EVATUATION SYSTEM OF EXHAUST FANS USED ON VENTILATION SYSTEM IN COMMERCIAL BROILER HOUSE

This study aim to develop a system, called FANS-N, for evaluation the exhaust fans in the ventilation system of broiler facilities. The system is divided into: 1) Mechanical Structure consisting of two stepper motors for positioning a anemometer sensor in the vertical and horizontal coordinates; 2) Electronic Interface control of the anemometer positioning and record data of wind speed; 3) Control Programming Module – accountable for the cursor movement, measurement and record the wind speed data with the anemometer at predetermined points; and 4) Analysis Programming Module responsible for the interpretation of wind speed values at each point. The software uses artificial neural networks (Multi-Layer Perceptron) for images analyses of data base. The output of neural network give to the user the following recommendations: "possible changing", "maintenance", "standard limit", and "within standard". The system was able to evaluate the exhaust fans, identify the failures and proposing solutions to farmers of a preventive diagnosis.


INTRODUCTION
The quantification of the ventilation rate is crucial for the control and maintenance of the environment in broilers production. Adequate management of ventilation rates helps to maintain optimal growing conditions inside the facility for the well-being and poultry production (Calvet et al., 2013).
However, determining the actual ventilation rate in a poultry facility is a difficult and complex task due to the effects of weather, unhealthy environment, maintenance lack of exhaust fans, dynamic and irregular effects of the winds, different positioning of the exhaust fans, variable number of exhaust fans, differentiated openings for air intake, existing cracks and the different typologies of the aviaries, although this one uses artificial ventilation system (Zhu et al., 2012;Zhao et al., 2013).
In this sense, several methodologies, standards and procedures were developed to determine the performance of exhaust fans and to measure airflow in laboratories (ASHRAE Standards, 1992, Wheeler & Bottcher, 1995AMCA, 1999;ASHRAE, 2001).
Afterwards, the methodology to evaluate the airflow of exhaust fans installed in broiler house in situs was developed by Simmons & Hanningan (2000) and later used by Gates et al. (2004). The developed system, called FANS (Fan Assessment Numeration System), allows accurate mapping of air velocity at the exhaust outlet. The system has proved to be adequate for evaluating the performance of the exhaust fans in environments protected from animals and plants (Morello et al., 2010;Zhi et al., 2015) including being adopted by the USDA (United States Department of Agriculture) as the standard methodology for assessing gas emission from poultry and pig farms in the United States of America.
Another technique to evaluate the ventilation system, zootechnical performance, and decision support are the Artificial Neural Networks (ANN). ANN are artificial intelligence tools that present a mathematical model inspired by the neural structure of intelligent organisms and that acquire knowledge through experience, adapts and learns to perform a certain task, or behavior, from a set 888 of given examples. Many studies have been carried out in the area of precision Zootechny using ANN to evaluate economic loss, evaluation and prediction of productivity, diet, ventilation system, among others (Faridi et al., 2012(Faridi et al., , 2014Sefati et al., 2014;Curi et al., 2014).
The use of artificial neural network is adequate in the analysis of figure patterns, since this technique is able to classify, organize and give answers related to the results to be obtained. The efficiency on finding patterns in figures and comparing them with a pre-established database is already very common in many situations, such as in the identification of people by their fingerprints or the iris (Al-Allaf et al., 2013;Godara & Gupta, 2013).
In this context, a system called FANS-N was developed for in situ evaluation of the performance of the exhaust fans in broiler house. This system, composed of automatic measurement of air velocity, neural networks for database interpretation and analysis, has the following recommendations: "possible switching", "maintenance", "standard limit", and "within standard".

MATERIAL AND METHODS
The system developed in the present study is divided into:

Mechanical structure
The assembly of the prototype was based on the recommendations proposed by Gates et al. (2004) for air velocity measurement at the exhaust outlet and improvements were included in the model that will be presented throughout the text (Fig. 1). FIGURE 1. Parts of the developed equipment: anemometer (a), motors for horizontal (b) and vertical (c) displacement, engine controller interface (drives) and analog / digital converter (d) and computer and software for engine movement, data logging and data analysis by neural networks (e).
The anemometer used was the thermistor. Its calibration principle is based on the technique known as "hot-wire" (Jorgensen, 2002;Valença, 2003). In this technique, the heating of a resistance is caused by the passage of electric current; the resistive element is maintained at a constant temperature. Then, when measuring the electric voltage in the resistor, the values are obtained proportional to the natural logarithm of the air velocity (Equation 1), observed in Valença, 2003. (1)

Electronic Interface and Control Programming Module
The sensor displacement system was made by means of belts positioned vertically and horizontally, being these moved by step motors of the type NEMA 34 (model KTC-5034-349). The driver controls the motors of the equipment which provides the necessary electrical voltage and the Evatuation system of exhaust fans used on ventilation system in commercial broiler house Eng. Agríc., Jaboticabal, v.37, n.5, p.887-899, sep./oct. 2017 889 correct energizing cycle of the motors spirals. The interface board between the equipment and the computer consists of a circuit based on the microcontroller PIC18F4550 (Microchip ® ) with serial communication via USB with analogical digital conversion (A / D) capability and digital outputs for the control of the motor drives.

