Mechanisms of Particle Deposition in Ventilation Ducts for a Food Factory

Abstract

An improved Eulerian model is proposed to predict particle deposition velocity in a typical mechanical ventilation system of a dairy food factory. For fully developed turbulent flow, the model is modified based on the three-layer model developed by Zhao and Wu (2006a), accounting for thermophoresis as well as turbophoresis, Brownian diffusion, turbulent diffusion, and gravitational settling. Based on in-situ measurements of the aerosol size distributions, particles discussed in this article were in the size range 0.3–20 μm. The measured mass concentration and the predicted particle deposition velocity were used to calculate the deposited particle mass flux in the ventilation duct. The results indicate that the effects of temperature gradients and of surface roughness should be considered in food factories ventilation ducts. For horizontal surfaces, it is shown that even a little difference between empirical and theoretical models can lead to a 2-fold difference in the predicted deposited particle mass flux. Findings of this work may help to identify the specific parameters for cleaning procedures.

1. INTRODUCTION

In recent years, the investigations concerning ventilation in the agro-food industry mostly look at the air distribution but very poor attention was paid to the ventilation duct systems. However, the control of the indoor air quality is directly related to the particle deposition in the ventilation ducts. Deposition onto duct surfaces may affect the microbial quality of the air in forced-ventilation food plants. Depending on the amount of dirt, the material, and the environmental conditions, the HVAC (heating, ventilation, and air-conditioning) system may become an active growth site for microorganisms and a source of biological contamination (Foarde and Menetrez 2002). As mentioned by Zhao and Wu (2006a) a number of problems associated with human health are caused by the pollution of ventilation ducts. Kozak et al. (1996) suggested that the presence of Listeria monocytogenes in finished product must be the result of post-pasteurization contamination from an HVAC system. Byrne et al. (2008) studied the airborne contamination in a pork processing plant and indicated that the contamination could be attributed to people in the area, condensation and mechanically generated aerosol, but they also mentioned that building ventilation factors would also need to be considered before a definite conclusion could be made. According to a study by Siegel and Nazaroff (2003), air turbulence in the duct leading up to a heat exchanger can induce deposition on heat exchanger surfaces. In dairy processing, this may reduce the processing efficiency, since the deposited material disturbs both the fluid flow and the heat transfer, and impair the product quality (Petermeier et al. 2002). Ventilation duct systems usually consist in very large cross section duct with branches, bends, reductions, and diverging sections. All of these elements which cause a flow perturbation and affect the nature of the turbulence are likely to have an influence on the deposition rate (Miguel et al. 2004; Sippola and Nazaroff 2005).

To meet the code of ventilation duct cleanness, the HVAC unit and ductwork should be cleaned on a periodic basis (Zhao and Chen 2006). Thus, the ability to predict the deposited particle mass flux (DMF) within ventilation ducts will help identify the specific parameters for cleaning procedures that are generally effective in reducing microbial contamination.

Recently, Zhao and Wu (2006a) proposed an improved model based on the three-layer model by Lai and Nazaroff (2000), accounting for turbophoresis as well as Brownian diffusion, turbulent diffusion, and gravitational settling. The agreement between improved model results and measured data collected by Sippola and Nazaroff (2004) in uninsulated steel and internally insulated ducts is acceptable for vertical and horizontal surfaces (Zhao and Wu 2006a). However, in food factories, temperature could considerably change between workshops. Zhao et al. (2009) suggested that when the size of the particles gets smaller and smaller, the effect of thermophoresis force becomes stronger while the effect of gravity drops. Besides, the thermophoretic mechanism for larger particles is subjected to the air slipping along the nonuniformly heated particle surface, where the nonuniformity is stipulated by the air temperature gradient and strongly depends on the thermal conductivity of particles (Bakanov 1991, Tsai 1999). Therefore, thermophoretic particle deposition needs to be considered carefully to get reasonable results.

