible changes in the air velocity were recorded by any of the sensing stations after second 40, the duration from 40 s to 60 s indicated the background kinetic energy (i.e., the kinetic energy of the air without the impacts of the door movement), which in turn helps to demonstrate the change in kinetic energy due to the door movement. Bhattacharya et al. (2021) demonstrated that air velocity magnitudes below 0.15 m/s can be considered as quiescent [35]. By second 40, for both door opening once and twice, the velocity magnitude went ≤0.15 m/s when there was no air supply in the chamber. Alternatively, when air was being supplied to the chamber, by second 40, the velocity magnitudes dropped to the level commensurate with the air speed before the door operation. Hence, second 40–60 were used to calculate the background kinetic energy.

2.6. Predictive Methodology Using Event-Based Model

As discussed earlier, the difficulty associated with modeling the intermittent transient events has been an unaddressed issue. The novel methodology of the Event-Based Model (EBM), developed by Mousavi and Bhattacharya (2022), takes an innovative approach to using the data collected for predicting the resultant flow fields from similar events under different initial conditions [25]. The fundamental concept behind the EBM is that it creates discrete events from the interactions between a subject and its environment. For instance, consider a person walking towards and exiting through a door in a room. Factors affecting airflow in this scenario include steady and transient events. The steady events, such as constant air circulation from the ventilation system, remain consistent, while transient events, such as human movement, change over time and involve higher computational costs if one wants to simulate the transient change in the flow properties numerically. Bhattacharya et al. (2021) [35], using experimental studies by measuring airflow around a moving body, showed that the change in flow velocity can reach a distance of 2 m from the walking human in the direction normal to the moving direction when the speed of the walk is at 1 m/s. Simulating airflow during movement, such as human walking or door operation, can be computationally expensive, with researchers reporting that simulating a 14 s door operation took approximately 168 h [5]. Instead of conducting experiments or developing CFD models to compute the transient changes, a steady-state model can be used to describe airflow before the onset of movement, which is easier to model, and then a 20 s walking model can be superposed on the steady-state model to understand airflow dynamics under the transient event. Thus, EBM offers a more efficient solution by modeling dynamic events independently and storing these results for future use, significantly reducing computational time. EBM leverages precalculated results and integrates them into the current flow field, making repeated simulations more efficient.

Let us consider the flow properties from a transient event to be denoted as $\overrightarrow{U_0}(\overrightarrow{x},t)$, where x represents the spatial variable and t represents the temporal variable. For simplicity, if $\overrightarrow{U_0}(\overrightarrow{x},t)$ can be denoted as $u^i$. In fluid dynamic simulations, the flow properties are Markovian, essentially making it possible to estimate the properties at the next time step from the values at the current time step. In this construct, an identifier can be defined to linearly approximate $u^{i+1}$ from $u^i$.

$${\left(u^{i+1}\right)}_{1\times n}={\left(u^i\right)}_{1\times n}\times {\alpha }_{n\times n}^i$$

(4)

$${\alpha }^i={\left(u^i\right)}^{p_{-1}}\times u^{i+1}$$

(5)

Here, the subscript is denoting the size of the vectors in a linear system. In Equation (5), the $p_{-1}$ is the pseudo-inverse operator. With this approach, a time-variable $\alpha \left(\in R^{n\times n\times m-1}\right)$ can be found for each known case that would make possible the prediction at each timestep based on the previous timestep. So, using this prediction methodology, if there are known cases of the transient events under different initial steady state conditions, one can use data-based prediction to estimate the resultant flow field from the transient event using the previously collected data from the transient events. Let us assume that a total of K previous cases with known solutions are available, i.e., ${\mathit{X}}_{\mathbf{1}}to{\mathit{X}}_K$. Hence, $\alpha$ can be calculated and stored for each case. To predict a new case (i.e., ${\mathit{X}}_{\mathit{n}\mathit{e}\mathit{w}}$), one only needs to approximate the ${\alpha }_{new}$ and use that to predict velocity magnitudes beyond the initial condition. Here, the goal is to apply EBM to predict door-opening events based on experimental data. For instance, in the case of door operation, EBM assumes each door opening follows the same conditions and allows the velocity field for new door events to be approximated using previous data. Thus, Algorithm 1 provides a methodology to dynamically calculate $\alpha$ at every time step by comparing the estimated velocity field with the closest velocity field in the existing database. This approach utilizes simplified Navier–Stokes equations, enabling efficient reuse of existing solutions. For more details on the EBM methodology and derivation, please refer to [25].

