Search

Detecting diseases early in pig farming is a promising way to help breeders monitor animal health, reduce antibiotic use and treat sick animals in a targeted manner. This study analysed individual water and feed consumption as a function of the live weight of weaned piglets and its relation to diarrhoea. Data were collected from 15 batches of 102 piglets each, using specific automata (i.e. connected feeders, connected drinkers, automatic weighing stations, RFID ear tags), and piglet health status was monitored daily. Analysis used seven machine learning methods. Mean water and feed consumption of healthy pigs did not differ significantly among weight categories, except in week five, due to atypical water consumption by a heavy pig. In addition, mean water consumption did not differ significantly between healthy and sick piglets, either on the day that diarrhoea was observed or two days before. However, mean feed consumption of healthy piglets and those with diarrhoea did differ significantly from day 4-9 post-weaning and even 24 and 48 h before the onset of clinical signs. Nevertheless, the machine-learning methods tested did not provide effective early detection of diarrhoea, due to insufficient sensitivity and specificity of results (74% and 58%, respectively) for field use. In future studies, new parameters such as automatic measurement of body temperature or location of piglets in the pen by image analysis could improve the performance of these methods., Détection précoce des diarrhées chez les porcelets sevrés à partir des données d'abreuvement, d'alimentation et de poids La détection précoce de pathologies en élevage porcin est une piste prometteuse pour aider les éleveurs à surveiller la santé des animaux, réduire les usages d'antibiotiques et traiter de façon ciblée les animaux malades. L'objectif de cette étude est d'évaluer la valeur prédictive des consommations individuelles d'eau et d'aliment, en fonction du poids vif, pour détecter précocement l'apparition de diarrhée chez des porcelets sevrés. Quinze lots de 102 porcs, logés dans des cases équipées de nourrisseurs, abreuvoirs et bascules connectées, ont été identifiés individuellement par des boucles RFID et leur état sanitaire a été suivi quotidiennement. Les résultats ont montré que les consommations moyennes d'eau et d'aliment des porcs sains n'ont pas été significativement différentes selon leur catégorie de poids (lourd, moyen, léger), sauf en cinquième semaine en lien avec le niveau d'abreuvement atypique d'un porc lourd. Par ailleurs, aucune différence significative n'a été observée entre les consommations moyennes d'eau des animaux sains et malades, ni le jour de l'observation de la diarrhée ni deux jours avant. A l'inverse, une différence significative a été notée entre les consommations alimentaires moyennes des porcelets sains et diarrhéiques sur la première période du post-sevrage, à la fois le jour même d'apparition de la diarrhée, 24 heures et 48 heures avant. Néanmoins, les méthodes d'apprentissage automatique testées n'ont pas permis une détection précoce des diarrhées de manière efficace, les résultats de sensibilité et de spécificité étant insuffisants (respectivement 74 et 58 %) pour une utilisation pertinente sur le terrain. Pour les études futures, de nouveaux paramètres comme la mesure automatique de la température corporelle ou la localisation des porcelets dans la case par analyse d'images pourraient améliorer les performances de ces méthodes. Early detection of diarrhoea in weaned piglets from individual feed, water and weighing data Detecting diseases early in pig farming is a promising way to help breeders monitor animal health, reduce antibiotic use and treat sick animals in a targeted manner. This study analysed individual water and feed consumption as a function of the live weight of weaned piglets and its relation to diarrhoea. Data were collected from 15 batches of 102 piglets each, using specific automata (i.e. connected feeders, connected drinkers, automatic weighing stations, RFID ear tags), and piglet health status was monitored daily. Analysis used seven machine learning methods. Mean water and feed consumption of healthy pigs did not differ significantly among weight categories, except in week five, due to atypical water consumption by a heavy pig. In addition, mean water consumption did not differ significantly between healthy and sick piglets, either on the day that diarrhoea was observed or two days before. However, mean feed consumption of healthy piglets and those with diarrhoea did differ significantly from day 4-9 post-weaning and even 24 and 48 h before the onset of clinical signs. Nevertheless, the machine-learning methods tested did not provide effective early detection of diarrhoea, due to insufficient sensitivity and specificity of results (74% and 58%, respectively) for field use. In future studies, new parameters such as automatic measurement of body temperature or location of piglets in the pen by image analysis could improve the performance of these methods.