In the frame of the national project CLEANAIRNET, the hygienic design of ventilation ducts in food factories is studied. The main purpose of this article is to modify the three-layer model for particle deposition by taking thermophoresis into account for predicting particle deposition in a typical ventilation duct of a food factory. To this end we have sought also to quantify the effects of various factors that may contribute to the particle deposition flux. After comparison of predicted deposition velocity with empirical model (Sippola and Nazaroff 2003) and measured data collected by Sippola and Nazaroff (2004), our model and the empirical model are applied to predict the DMF based on in-situ measured physical parameters (flow conditions and particle mass concentration).

2. MODEL DESCRIPTION

2.1. Particle Deposition Flux

For fully developed flow, Zhao and Wu (2006a) proposed a three-layer model by considering turbophoresis as well as Brownian diffusion, turbulent diffusion, and gravitational settling. In this Eulerian model, the particle flux within the boundary layer is assumed to be one-dimensional, steady, and constant, i.e., there are no sources of particles. To model the thermophoresis as a mechanism of particle deposition, the particle flux should be modified as follows:

where D _B is the Brownian diffusivity, ξ _p is the turbulent diffusivity which is assumed to be equal to the eddy viscosity of air ξ _a (see Equation (18)), v _ther the thermophoretic velocity, v _g is the particle gravitational settling velocity, v _tp is the turbophoretic velocity, y is the normal distance from the wall, is the time-averaged particle concentration, and i is used to characterize the orientation of the surface, i.e., for a upward facing horizontal surface, i = 1; for a vertical surface, i = 0.

For spherical particles, the Brownian diffusivity D _B can be calculated by the Stokes-Einstein relation, corrected for slip:

Here, T is the fluid absolute temperature, σ _B = 1.38× 10^{− 23} J/K is the Boltzmann constant, d _p is the physical diameter, μ is the dynamic viscosity of the air, and C _c is the Cunningham coefficient caused by slippage (Hinds 1998):

where λ is the air molecular path.

The thermophoretic velocity v _ther is given by the following equation:

where ν is the fluid kinematic viscosity, dT/dy is the temperature gradient, and H _ther is the thermophoretic coefficient which is given by (Talbot et al. 1989):

where k _a and k _p are the thermal conductivities of the air and particle, respectively. Kn is the Knudsen number defined by Kn = 2λ/d _p . The temperature gradient dT/dy can be estimated by the following equation (Nerisson 2009):

where y ⁺ = yu*/ν is the dimensionless distance from the wall to the turbulent core with u* is the fluid friction velocity (Equation (9)), Pr = 0.7 is the Prandtl number of air, σ _t = 1 is the turbulent Schmidt number, κ = 0.41 is the von Karman's constant, C _p is the specific heat capacity of air, and Δ T is the temperature difference between the upper boundary of the calculation domain and the duct wall.

For rough surfaces, according to Zhao and Wu (2006b) the wall boundary condition y ₀ ⁺ is evaluated at:

where k is the mean microscale roughness height of the rough wall and e is the effective roughness height, which can be calculated for different turbulence regime by:

In Equation (7), the fluid friction velocity is calculated by:

where, U _∞ is the average air speed in the axial direction and f is the friction factor, which can be obtained by the following equation presented by White (1986):

where D _h is the duct hydraulic diameter and Re is the Reynolds number for the duct flow.

A balance of the drag force with the gravitational force on a particle leads to a simple expression for the particle gravitational settling velocity:

where g is the gravitational acceleration. The particle relaxation time,τ _p , may be calculated for spherical particles in the Stokes flow regime as follows (Baron and Willeke 2005):

where ρ _p is the particle density and C _d is the drag coefficient, which can be calculated by the following equations (Hinds 1998):

where Re _p is the particle Reynolds number.