It should be noted that this study employs the previously established EBM framework within controlled boundary conditions to demonstrate its applicability to door-opening events [6]. Therefore, an event dataset was constructed from the measured door opening dataset under various ventilation states. The velocity fields were predicted with low mean relative error, as reported in the Results.

Algorithm 1 EBM Approximation.

Let < $X_1,X_2,X_3,\dots ,X_k$> be the known cases

Let$V_i$be the velocity vectors at all the nodes of the entire domain at timestep I for each of the known cases, i.e., the IC

$fori=1:Tdo$ ($\%T=numberofthetimesteps\%$)

$forj=1:30do$ (% j is the no. of data points per experiment %)

$fork=1:Kdo$ (% k is the no. of known cases %)

$d=norm({IC}_{X{1}_i}-{IC}_{X{K}_j};)$

$endfor$

$Q=Index\left(min\right(d\left)\right)(\%Determinetheclosestcaseateverytimestep\%)$

$endfor$

${\alpha }_{new}$ = αQ

$u_{new}^i$ = $u_{new}^{i-1}$ × ${\alpha }_{new}$

$endfor$

3. Results

The velocity magnitude data obtained for 60 s at every location during different experiment setups were analyzed to understand the alterations in indoor airflow characteristics influenced by the door movement. The following sections describe the spatiotemporal changes in the velocity profile within the test chamber under different experimental conditions and show the potential of the EBM method to predict the velocity fields using a data-driven approach.

3.1. Door Opening Once—Still Air

When the door was opened once in still air, all sensors recorded delayed velocity increases after the airflow passed each sensor (Table 1). Sensors closer to the door (L11, L21, L31, and L41) detected increases more rapidly than those farther away, with clear spatial differences in velocity readings.

Sensing Systems at Lower Elevation

The temporal characteristics of flow velocity, as captured by sensors at the lower level, displayed location-specific trends according to their arrangement (Figure 1). As can be seen from Figure 3, sensors closest to the door in rows L1 and L2 (LL11 and LL12) detected velocity increases sooner, with LL11 registering a rise at second 3, LL12 at second 5, and sensors located farther from the door (LL13 and LL14) around second 11. Both LL11 and LL12 recorded peak velocities of 0.51 m/s at seconds 15 and 19, respectively, with sharper drops in velocity afterward. In contrast, LL13 and LL14, positioned farther away, exhibited smoother declines, maintaining velocities above 0.1 m/s until nearly 50 s, while LL11 and LL12 dropped to near zero by 40 s. These observations indicate a progressive spread of velocity fields over time.

At row L2, the peak velocity recorded by LL22 (0.29 m/s) was around 45% lower than LL21’s, with similar patterns at LL23 and LL24 compared to LL13 and LL14. In row L3, the nearest sensors, LL31 and LL32, matched the peak velocities of rows L1 and L2 but showed a quicker peak at LL32. LL33 and LL34 showed minimal change, except for a minor increase at 23 s. In row L4, all sensors logged velocity changes almost simultaneously, contrasting with the staggered timing of previous rows. The peak velocities at LL41 and LL42 were slightly lower but occurred at 15 s, about 4 s earlier than LL12. Similarly, LL43 and LL44 peaked 5–6 s sooner than the corresponding sensors in previous rows.