International audience; Suboptimal resource utilization due to outdated building standards, control systems and barn management approaches results in high emissions and losses within EU pig production. Additionally, with increasing concern for animal welfare, pig farmers are struggling to maintain competitiveness. The PigSys project addresses these issues by considering a whole system improvement towards a future sustainable and animal friendly pig production. The developed data warehouse platform is accessible to end-users to collect information on building characteristics and data from on-farm equipment, to connect them with a decision support system (DSS based on a bioclimatic model called ThermiPig). The DSS simulates the thermal balance and climatic conditions of a fattening room depending on a farmer’s decisions. A system using deep learning and machine vision techniques was developed to monitor pig behavior, providing a tool for early warning of problems related to pig welfare and health. New sensor data was used to evaluate level of noxious gases that impact pig performance and welfare. Pig cooling with showers over the slatted area and increasing air velocity at pig lying area were cost-effective options improving hygiene and mitigating ammonia emissions, both being crucial for improved environmental and social acceptability. Models to assess the trade-offs between environmental and bio-economic impacts of different pig housing and manure management were developed.

International audience; The dynamic model ThermiPig was developed to simulate thermal balance at the fattening room scale, as the result of the heat produced by the group of pigs or provided by incoming air and equipment and that lost due to air renewal or thermal conductivity of the walls and ceiling. A growth model (assuming thermoneutral conditions, InraPorc) and a bioclimatic model (ThermiSim) were combined to create ThermiPig. It considers a common time step (1 hour), definition of the pig thermoneutral zone (based on body weight and type of floor), evaluation of impacts of cold and hot exposure on the amount of available energy for growth, and the circadian distribution of heat produced daily by each pig (based on meal partition over the day and post-prandial thermic effect of feed). Data collected in vivo from a group of 96 pigs of the IFIP experimental station were used to evaluate the accuracy of the prediction of ambient temperature (T). Characteristics of the fattening room, regulation rules of the climate box control of the ventilation system, nutritional characteristics of the 2-phase feeding sequence, and outdoor conditions (hourly T and relative hygrometry) observed during the in vivo trial were used as inputs of the model, and simulation were performed on 30 virtual groups of pigs from the same genders and crossbreed. With average in silico growth performance of the 30 groups similar to observed ones, the error of prediction (RMSEP) on ambient T averaged 0.6°C. Thereafter, ThermiPig was used to simulate the impacts of alternative regulations for ventilation or the use of new equipment (heater, pad cooling) on growth performance of pigs, N excretion, and indirect energy (from feed intake) and direct energy (from electricity consumed by equipment) consumption. Comparison of multi-criteria performance in silico helps identify the most interesting option. Economic performance depends on carcass value minus costs of feed intake and electricity (depending on the energy source). Therefore, more accurate prediction of carcass value is expected in the next version of the model by considering effects of ambient temperature on the distribution of body fat and protein metabolism.