The turbophoretic velocity is given as (Caporaloni et al. 1975):

is the particle wall normal fluctuating velocity intensity, which could be approximately estimated according to Guha (1997) to be:

where _ay is the air wall normal fluctuating velocity intensity, and τ _f is the Lagrangian timescale of turbulence. These parameters are parameterized in the boundary layer as follows (Guha 1997):

where ξ _a is the air turbulent viscosity. For rough walls, Zhao and Wu (2006b) have proposed the correlation of Johansen (1991) to calculate this parameter. To avoid discontinuities in that correlation, the equation given by Davies (1966) was used for this study:

2.2. Particle Deposition Velocity

The deposition velocity V _d (m s^{− 1}) is the total particle flux J normalized by the time-averaged particle concentration C _∞ (μ gm^{− 3}) in the center line duct:

The deposition velocity is made dimensionless with the aid of the fluid friction velocity u* :

Thus, the dimensionless form of Equation (1) can be written:

where v _ther ⁺ = v _ther /u*, v _tp ⁺ = v _tp /u*, v _g ⁺ = v _g /u*, and C ⁺ = /C _∞.

To solve the model, Equation (21) can be integrated from the duct wally ₀ ⁺, where C ⁺ = 0, to the turbulent core y ⁺ = 200, where C ⁺ = 1. With these boundary conditions, the solution of Equation (21) is as follows (Zhao and Wu 2006a):

where

In this study, the numerical calculations will be compared with the empirical model developed by Sippola and Nazaroff (2003). For uninsulated galvanized steel ducts, the empirical equations for predicting deposition rates to floor and wall surfaces are given as:

Here, k ₃ = 0.13, k ₂ = 1.1× 10^{− 3}, k ₄ = 1.13, k ₁ = f/2, Sc = ν νD _B is the Schmidt number, τ _p ⁺ = τ _p u* ² /u* ² ν is the dimensionless particle relaxation time and g ⁺ = gν/u* ³ is the dimensionless gravitat3ional acceleration.

3. RESULTS AND DISCUSSION

3.1. Case Study Description

To study the influence of the different mechanisms on particle deposition in a typical food factory ventilation system, the above model requires several input parameters, such as flow conditions, particle mass concentration and characteristics of the ventilation duct. These parameters were obtained by one partner (Cemagref) in the project CLEANAIRNET from in-situ measurements. Part of these data is still confidential, but some useful information is available. Figure 1 shows a schematic diagram of this industrial ventilation system. All supply ducts having diameter φ of 355 mm are galvanized steel and are not insulated. The microscale roughness, k, of the galvanized steel ducts is estimated to be in the range 6–60 μ m. Outdoor air, together with a proportion of the air returned from the air-conditioned workshops (re-circulated air) is drawn into a central air handling unit (AHU). This mixed air is cleaned by means of two types of filters; the G filter and F filter with a 40% and 85% ASHRAE dust spot average efficiency, respectively. The mean air velocity U _∞ at the centerline of the supply duct is equal to 5.4 ± 0.1 m.s⁻¹, giving a Reynolds number of 1.3× 10⁵. The central AHU delivers air at the temperature and the relative humidity close to 20°C and 20%, respectively.

FIG. 1 Schematic diagram of the ventilation system.

During this study, temperature was also measured at locations A, B, C, and D within the technical room. The measured temperature difference between the air in the supply duct and the air in the technical room was in the range 0–10°C and it depends largely on seasonal temperature variability.

Optical particle counter (Grimm OPC, Model 1.108) and cascade impactor (Andersen sampler) were used to measure the mass- and number-size distributions of aerosols, at an air-flow rate of 1.2 L/min and 22 L/min, for 648 and 672 h, respectively. Thus, the Grimm OPC recorded a particle number concentration size distribution every 5 s. Two isokinetic probes were used to sample the aerosol, at a close location in the supply duct (point C in Figure 1). The corresponding mass and number concentrations of the aerosol particles in different size ranges are given in Table 1 and Table 2, where the number concentration is calculated from the average weekly number concentration. Hence, the particles that will be discussed in this article are in the size range 0.3–20 μ m.