With the initial condition of quiescent air inside the chamber, Figure 4 displays velocity changes for higher elevation sensors in rows L1 and L2. Sensors closest to the door (UL11, UL21, etc.) recorded immediate velocity increases, recording a peak velocity magnitude of around 0.49 m/s. In contrast, sensors in positions 2, 3, and 4 showed delayed responses, with time lags correlated to their distance from the door. Only sensors at positions 1 and 2 in each row showed substantial speed changes, while sensors at positions 3 and 4 registered above 0.1 m/s only in row L1. Notably, row L4 sensors logged peak velocities more quickly than those in row L1.

Table 3. The maximum velocity associated with the sensing stations and the time it occurred door opening once.

Sensors at Low Elevation					Sensors at High Elevation
	Range of		Maximum Velocity		Range of		Maximum Velocity
Sensor	Non-Zero Entries	Lag (s)	[Time It Occurred]	Sensor	Non-Zero Entries	Lag (s)	[Time It Occurred]
ID	(s)		(s)	ID	(s)		(s)
LL11	26	2	0.51 [15]	UL11	22	2	0.49 [13]
LL12	34	6	0.51 [19]	UL12	34	6	0.25 [19]
LL13	35	8	0.23 [14]	UL13	20	10	0.11 [14]
LL14	40	10	0.25 [22]	UL14	23	12	0.13 [22]
LL21	26	<1	0.53 [13]	UL21	26	<1	0.49 [13]
LL22	28	6	0.29 [19]	UL22	33	2	0.17 [19]
LL23	26	8	0.24 [19]	UL23	20	6	0.07 [19]
LL24	30	10	0.25 [24]	UL24	20	10	0.06 [14]
LL31	24	<1	0.5 [13]	UL31	24	<1	0.48 [13]
LL32	26	2	0.29 [15]	UL32	22	2	0.13 [15]
LL33	28	2	0.12 [19]	UL33	24	2	0.05 [15]
LL34	19	11	0.087 [22]	UL34	20	6	0.04 [17]
LL41	22	2	0.49 [13]	UL41	26	<1	0.48 [13]
LL42	30	2	0.31 [15]	UL42	26	2	0.17 [13]
LL43	24	4	0.14 [15]	UL43	24	4	0.11 [15]
LL44	26	6	0.12 [17]	UL44	22	6	0.08 [17]

indicates that lower elevation sensors consistently recorded higher peak velocities (0.087–0.53 m/s) compared to higher elevation sensors (0.04–0.49 m/s). Sensors closest to the door at both elevations (LL11 and UL11) detected peaks at 0.51 m/s and 0.49 m/s, respectively. Sensors at higher elevations, farther from the door, showed lower velocities, likely due to decreased airflow interactions. This trend underscores the impact of sensor height in capturing airflow dynamics from door movements and sheds light on the velocity distribution at the vertical plane.

3.2. Opening Twice—Still Air

The observed temporal trends for air movement characteristics corresponding to opening the door twice with still air was comparable to that of opening the door once with quiescent conditions inside the test chamber, albeit door opening twice resulted in higher velocity magnitudes and a greater area under the curve of velocity plotted against time. It was found that no drop in velocity magnitude was recorded at any position due to the 2 s recess time between the two consecutive door opening activities. Even though both sets of door operations were completed close to second 12 and presumably the wakes carried by moving air hovered past the sensors sometime close to that, a lag of at least 5 s was observed. Assuming the highest velocity was obtained from consecutive door openings, the earliest recorded maximum velocity magnitude was at the first sensors located the closest to the door at second 17.

3.2.1. Sensing Systems at Lower Elevation

As with the single opening, the closest lower elevation sensors (LL11 in L1 and LL12 in L2) recorded the highest velocities, including 0.79 m/s for L1 and 0.76 m/s for L2, L3, and L4. The second sensors (LL12, LL22, etc.) also detected significant airflow increases. In rows L3 and L4, sensors at positions 3 and 4 showed minimal changes, indicating limited airflow impact further from the door. Notably, in rows L1 and L2, the nearest sensors recorded peak velocities sooner than those farther away, establishing a pattern where velocity peaks propagate outward as a function of distance, suggesting the temporal spread of airflow fields through the chamber (Figure 5).