Analysis Programming Module
The development of the entire FANS-N system was performed in the Delphi ® 6 programming language with two internal components: the first was the serial control called ComPort used to communicate the software with the project equipment via USB and the second component was the neural networks Multi-Layer Perceptron, for analysis by artificial neural network with Perceptron modeling.
The FANS-N system, as a whole, performs multiple functions, among them: Function 1movement of the motors for the positioning of the sensor; Function 2 -capture and storage of values measured by the anemometer and, Function 3 -use of algorithms and neural networks to interpret the results by searching for image patterns, relating to previous standards and efficiency data provided by the manufacturers of the studied equipment.

Neural networks
The neural network model, Multi-Layer Perceptron (MLP), was built in the WEKA ® program (Waikato Environment for Knowledge Analysis) version 3.6.9 (2013) through the backpropagation algorithm. The Cross-validation test was used for the model construction and validation. The crossvalidation technique uses the method of partitioning the data set into mutually exclusive subsets, then some of these subsets are used to estimate the model parameters (training data) and the rest of the subsets (Validation data) is used in the model validation.
MLP modeling provides sufficient information for decision-making in exhaust systems in broiler house (Curi et al., 2014). The use of neural network in MLP modeling is therefore satisfactory for image recognition and treatment (Lima et al., 2010).
In this study 500 images were collected and were used in the neural network modeling, with an average of 20 images for each related static pressure value. The training and testing of the neural network were performed subsequently.
The network parameters used were: ten epochs, zero neurons of hidden layer, ninety of learning rate and thirty of inertia rate, seven input levels (corresponding to each static pressure value) for each expected result (within the standard, standards limit, maintenance and possible changing). The neural network evaluation was performed by the accuracy of the model, that is, mean error less than 1% (0.00012) (Leal et al., 2015).
The inputs of the neural network were: air velocity, exhaust ventilation airflow, electric current, and static pressure.
The statistical method of analysis of variance (ANOVA) and the Tukey test were used to verify if the inputs to the neural network developed in the FANS-N program had significant importance and thus to validate the use of the chosen neural network.
With the result of the obtained flow and the graphical analysis performed by the neural network, the program provides as output, results of generated static pressure and energy efficiency (relation between the flow and the consumption of the exhaust fan). As the last return four types of status classification are made regarding the exhaust: "good", "need for maintenance", "need for technical adjustment" or "change exhaust ".
The validation of the equipment was performed in Blue House sheds with negative pressure ventilation of tunnel type with the following characteristics: Installation typology: broiler house with negative pressure artificial ventilation system with air inlet on the opposite side to the exhaust.
Location: municipality of Elias Fausto-SP.
Isolation: Roof made of asbestos cement tiles with slope of 14°, polyethylene curtain lining in blue color, polyethylene curtain side walls in the blue color on the inner face and silver on the outer face.
Building materials: Masonry structure in pillars and beams, wooden structure to support the roof, concrete floor, wall with 0.30 m of masonry height and anti-bird screen.
Ventilation system: composed by exhaust fans Big Dutchman ® model.
The study of the exhaust fans in broiler house consisted in the identification of those from 1 to 10 to start counting from left to right (Fig. 2). The air velocity was measured in certain exhaust fans, subsequently (Table 1) as a function of the number of exhaust fans and also of the static pressure produced by them. The methodology used was adapted from the method proposed by Morello et al. (2010).  Table 2), it is possible to observe that there are differences between the regions of the exhaust surface in relation to the air velocity as a function of the static pressure. Therefore, it is considered that the variations change due to the increase of the static pressure and the regions of the surface maintain their differences in relation to the means of their velocities (Fig. 3). There is convergence of the curves when the static pressure increases, helping in the confirmation of the feasibility of using the regions velocities mean with inputs to the neural network created in FANS-N.  Considering the total set of data concerning the exhaust fan surface, the differences between the regions stand out, as observed in the mean values of Table 2.