TABLE 1 The measured mass concentration for different stages of the Andersen sampler

Stage	Size range of aerodynamic diameter (μ m)	Mass concentration dm in μ g.m^− 3
1	0.65–1.1	0.64 ± 0.04
2	1.1–2.1	1.23 ± 0.04
3	2.1–3.3	0.74 ± 0.04
4	3.3–4.7	0.57 ± 0.04
5	4.7–7	0.19 ± 0.04
6	7–20	0.28 ± 0.04

TABLE 2 The measured particle concentration by the Grimm OPC

Size range of optical diameter (μ m)	Number concentration (Average ± Standard Deviation) dn in Number. m^− 3
0.3–0.4	18568000 ± 300000
0.4–0.5	5673000 ± 100000
0.5–0.65	2110000 ± 300000
0.65–0.8	525000 ± 50000
0.8–1	286000 ± 20000
1–1.6	108000 ± 8000
1.6–2	60000 ± 8000
2–3	52000 ± 1000
3–4	12000 ± 4000
4–5	4000 ± 2000
5–7.5	1200 ± 600
7.5–10	45 ± 30
10–15	3 ± 2
15–20	0.13 ± 0.2
> 20	0

The measured size distributions can be presented through a differential plot between dM/dlog d _p (dn/dlogd _p ) and d _p , which shows the modes of particle size distribution as well as the amount of sample which is found in each size range along a continuous spectrum. To provide a better presentation of the distribution of mass concentration of particles in the size interval from 7 to 20 μ m, we assumed that the relative concentration is constant (i.e., dM/dlogd _p = 0.614 μ g/m³) for the size ranges of particles: 7–10.5, 10.5–15, and 15–20 μ m.

Figure 2 shows a comparison between the mass- and number-size distributions of aerosols computed from the samples collected using Andersen sampler and Grimm OPC. The mass size spectra show bi-modal distribution with a mode around 1.6 μ m and second mode around 4.0 μ m. Generally, aerosols of coarse sizes originate from the nature of the product and processing activities (i.e., mechanical operations). Finding coarse particles after F filter could be considered as not acceptable. Nevertheless, this observation is not unusual in real industrial conditions. In fact, many common HVAC filters are partially efficient for particle sizes less than 10 μ m (Hanley et al. 1994). Moreover, the bypass of air around filters has been observed, but rarely quantified (Sippola 2002).

FIG. 2 Comparison between size distributions retrieved from Andersen sampler and Grimm OPC.

The visual inspection of the inside of the ducts showed that the upward-facing surfaces were covered with heavy milk powder deposits. The deposition rates in side-facing surfaces and downward-facing surfaces are nearly the same and less important than those in upward-facing surfaces. This is metabolically favorable for microorganisms to attach to surfaces, replicate themselves and biofilm formation, as nutrients are more concentrated at an interface. The consequences of biofilm communities forming in food processing environments are substantial (Bower et al. 1996).

3.2. Mechanisms and Parameters Affecting Particle Deposition

The above measurements and corresponding parameters (u* = 0.25 m.s^{− 1}, 0.1 < k ⁺ < 1 and 0 < Δ T < 10) were introduced in the present model to predict particle deposition velocity in ventilation ducts of this dairy factory. The results are presented for particle deposition from fully developed turbulent pipe flow. The analysis is restricted to dilute suspensions meaning that the fluid motion is unaffected by the presence of the particles and that particle–particle collisions can be neglected. From the above description, the calculations were performed for spherical particles of milk powder with the fixed values of particle density ρ _p = 700 kg.m^{− 3} and thermal conductivity k _p = 0.06 Wm^{− 1}K^{− 1} (Chong and Chen 1999; MacCarthy 1985). Air parameters were taken at standard conditions of ρ _a = 1.2 kg.m^{− 3}, C _p = 1004 J.kg^{− 1}K^{− 1}, μ = 1.8× 10^{− 5} Pa.s, λ = 0.07 μ m, and k _a = 0.026 Wm^{− 1}K^{− 1}.