International audience; A mechanistic, dynamic and determinist model, called ThermiPig, has been developed by combining a growth model and a bioclimatic model. It allows for the prediction of the within-room thermal balance at the room level and its consequences on the daily and individual performance of group-housed pigs. As this research is part of the Pigsys ERA-Net project, co-funded under EU Horizon 2020 RI program (SuSan, www.era-susan.eu, Grant Agreement n°696231) by the French ANR (grant n°ANR-16-SUSN-0003-02), a survey was performed in different European countries by the partners of the project in order to describe different types of fattening rooms, feeding strategies and climatic conditions. Afterwards, the objective of the study was to use the results of the survey and to combine different characteristics of the fattening room (insulation, cooling and/or installed heater capacity when available, regulation rules of the climate control box) to simulate the corresponding impacts on (i) the thermal balance at room scale, (ii) growth performance, (iii) direct and indirect energy use, and (iv) nitrogen output in pigs. Simulations were performed over the whole fattening duration with each combination using 30 virtual batches of pigs, generated from an average animal growth profile. Based on outputs of the model, it is possible to evaluate the influence of insulation of walls or heater capacity under different climates or to optimize regulation rules of the climate box control of the ventilation system in terms of feed efficiency, margin on feed cost and nitrogen output.

International audience; Pigs in the EU are commonly reared in climate-controlled systems mainly dependent on air ventilation based on a single point (one sensor) temperature set value. These systems assume even spatial distributions of the indoor climate. Since pigs are sensitive to both temperature and other environmental aspects such as humidity and noxious gas concentration, adverse climate conditions can decrease animal welfare and consequently lead to economic losses. In an on-farm study, we measured temperature, relative air humidity, air velocity, ammonia and carbon dioxide concentrations continuously on five different measuring points in one room for a total of three consecutive batches (672 pigs per batch). Fresh air was supplied through ceiling diffuse inlet by a negative pressure ventilation system with two exhaust air ducts which were automatically controlled by the climate control unit The animals were continuously monitored using four network cameras (CCTV) during the whole batches (growing period). Deep learning techniques were developed to automatically detect the activity level of the animals in the recorded video data. Severe spatial differences (horizontal and vertical) in environmental conditions between the measurement points and temporal changes within and across batches were observed, which directly affected the animals’ activity levels. The results demonstrate that environmental control based on single point measurements can have severe shortfalls. Furthermore, environmental information beyond air temperature need to be integrated into climate control to assure appropriate conditions for the animals' health and welfare.

National audience

The dynamic model ThermiPig was developed to simulate thermal balance at the fattening room scale, as the result of the balance between the heat produced by the group of pigs or provided by incoming air and equipment and that lost due to air renewal or thermal conductivity of the walls and ceiling. A growth model (assuming thermoneutral conditions, InraPorc) and a bioclimatic model (ThermiSim) were combined to create ThermiPig; it considers a common time step (1 hour), definition of the pig thermoneutral zone, evaluation of impacts of cold and hot exposure on the amount of available energy for growth, and the circadian distribution of heat produced daily by each pig. Data collected in vivo were used to evaluate the accuracy of the prediction of ambient temperature, which averaged 0.6°C (RMSEP, error of prediction). ThermiPig was used to simulate impacts of alternative regulations for ventilation or the use of new equipment (heater, pad cooling) on growth performance of pigs, N excretion, and indirect energy (from feed intake) and direct energy (from electricity consumed by equipment) consumption. Comparison of multi-criteria performance in silico helps identify the most interesting option. Economic performance depends on carcass value (based on the French payment grid) minus costs of feed intake and electricity (depending on the energy source). However, more accurate prediction of carcass value is expected in the next version of the model by considering effects of ambient temperature on the distribution of body fat and protein metabolism., Un modèle de croissance du porc élevé dans des conditions de thermoneutralité (InraPorc) est combiné à un modèle permettant de simuler l’équilibre thermique de la salle d’engraissement selon la chaleur apportée par l’air entrant, les équipements ou les animaux, et les déperditions par la ventilation ou les parois (ThermiSim). La jonction de ces deux modèles repose sur le choix d’un pas de temps commun (1 heure), la définition de la zone de confort thermique pendant l’engraissement, la quantification des effets de l’exposition au chaud ou au froid sur la quantité d’énergie disponible pour la croissance, et enfin la répartition dynamique de la chaleur produite par les animaux au cours de la journée. Les résultats d’un essai in vivo sont utilisés pour évaluer la précision de prédiction de la température ambiante par ce modèle, appelé ThermiPig ; elle est en moyenne de 0,6°C (RMSEP, erreur de la prédiction). Des simulations sont ensuite réalisées pour évaluer l’incidence d’un changement des règles de régulation du boîtier de ventilation et de l’installation de nouveaux équipements (chauffage ou pad cooling) sur les performances de croissance du groupe de porcs, leurs rejets azotés, leur consommation d’énergie directe (électrique) et indirecte (aliment). En comparant les résultats des simulations, il est possible d’identifier la situation la plus intéressante sur la base de ces différents critères. La différence entre lavalorisation des carcasses dans la grille de paiement et le coût alimentaire et en électricité pour faire fonctionner les équipements, qui dépend de la source d’énergie utilisée pour produire l’électricité, permet d’évaluer la performance économique. Toutefois, la prédiction de ce critère doit encore être affinée par une prise en compte plus précise de l’influence de la température sur la conformation des porcs et sur leur métabolisme protéique.