3.2.2. Sensing Systems at Higher Elevation

Similar to the single door opening (test 1), data from test 4 (door opening twice–no air supply) showed consistent velocity patterns across both elevations, though with lower magnitudes at higher elevations. As with lower sensors, those closest to the door recorded velocity increases with shorter lags, which increased with distance from the door (Figure 6).

Table 4. Change in velocity proportion from opening once to twice at all sensor locations.

Sensor	Door Opening	Door Opening	Velocity	Sensor	Door Opening	Door Opening	Velocity
ID	Once	Twice	Proportions	ID	Once	Twice	Proportions
Column	A	B	B/A	Column	A	B	B/A
LL11	0.1504	0.1962	1.3	UL11	0.1431	0.268	1.87
LL12	0.1824	0.1821	1	UL12	0.1318	0.1518	1.15
LL13	0.1258	0.1342	1.07	UL13	0.0739	0.1012	1.37
LL14	0.1426	0.1629	1.14	UL14	0.0733	0.0937	1.27
LL21	0.1597	0.1911	1.2	UL21	0.1547	0.2731	1.77
LL22	0.1029	0.1103	1.07	UL22	0.0766	0.1125	1.47
LL23	0.0932	0.1122	1.2	UL23	0.0504	0.0722	1.43
LL24	0.1063	0.1088	1.02	UL24	0.0476	0.0679	1.42
LL31	0.1556	0.2033	1.31	UL31	0.1645	0.2808	1.71
LL32	0.116	0.1401	1.21	UL32	0.0634	0.0929	1.47
LL33	0.0683	0.0715	1.05	UL33	0.0282	0.0404	1.43
LL34	0.0694	0.0789	1.14	UL34	0.0234	0.0344	1.53
LL41	0.1495	0.2218	1.48	UL41	0.1563	0.2971	1.9
LL42	0.112	0.1485	1.33	UL42	0.0753	0.1242	1.65
LL43	0.0745	0.0758	1.02	UL43	0.0695	0.0832	1.2
LL44	0.0613	0.0546	0.89	UL44	0.0395	0.0397	1.01

compares the time-averaged velocity values ($\overline{v}$) at each location from door opening once and twice.

Since, for most of the experiment scenarios, after 40 s, the velocity magnitude values returned to the initial values, the time-averaged calculation was carried out using the following equation:

$$\overline{v}={\displaystyle \frac{{\sum }_{t=1\mathrm{s}}^{40}v}{N}}$$

(6)

Here, $v$ is the velocity magnitude, recorded at each timestep (2 s), and $N$ is the total no. of data points from second 1 to 40.

Table 4 shows that consecutive door openings significantly raised peak velocities, especially near the door. Lower elevation sensors like LL11 and LL21 had velocity ratios of 1.3 and 1.2, while higher elevation sensors such as UL11 and UL21 displayed larger increases, at 1.87 and 1.77 times, indicating greater sensitivity to repeated openings. Notably, in some cases (e.g., LL23 and LL34), lower elevation sensors farther from the door showed similar velocity ratios to those closer. This anomaly likely reflects increased airflow interaction along the floor, which dissipates with distance.

3.3. Effect of Initial Conditions

Velocity magnitudes from both door-opening scenarios (single and double) were compared across initial conditions: still air, 70% fan, and 100% fan. The results indicated similar spatial and temporal trends among sensor rows and positions across all conditions, though with notable variations in airflow response based on fan operation levels (Figure 7).