Figure 3 presents the variation of the nondimensional deposition velocity as a function of particle diameter for different mechanisms. Equations (22)–(23) are solved for isothermal flow, with k ⁺ = 0.1 (k = 6 μ m). The molecular and turbulent diffusion effect is calculated by assuming that turbulence is homogeneous. In this case, Equation (23) is solved by taking the lower boundary at y ₀ ⁺ = 0, with i = 0. Similar to previous studies, the deposition velocity monotonically decreases with increasing particle diameter.

FIG. 3 Comparison of the dimensionless deposition velocity for different mechanisms. For all computed curves, k ⁺ = 0.1, u* = 0.25 m.s⁻¹, and Δ T = 0.

Including interception effect changes the behavior of the deposition velocity. When the lower boundary is given by Equation (7) with k ⁺ = 0.1, the calculated deposition velocity increases substantially with increasing particle diameter.

In Figure 3, total deposition for vertical surface is calculated by retaining all terms in Equation (22), with i = 0. It merges with the “diffusion + interception” case for very small particles. For particles ranging from 4 to 20 μ m, the turbulent eddy fluctuations effect becomes important, and the deposition rate increases sharply as the particle diameter increases further (Zhao and Wu 2006b).

Line predicting total particle deposition for horizontal surface is also included in Figure 3. Equation (22) is now calculated with i = 1.When the particle diameter is large enough, the gravitational sedimentation strongly affects the motion of particles and controls the particle deposition much more than diffusion and turbophoresis effect.

Figure 4 shows the results of calculations where all transport mechanisms, have been included (Equation (22)). Four cases are considered: Δ T = 0, Δ T = 1 K, Δ T = 5 K, and Δ T = 10 K. For horizontal surfaces, thermophoresis would tend to increase deposition of particles less than 3 μ m in diameter. For vertical surfaces, the effect of thermophoresis is significant over the entire range of particle sizes. Figure 4b demonstrated that for vertical surfaces, the significant role of thermophoresis in enhancing particle deposition rates is especially effective for particles with a size of 0.3 μ m ≤ d _p ≤ 10μ m. These results corroborate calculations performed by Tsai and Lin (1999). From the plot (Figure 4b), when Δ T = 10 K, one can see that the deposition rate of particles less than 10 μ m in diameter increases by approximately one order of magnitude. In addition, thermal conductivity and particle density can affect thermophoretic deposition. Thus, in Figure 5 we show the comparison of dimensionless deposition velocity onto vertical wall at three thermal conductivities of particles and for Δ T = 5°C. For particles with a size of 0.875 ≤ d _p ≤ 5.85μ m, the model predicts a greater thermophoretic deposition rate of low conductive particles and proves the importance of this parameter in predicting particle deposition flux.

FIG. 4 Effects of thermal diffusion on the predicted deposition rate for (a) horizontal and (b) vertical. For all computed curves, k ⁺ = 0.1, u* = 0.25 m.s⁻¹, and k _p = 0.06 W.K⁻¹.m⁻¹.

FIG. 5 Comparison of predicted deposition velocity at three thermal conductivities of particles. For all computed curves, k ⁺ = 0.1, u* = 0.25 m.s⁻¹, Δ T = 5°C, and ρ _p is equal to 700, 920, and 1200 for Milk powder, Olive oil, and Glass, respectively.

Figure 6 illustrates the variation in total deposition velocity with particle diameter for horizontal surface and different roughness parameters: k ⁺ = 0.1, k ⁺ = 0.5, and k ⁺ = 1, which correspond to roughnesses k = 6, 30, and 60 μ m. Equations (22)–(23) are solved for isothermal flow. The calculations show that the presence of small surface roughness significantly enhances deposition especially of small particles. When k ⁺ = 1, the deposition velocity of 2 μ m particles varies by approximately two orders and one order of magnitude for vertical and horizontal surfaces, respectively.