International audience; Purpose. Feed production and excretion of nutrients by the animals are major sources of the environmental impacts of pig production. New communication and information technologies (NTIC) allow the development of precision feeding (PF) in pig-fattening units, which appears as a promising way to decrease environmental impacts of pig production. Therefore, we performed a life cycle assessment (LCA) of pig production with either conventional or PF applied to growing-finishing pigs.Methods. Two sources of data were involved: experiments and simulations performed using a model of pig fattening. A first experiment compared two-phase (2P) ad libitum (AdLib) feeding to ad libitum PF. A second experiment compared 2P restricted (Res) feeding to restricted PF. With the model, four feeding programs – 2P, 2PLCP (2P with low crude protein content), MP (daily multiphase group feeding) and PF – were simulated according to two feeding levels: AdLib and Res. A cradle-to-farm gate LCA was performed for each strategy, with the kg of live pig at farm gate being the functional unit. Five categories of impacts were evaluated: ILCD climate change (CC), ILCD acidification (AC), CML 2001 eutrophication (EU), CML 2001 land occupation (LO) and CED V1.8 non-renewable and fossil energy demand (CED).Results and discussion. PF slightly increased feed conversion ratio (FCR) and total feed intake compared with their reference 2P strategies, excepted in the in vivo experiment comparing PF-AdLib to 2P-AdLib. In this latter case, FCR was improved with PF. PF resulted in reduced N intake and N excreted relatively to 2P feeding in both in vivo and virtual experiments, whatever the rationing plan applied. PF always decreased AC and EU by reducing nitrogen excretion. With experimental data, CC was reduced in PFAdLib but not in PFRes compared to their respective 2P strategies. Simulated results showed a decrease in AC, EU and LO impacts, but not in CC, what is probably due to the small difference in total feed consumption between ad libitum and restrictedly-fed pigs (5 kg/pig). Conclusions. PF can reduce environmental impacts of pig production, particularly AC, EU and LO. Nevertheless, the definition of constraints on amino-acid and crude-protein contents and feed formulation should be made in accordance with local practices (i.e., feeding level) in order to obtain environmental gains expected with PF.