Figure 7 shows an example from lower elevation sensors in row L2 for both door-opening scenarios at 70% and 100% fan capacities. Two key differences emerged between the still air and fan conditions. First, the increased airflow supply reduced the intensity of door-induced disturbances due to a higher-pressure differential, where the high velocity of ventilated airflow quickly pushed the wakes from the door opening. This effect is evident in the velocity dip around 17 s (for door-opening twice), corresponding to the 2 s pause after the initial door movement. Additionally, with a 100% air supply, sensors farther from the door (positions 3 and 4) showed minimal response under still air or 70% fan and captured notable velocity changes. This increase in response can be attributed to the increased interaction between the primary flow from the ventilation system and the secondary flow originating from the door opening, which created strong turbulence and high-velocity wakes that propagated further into the room.

3.4. Spatial Distribution

For both the door opening once and twice with the quiescent air initial condition (Test 1 and Test 4), the distribution of velocity fields is depicted in Figure 8 and Figure 9. For the door opening twice, the isolines were associated with higher magnitudes of velocity, and isolines with higher speeds were found deeper inside the test chamber when compared to the door opening once. It is worth noting that the highest velocity flow fields were located near the door movement zone quickly after the door movement was stopped, and with time, as these wakes expanded deeper into the room, velocity values dropped gradually.

For the experiments involving the door opening once with no fan until the second 11, the areas with significant velocity values (≃0.3 m/s) were found to be within 1.5 m from the door. By second 19, the velocity values of the range ≃0.5 m/s were found further into the room. At second 21, isolines of ≃0.3 m/s could be found in the middle of the room, which started to drop down from second 29. But it is interesting to observe that even though the velocity values have decreased to ≤0.1 m/s, the highest velocity magnitudes correspond to a location in the middle of the room, suggesting that in the absence of ventilation, the wakes become stagnant. For door opening twice, the flow fields containing higher velocity magnitude, in the range of ≃0.7 m/s, are found to be in close proximity to the door movement region until late, compared to opening once, as seen from the surface plots at time 19 s (Figure 8 and Figure 9), due to additional motion generated by opening the door a second time. By second 33, wakes with ranges ≤0.15 m/s were observable at a distance of ≃2.5 m from the door.

The presence of wakes with higher magnitudes deeper into the experiment chamber was due to the interaction between already moving air with the wakes generated from opening and closing the door for a second time. Importantly, the surface plots were obtained by interpolating the data in a 0.1 × 0.1 grid throughout the entire chamber, considering zero velocity values at the boundaries. It may also be noted that these surface plots were approximated using the data from the lower elevation sensors only.

A Radial Basis Function (RBF) interpolation method was employed to ensure a smooth and continuous representation of the flow field between discrete sensor measurements. This technique has been widely validated in the context of scattered data approximation [36,37]. It is also noteworthy that these surface plots were derived solely from data collected by the lower elevation sensors.

3.5. Kinetic Energy

The kinetic energy at the onset of the door opening initiation was comparable for both the test scenarios, but admittedly, door opening twice was associated with a higher kinetic energy transfer involved for a longer duration (Figure 10). The ratio of kinetic energy between the two door opening scenarios was 2 for the three different initial conditions, as shown in

Table 5. Time-averaged Kinetic Energy for Two Door Opening Schemes.

Initial Condition	Still Air		70% Fan		100% Fan
Door Opening	Opening	Opening	Opening	Opening	Opening	Opening
Scheme	Once	Twice	Once	Twice	Once	Twice
$K_{do}$	0.0242	0.0516	0.0249	0.0483	0.0411	0.0682
$K_{bg}$	0.0012	0.0022	0.0042	0.0052	0.0129	0.0145
$∆K=(K_{do}-K_{bg})$	0.023	0.0494	0.0206	0.0431	0.0281	0.0537
$∆K\left(Twice\right)/∆K\left(Once\right)$	2.15		2.09		1.91

. It can be noted that for zero initial condition (i.e., no fan operating) and 70% air, the kinetic energy for door opening is >2 times that of the steady-state condition without any interference in terms of door operation. But with the maximum air supply (i.e., the 100% fan), a part of the energy is spent to overcome the heavy resistance provided by the air mass inside the test chamber when opening the door, taking the proportion below 2.