FIG. 6 Effects of surface roughness on the predicted deposition rate for (a) vertical and (b) horizontal surfaces. For all computed curves, Δ T = 0 and u* = 0.25 m.s⁻¹.

The use of expressions validated only for fully developed turbulent flow seems to be the main drawback of theoretical models. For example, the friction velocity is a major parameter which determination was obtained by a simplified approach (Equation (9)). To illustrate its influence, we performed calculations of the dimensionless deposition velocity onto vertical wall for three friction velocities (Figure 7). Equations (22)–(23) are solved for isothermal flow with k ⁺ = 0.1. For submicron particles, the dimensionless deposition velocities are nearly the same. The reason is that the effect of turbophoresis on particle deposition is far less than that of Brownian diffusion in this range of particle diameters (Zhao and Wu 2006a). For larger particles, dimensionless deposition velocities are highly dependent on the magnitude of the friction velocity. Besides, the friction velocity indicates the level of turbulence in the flow and the degree to which particles may be transported and deposited to walls by turbulent fluctuations. Thus, for accurate evaluation of the deposition rate, this parameter needs to be properly modeled.

FIG. 7 Comparison of the dimensionless deposition velocity onto vertical wall at three friction velocities. For all computed curves, Δ T = 0 °C and k ⁺ = 0.1.

3.3. Comparison Between Models

Experiments using real HVAC materials for the evaluation of the deposition velocity are rare. As this study considers real ventilation duct, the only available measured data that could be used to validate the present model are those of Sippola and Nazaroff (2004). Figure 8 shows the comparison of the predicted deposition velocity with empirical model (Sippola and Nazaroff 2003) and measured data collected by Sippola and Nazaroff (2004) for vertical and horizontal surfaces. Here, Equations (22)–(23) are solved based on experimental parameters (Sippola 2002); k = 5 μ m, u* = 0.27 m/s, and Δ T = 0.4°C. It could be firstly mentioned that predicted deposition rates from empirical equations do not perfectly follow the trends established by the experimental data. These equations have the drawback that they cannot account for changes in the microscale roughness character of the ducts. In engineering practice, usually more than one mechanism can act simultaneously and their interactions must be considered for the accurate prediction of deposition rates. As the accuracy of the measurement data was within ± 12%, the predicted results by the present model agree well with the measured data, for horizontal surfaces owing to the influence of gravity on large particles. Figure 8b shows that a small temperature gradient (Δ T = 0.4°C) produces a better agreement between the prediction and the experimental data.

FIG. 8 Comparison of predicted results by the present model with the empirical model and measured data collected by Sippola and Nazaroff (2004) for (a) horizontal and (b) vertical surfaces.

These results confirm that developing theoretical models is still a promising way to predict the deposition velocity. Improvements of these models would probably come from a better consideration of turbulence and friction velocity.

3.4. Predicting Particle Deposition Flux

The present model and empirical models were applied to predict the DMF at vertical and horizontal surfaces of the ventilation ducts of the dairy food factory.

Figure 9 shows the DMF (μ g/m².year) at vertical and horizontal surfaces calculated by theoretical and empirical models for different mean particle diameters. Values of the DMF were calculated using Equation (19), where the values of C _∞ are the measured and estimated (i.e., in the range 7–20 μ m) mass concentration. Here, the theoretical values of V _d are calculated using Equations (22)–(23) for Δ T = 0.4°C, k = 5 μ m, and u* = 0.25 m/s. The predicted results of the total DMF at vertical and horizontal surfaces are given in Table 3.

FIG. 9 Comparison of deposited particle mass flux for the present model and empirical models at (a) horizontal and (b) vertical surfaces.