International audience; A test was performed with 96 growing pigs to implement precision feeding (PF) of restricted-fed growing pigs to characterise growth performance and N output using a PF system developed in the Feed-a-Gene project (H2020, no. 633531) compared to a 2-phase strategy (2P). A decision support tool was used to manage data (mainly individual and daily body weight (BW) measured with an automatic scale), to forecast performance on the following day, to assess corresponding amino acid requirements, and to adapt the quality of the diet delivered on the next day through the proportion of two diets used in a blend. Both diets were formulated to 9.75 MJ net energy (NE)/kg, with contrasting concentrations of digestible lysine and crude protein (A: 1.0 g lysine/MJ NE and 16.6% N x 6.25, B: 0.5 g/MJ NE and 10.9% N x 6.25). Forecasting BW and BW gain for PF pigs was interrupted for 2 weeks due to a problem with an electronic device for 2 consecutive days; it identified a problem of robustness in how forecasting methods were applied. A solution was found and applied to improve the entire system. However, no significant differences were observed between strategies for growth rate (2P: 722 vs PF: 716 g/d, P = 0.62), feed conversion ratio (2.64 vs 2.70, P = 0.063) or carcass leanness (62.0 vs 61.4 units, P = 0.18). The crude protein content of the blend provided to 2P during the growing and finishing phases was 15.8% and 13.4%, respectively. With this new 2P diet management, N output can potentially be reduced by 10% compared to a 2P with standard diets (16.0% and 15.0% N x 6.25). Despite the technical problem, which might have minimised the difference in N intake betweenstrategies, the PF contributed an additional decrease of 6%.

International audience; Precision feeding (PF) in growing-finishing pigs appears as a promising technology to decrease environmental impacts of pig production. However, few studies have compared environmental impacts of conventional and PF. Therefore, the aim of this study was to perform life cycle assessment (LCA) of pig production with either conventional or PF applied to growing-finishing pigs. Two approaches were used: using data from two experiments or relying on simulations performed with a fattening-unit model. In the ExpAdLib experiment, two-phase ad libitum (AdLib) feeding was compared to ad libitum PF. In the ExpRat experiment, two-phase restricted (Res) feeding was compared to restricted PF. In the model, four feeding programs − S2PREF (reference two-phase), S2LowProt (two-phase with low protein content), SMPGr (daily multiphase group feeding) and SMPInd (daily multiphase individual feeding) − were simulated according to two levels of supply: AdLib and Res. PF always decreased acidification and eutrophication impacts by reducing nitrogen excretion. Experimental results showed that climate change was reduced in PF AdLib but not in PF Res. Simulated results showed decrease in acidification, eutrophication and land occupation impacts, but not of climate change because of a worse feed conversion ratio. This LCA confirms that PF is a technology for reducing environmental impacts of pig production, particularly acidification and eutrophication. Nevertheless, definition of constraints on amino-acid and crude-protein contents, feed formulation and definition of the level of feed restriction should be made carefully in order to obtain the expected environmental gains with PF.; L’alimentation de précision (AP) des porcs charcutiers pourrait réduire les impacts environnementaux de la production porcine. Pourtant peu de résultats ont été publiés sur ce sujet. Cette étude visait à évaluer par analyse du cycle de vie les impacts de l’AP de porcs charcutiers en comparaison d’une alimentation conventionnelle. Deux approches sont utilisées : l’une reposant sur des données obtenues dans deux expérimentations et l'autre sur des simulations réalisées avec un modèle de l’atelier d’engraissement. Dans l’expérimentation ExpAdLib, une stratégie biphase (2P) ad libitum (AdLib) a été comparée à une AP AdLib. Dans l’expérimentation ExpRat, une stratégie en biphase rationné (2Prat) a été comparée à une alimentation de précision rationnée (Rat). Quatre programmes d’alimentation ont été simulés : S2PREF = biphase deréférence, S2PBasProt = biphase à basse teneur en protéines, SMPGr = multiphase journalier groupe et SMPInd = multiphase journalier individuel, selon deux plans de rationnement : AdLib et Rat. L’AP réduit de façon systématique les impacts acidification et eutrophisation, grâce à la réduction de l’excrétion d’azote. Pour l’impact sur le changement climatique, les résultats expérimentaux suggèrent une amélioration en AP AdLib mais pas en AP Rat. Les résultats simulés font état d’une amélioration claire sur l’acidification, l’eutrophisation et l’occupation des sols, mais pas d’amélioration sur le changement climatique en raison de la légère augmentation de l’indice de consommation. Cette étude confirme que l’AP est un levier de réduction des impacts environnementaux de la production de porcs charcutiers, particulièrement pour l’acidification et l’eutrophisation. L’AP doit faire l’objet d’une approche globale incluant la définition des niveaux d’acides aminés des aliments, la formulation des aliments, et la définition du niveau de rationnement, pour obtenir le meilleur gain environnemental.