3.6. Prediction of Velocity Fields Using EBM

To evaluate the accuracy of the EBM model, 2D velocity fields from both experimentally obtained data and EBM-approximated data were visualized using filled contour plots. Figure 11 illustrates the spatial velocity fields derived from the radially arranged sensor locations, assuming zero slip conditions at all boundaries, such as walls. Since velocity values were only available at sensor locations, they were interpolated throughout the testing facility. The test chamber area was discretized into grids of 110 × 109, with each grid measuring 5 cm × 5 cm. These visualizations represent the velocity fields at two specific instances, namely timesteps 4 and 8. Approximately 18% of the grid area, mainly near corners and upper boundaries, was extrapolated due to sensor spacing. These extrapolated regions may increase local deviations, but did not change the mean error (<5%) or the spatial patterns reported. The comparison demonstrates that the EBM can accurately approximate the velocity fields, with some deviations observed near the walls and around the door movement area.

To evaluate the accuracy of the estimated velocity using Algorithm 1, the predicted velocity field ($u_{new}$) was compared to a test case ($u_{test}$). The absolute error (e) was defined as the difference between the predicted and measured velocities at each of the n locations and m timesteps (Equation (7)):

$${e=|u}_{new}-u_{test}|ϵR_{n\times m}$$

(7)

To make the error dimensionless, the Relative Error (ε) was calculated by dividing the absolute error by the measured velocity matrix ($u_{test}$) which is presented in Equation (8). Mathematically, this is defined as follows:

$$\mathsf{\epsilon }={\displaystyle \frac{e}{u_{test}}}ϵR_{n\times m}$$

(8)

Although the detailed analysis of e and ϵ provided granular insights, in some parts of this paper, we presented mean spatial (row average) or temporal (column average) errors. Additionally, all individual error data (n × m) were visualized using statistical histograms. In this context, we used the algebraic value of ${|u}_{new}$−$u_{test}|$ instead of the absolute value.

Figure 12 illustrates the time-based average absolute error between the EBM predictions and the measured data for a double door-opening scenario under 70% fan operation. Throughout the 30 timesteps, the average of the error consistently remains below 5% of the measured velocity, reflecting an acceptable level of agreement between model and experiment. The error is low at the start, then rises to a peak of around 0.07 m/s in the midpoint when door-induced turbulence is most pronounced. It subsequently declines and stabilizes at approximately 0.02 m/s, indicating the model’s ability to capture transient flow behaviors more accurately once the flow begins to settle. These results underscore the EBM’s capability to approximate complex, overlapping airflow events efficiently. Overall, the model offers a robust predictive tool with minimal computational overhead, suitable for practical indoor airflow analyses.

Figure 13 shows the signed error distribution for the EBM model, along with a normal fit. The distribution is centered near zero, indicating that the model does not consistently overestimate or underestimate the measured data. The narrow spread around the center suggests a good match between the predictions and the actual values. Positive and negative errors occur with similar frequency, implying that deviations are random rather than systematic. The close alignment of the normal curve with the observed data further supports the model’s accuracy in capturing the door-induced airflow patterns.

4. Discussion

Since door movement was the only physical change in the test chamber, all observed velocity variations can be attributed directly to the door’s operation. Figure 3, Figure 4, Figure 5 and Figure 6 show that sensors closer to the door consistently recorded higher air velocities than those farther away. As the door opened, it imparted momentum to surrounding air, generating wakes primarily from the leading edge. These wakes propagated through the chamber, with weaker disturbances trailing. Leaving the door ajar for one second allowed for partial wake dissipation; however, the subsequent closing motion added momentum in the opposite direction, creating turbulent vortices and increasing overall air velocity. This pattern aligns with fluid dynamics principles, where moving boundaries induce disturbances in nearby air.