TABLE 3 The predicted results of total deposited particle mass in vertical and horizontal surfaces

Total deposited particle mass flux (mg/m^2.year)
Models	Vertical surfaces	Horizontal surfaces
Present model	5 ± 1	151 ± 5
Empirical model	9 ± 2	86 ± 3

For the same reason mentioned above, Figure 9 indicates a discrepancy between theoretical and empirical models. Using the present model, for super-micron particles, the DMF at the floor is about 6 to 60 times larger than that at vertical walls. For each particle size, the empirical model predicts the DMF at the floor about 5–12 times larger than that at vertical walls. Thus, the deposited particle mass on the floor is sufficient to check if a ventilation duct should be cleaned.

Despite the high mass concentration of small particles (d _p < 5 μ m), the corresponding values of the DMF are significantly smaller than that predicted for larger particles. As discussed above, the gravitational settling velocity is dominant, for most of the particle sizes studied here, and can vary over a few orders of magnitudes (Figure 3). This also explains why the DMF increases with increasing particle diameter, as shown in Figure 9a.

In Figure 9b, for particles diameter less than 8.75 μ m, the maximum DMF was obtained with the present model for the 1.6 μ m diameter particles. As the deposition velocity is quite constant in the range 0.65 to 8.75 μ m, the deposition is controlled by the mass concentration distribution. For particles larger than 8.75 μ m, the DMF increased with the diameter owing to the turbophoretic effect.

Errors occurring in the prediction of the deposition velocity would significantly alter the accuracy in predicting the particle accumulation by deposition in ventilation duct. Thereby, the total DMF predicted by the present model is two times larger at horizontal surfaces, and two times less at vertical walls than that predicted by empirical model (Table 3).

Ventilation systems can be potential sources of pollutants and contamination due to the accumulation of particles in their duct ventilation. Ventilation systems must therefore be maintained under optimal cleanliness conditions. According to Lavoie et al. (2007), the established cleaning initiation criteria correspond to 300 mg/m² for the ASPEC method (ASPEC, French association for the prevention and study of contamination). We can therefore state that the particle accumulation in a typical ventilation duct of a food factory would take approximately two years to meet the cleaning code. However, particles films on duct surfaces may enhance the interception and electrostatic effects, owing to microscale morphological changes in the surface. This should increase the particle deposition rates and advance the cleaning time.

4. CONCLUSION

In the Agro-food industry, heating, ventilation, and air conditioning (HVAC) systems are often used to limit worker and food product exposure to airborne particles by removing contaminated air. However, HVAC systems have proved to be a source of contamination with pathogenic microorganisms. In order to prevent contamination, proper attention must be paid to determine particle deposition rates in ventilation ducts and to determine the best methods for predicting these rates. Based on the presented results, the following conclusions may be drawn:

• Thermophoresis may play an important role on the deposition rate of coarse particles for non-isothermal cases. In dairy industry, the low thermal conductivity of milk powder increases the thermophoretic deposition.

• Calculations confirm that the presence of small surface roughness significantly enhances deposition of small particles. The micro topography of a surface can also complicate the cleaning process as crevices and other surface imperfections can shield attached cells from the rigors of cleaning.

• The predicted total deposited mass flux to the floor (upward wall) is about 30 times larger than that onto vertical walls. In this case, the most deposited particle mass is associated with the coarse particles, which was determined by the nature of the product (milk powder) and the industrial operations.

At present, empirical and theoretical models can not reliably be applied to predict accumulation rates of particle deposits in ducts. For the purpose of studying the particle deposition in ventilation ducts by theoretical models, knowledge on roughness structure, temperature gradient, flow characteristics need to be further investigated. Thus, new experiments will be performed in laboratory at real scale with HVAC materials. In addition, computational fluid dynamics (CFD) will be adopted to better analyze turbulence and to determine the friction velocity.

This research was carried out with financial support from the ANR (Agence Nationale de la Recherche) for the specific programme “Programme National de Recherche en Alimentation et nutrition humaine,” project CLEANAIRNET “Hygienic design of ventilation duct systems in food factories,” in which nine partners are involved: Adria Normandie, Cemagref, CERTES, Clauger, CSTB, GEPEA, Lactalis R&D, Laval Mayenne Technopôle, and Soredab.