Interpolation surfaces
The interpolation surface is a mathematical tool that enables the study of air velocities on the surface of the exhaust fans, and its interpretation is able to identify the operating system of them. The generation of the interpolation surfaces in this study was performed by Surfer ® software version 10. The studied exhaust fans presented large central regions of low velocity, less than 3m s -1 and peripheral regions with higher velocities, above 8 m s -1 , similar to the results by Morello et al. (2010), and Wheeler et al. (2006). On Fig. 4, 5 and 6 it is possible to observe the precise details of the ventilation surfaces on the exhaust fans, such figures show the air flow produced by them. Details obtained by the interpolation figures indicate the influence of external factors on the exhaust fans as well as on the performance of them. Broiler house exhaust fans suffer greatly from heavy use and their engines undergo extreme wear and tear conditions, mainly due to the humidity and dust in which they are subjected.
Through the analysis of the interpolation surfaces it is possible to find the relation of this with the static pressure inside the broiler house, or to relate the height of the curtains with the flow produced by the exhaust fans.

Results of static pressure variation in exhaust fan 1
With a negative pressure of at 4 Pa there is a low interference in the operating system of the exhaust fans, so that

Results of static pressure variation in exhaust fan 3
In Fig. 6 (a) and (b) air velocities are practically zero in the centers and large amplitude in the periphery. The lower region in which the higher air velocities appear (shown in the previous figures for the exhaust fan 1) is less evident for this exhaust at 16 Pa. It is also observed that at 16 Pa there are several points in distinct regions with peaks of air velocities around 14 m s -1 , and it can be considered that there is more turbulence in the flow regime in this exhaust fan. Fig. 6 (b) and (c) maintain the central region feature at zero velocity, but the generalized air velocity drop can be observed in relation to the previous one (Fig.6 a)

Results of static pressure variation in exhaust fan 5
As in the previous analyzed figures, Fig. 7 (a) and (b) have the central regions with low velocity. Many "nodes" appear, corresponding to spikes of velocities scattered in both figures, mainly in Fig. 7 (b). has its air velocities on the surface smaller than on Fig. 7 (c), and the verification will be evidenced in the flow analysis of the two situations presented here.  Table 3 presents the flow results obtained by exhausts 1, 3 and 5 following the methodology of data collection adapted from Morello et al. (2010), where the activation and shutdown of the exhaust fans occurred to control static pressure and data collection. For statistical analysis, the Tukey test was performed at 5% of significance.  Table 3 shows that the values of air velocities obtained by the system are directly related to the values of static pressure of the broiler house, and the increase of the mean value of the air velocity is accompanied by the direct ratio of the flow rate and the inverse of the static pressure. The calculation of flow in this study (Equation 2) presented similar values to those obtained by the study done by Gates et al. (2004) and Morello et al. (2010). Ventilation is one of the environmental parameters and may contribute to the elevation of zootechnical indexes and well-being. The study of ventilation allows better understanding on the distribution of climatic variables inside the installations (Liang et al., 2014;Purswell et al., 2013;Mostafa et al., 2012;Guerra-Galdo et al., 2015).

Difference between exhaust fans and operating adequacy
It can be observed in Fig.8, obtained by FANS-N the flow drop as a function of the static pressure increase for each studied exhaust fan. As well as Morello et al. (2010), the curves obtained for the exhaust fans are linear relationships that can certainly characterize the exhaust fan and its operating conditions, since the curve establishes an exhaust fan behavior equation compared to the static pressure changes.  Table 4 shows that the obtained equations by both Minitab program as for FANS-N program it is verified a lot of similarity between the linear and angular coefficients of the calculated lines.  Table 4 shows the similarity between the linear and angular coefficients of the calculated lines, which indicates that both models are suitable for reference as the basis for the decision making / classification of the exhaust fan.
The generated interpolation figures followed as reference for analysis of all exhaust fans studied in their various static pressures and associated flow. For purposes of comparison and classification of the interpolation figures it became necessary to establish a standard figure. To do so, the exhaust fan 6 was chosen among the studied exhaust fans to originate the standard, since it is new and factory calibrated. The FANS-N system divided the interpolation figures into areas of greater importance to be considered as inputs to the developed artificial neural network. With the data of the sectors on generated figures for the exhaust fan 6 was established the results table of inputs for the neural network of the FANS-N system.
As a probabilistic responses to the performed tests, the obtained output of the neural network by the FANS-N system resulted in the values showed on Table 5. The response of the neural networks as observed on Table 5 helps in the understanding of what action should be taken so that the exhaust fans operate efficiently. To these four attributes (possible changing; maintenance; standard limit; within the standards) different values of probability are conferred, being the one that present greater probability should be the guideline of the attitude to be taken by the equipment user. As an example it is possible to observe that the exhaust fan 6 was most likely to be operating in the operating standard (limit -15% or within -68%) which corresponds to reality, since previously it was known that this was the calibrated exhaust fan.

CONCLUSIONS
The proposed FANS-N system was able to classify the exhaust fan in different operating modes, presenting the final status of the exhaust fan for decision making by the broiler house owner.