The sensors’ moderate resolution (one data point every two seconds) limited temporal detail, but Figure 3, Figure 4 and Figure 5 reveal location-specific velocity fluctuations reflecting cumulative door movement effects. Sensors along line L1 (parallel to the door) recorded higher velocities, likely due to stationary boundaries, such as walls, which facilitate wake propagation along their surfaces. This interaction produces higher velocity gradients near walls, supporting Eames et al. (2009), who found that stationary boundaries significantly affect airflow patterns by directing wake movement [37].

The timing and magnitude of peak velocities were influenced by the sensors’ distance from the door, particularly in row L4 (perpendicular to the closed door), where airflow disturbances spread more uniformly, causing simultaneous peak velocities across all sensors. After the door opened, the air moved along row L4, but the closing motion reversed the direction, drawing air back toward the sensors. This bidirectional movement created synchronized peaks across sensors, emphasizing the impact of door closing on airflow in perpendicular orientations.

The 2 s sampling interval was appropriate for low-speed indoor flows and produced stable, reproducible patterns across 30 repetitions. However, very short peaks in the first 1–2 s after the door starts moving may be under-resolved; so, the recorded series may underestimate the instantaneous maxima in that window. As a result, the time series reported here captures the dominant trends and integrated intensity but may underestimate the instantaneous maxima within the first few seconds. Moreover, in this study, the spatiotemporal impact of the transient event was studied and modeled, and as such, instantaneous maxima were not as crucial for the purpose of this study. Future studies will incorporate automated door actuation with angle tracking and higher-frequency measurements to resolve these rapid fluctuations and further improve EBM initialization.

Lower elevation sensors recorded consistently higher velocities than those at higher elevations due to increased interaction with the floor, a stationary boundary that enforces a zero-slip condition, which restricts airflow dissipation and maintains higher velocity gradients near the surface. The presence of particles on the floor may also contribute to increased airflow interactions at lower levels. In contrast, higher elevation sensors located away from these boundaries experienced weaker disturbances. These observations support previous findings that surfaces influence airflow fields by directing and sustaining airflow along their planes [37,38,39].

Opening the door twice resulted in higher velocity magnitudes than a single cycle. This effect is due to residual disturbances from the first door opening, which served as the initial condition for the second, amplifying turbulence and forming more intense vortices. Figure 5 and Figure 6 show that double door operations intensified airflow disturbances across multiple sensors, illustrating the principle of momentum superposition in fluid dynamics, where successive disturbances compound airflow patterns. The addition of the Event-Based Modeling (EBM) approach provided an effective predictive tool for assessing airflow due to door movements. EBM approximates airflow by treating each door movement as a discrete event, leveraging experimental data to estimate velocity fields for subsequent movements. This method efficiently models flow changes without requiring complex recalculations. Using precomputed data, EBM can generate real-time airflow predictions by superimposing prior results, enabling practical assessments in real-world settings. EBM’s computational efficiency makes it suitable for environments with frequent door movements, such as healthcare facilities, where maintaining controlled airflow is essential for hygiene and safety.

This study also examined how varying ventilation rates (70% and 100% air) impact door-induced airflow disturbances. Higher ventilation rates dampened door-induced disturbances more rapidly due to increased pressure differentials. Under 100% air conditions, sensors farther from the door detected stronger disturbances, as the pressure differential amplified turbulence, allowing high-velocity wakes to reach further sensors (Figure 7). This observation is consistent with findings from previous studies on airflow under varying ventilation conditions.

Lastly, surface plots in Figure 8 and Figure 9 illustrate the temporal distribution of air speed for single and double door openings. Initially, wakes concentrated around the door, but over time, and without additional disturbances, airflow fields spread and dissipated. In double door openings, residual momentum from the first cycle enhanced the second, sustaining higher velocities. Near walls, airflow experienced slower momentum transfer due to fewer surrounding air molecules, leading to abrupt changes in velocity along boundaries. This study confirms that door movements significantly impact airflow patterns in enclosed spaces, with EBM offering an efficient predictive model for managing these effects in high-occupancy or controlled environments, where precise airflow control is crucial.

5. Limitations

This study analyzed the effects of swing door movements on a positively pressurized chamber with three air supply rates. Several constraints shaped these experiments. Limited access to the chamber restricted testing to three initial conditions with only two defined door-opening scenarios. Due to the limited number of sensors, data collection occurred at sparse locations spaced one meter apart, intended to maximize the data coverage area around the door periphery. The sensor placement was designed intuitively without a statistical design of experiments due to time constraints. The chamber’s ventilation setup presented additional limitations. The research team had no control over air distribution arrangements, relying on wall-mounted grilles for air supply despite practical limitations. Supply air fluctuations could influence airflow regimes, but these could not be measured. The results were based on a gauge in the AHU, which indicated fairly consistent flow rates, although no system regulated the pressure differential post-door opening. These experiments also focused solely on door movement effects, excluding occupant movement, which has been studied elsewhere for its role in contaminant transport [40,41].

Directional velocity data were not available, as the ultrasound anemometer (Arens et al., 2020 [5]) was used only to validate the sensor calibration for future experiments. The moderate frequency of the sensing system limited its ability to capture all air velocity fluctuations, and discussions on turbulence are based on theoretical assumptions rather than direct measurement. Data collection was capped at 60 s per cycle, with a 40 s cutoff, limiting extended analysis. While these constraints affect generalizability, consistent findings across repeated experiments support this study’s validity [5]. The goal of this study was to examine the general flow behavior during door-induced transients rather than to capture high-frequency turbulence; therefore, the chosen setup represents a pragmatic balance between temporal fidelity, equipment limitations, and experimental repeatability.

6. Conclusions

This study experimentally and computationally investigated door-induced, transient indoor airflow under still air and mechanically ventilated conditions and evaluated a data-driven EBM framework for prediction. Velocity fields were measured in a full-scale chamber and visualized on a 110 × 109 grid (5 cm × 5 cm) using RBF interpolation with no-slip boundaries. Next, EBM reconstructed transient airflow distribution by composing discrete door-opening events over the steady background. Based on the analysis of the experiments and EBM predictions, the following conclusions are drawn:

• Door motion produced location-specific wakes. In this case, wakes appear earlier at near-door sensors, and after a few seconds, they show up at the sensors further away. The farther away the sensors are from the door, the lower the peak magnitudes are recorded. In still air, high-speed isolines (0.3–0.5 m/s) initially stayed within 1.5–2.5 m of the door, then migrated inward and weakened. Also, double openings kept higher-speed contours near the door for longer.
• Consecutive openings amplified velocities across the field typically by about 1.2 to 1.9 times depending on location and extended disturbance duration. Time-averaged kinetic energy increased by about 2 for double vs. single openings across initial conditions including 2.15, 2.09, 1.91 for still, 70%, 100%.
• Higher supply air partially mitigated peaks by speeding exchange and recovery but could carry wakes deeper so far that sensors (positions 3–4) still responded. Lower elevations consistently recorded higher peaks (near-door ≈0.49–0.51 m/s), and near-wall/floor regions showed stronger turbulence, indicating elevated resuspension risk at boundaries.
• EBM performance. EBM reproduced transient fields with mean relative errors generally <5% over 30 timesteps and no systematic bias in signed-error distributions while substantially reducing computational cost relative to conventional transient simulations by reusing discrete event solutions.

In summary, door-induced wakes are organized, measurable, and predictable within the tested range. The findings support placing sensors near doorways and at lower elevations. They also support choosing ventilation setpoints that limit wake persistence without pushing disturbances deep into occupied zones. Finally, integrating EBM for transient prediction enables responsive airflow management in contamination-sensitive spaces such as healthcare, laboratories, and cleanrooms.