International audience; Within the Horizon 2020 EU program Feed-a-Gene, a decision support system (DSS) was developed to implement precision feeding (PF) in commercial pig farms and to help improve feed efficiency. This study aimed to perform PF with the DSS in practical conditions with growing pigs fed ad libitum and to assess consequences on performance and nutrient use. Sixty-four pigs were reared from 77 to 161 days of age (33.5 to 108.8. kg body weight, BW) in a single pen equipped with an automatic weighing-sorting system and eight automatic feeders that register feed intake and deliver a tailored blend of two diets (A and B, respectively 1.0 and 0.4 g SID Lysine(Lys)/MJ net energy (NE), and 9.7 MJ NE/kg) to individual pigs. The control group received a blend providing 0.9 g Lys/MJ NE until the group weighed 65 kg on average (growing phase) and 0.7 g Lys/MJ NE thereafter (finishing phase). For the PF group, the Lys requirement was assessed individually and on a daily basis, based on up to 20 previous records of BW and feed intake, and diets A and B were blended accordingly. Daily feed intake, average daily gain, and feed conversion ratio did not differ between treatments. During the growing period, Lys and nitrogen (N) intake and N excretion were 11%, 9%, and 14% lower in the PF group than those in the control group, respectively (P < 0.05). During the finishing period, these values were only numerically lower (difference 0.66). These results could be explained by the slightly higher feed intake in the PF group (+100 g/d, P = 0.24) and the lower Lys content used during the finishing period of the 2-phase strategy compared to standard diets.

Since the new welfare regulation, farmers have to breed pregnant sows penned in the group. Thus, breeders observed more heterogeneity in the backfat thickness of sows when they are entering the farrowing units, implying more losses of piglets. It is also more difficult to observe lameness issues in large groups of sows. The objectives of this project are (i) to develop an activity sensor to feed each sow according to the energy it spends and (ii) to create an early detection system for lameness problems. The first step of this project was to develop a sensor able to record the individual activity level of sows penned in a group. As a result, Acti’Sow has been created. It is an ear tag accelerometer offering to know the daily time spent lying, standing and walking by a sow with a global accuracy close to 85 %. This project offers a better knowledge about sow behavior thanks to automatic feeders, connected drinkers, weighing scale and activity sensors. On average, the daily water consumption is 8.2 l/day/sow, but this result hides a huge variability close to 50 % when comparing a sow to another and 38 % for the same sow from a day to the next one. About their activity, an average sow spends 67 % of its time lying down, a bit more than 28 % standing up without moving and less than 5% walking. According to that, between the laziest one and the more active one, energy expenditure represent more than 500 g offeed. It means, with the same fixed objective of backfat thickness with these two sows, a farmer will need to give 500 g more feed/day for the very active sow. Early warning system for lameness issues was the other main result of this study. Through the use of watering and feeding behavior (number of visits per day, time of each visit, quantity of water/feed consumed, access rank to the feeder), individual weight and activity level, we built a first model able to individually predict lameness issue 24 hours before the farmer can observe it. The accuracy is close to 77 %. It means, that a cell phone app can alert farmers when a sow needs to be checked.; Depuis la mise en groupe des truies gestantes, les éleveurs observent plus d’hétérogénéité de l’état corporel des truies lors de leur entrée en maternité impliquant plus de pertes de porcelets. Il est également plus difficile d’observer les problèmes d’aplombs dans des grands groupes de truies. La première étape de ce projet avait pour objectif le développement d’un capteur capable d’enregistrer le niveau d’activité individuel des truies logées en groupe conformément à la réglementation bien-être. Cette étude a donc permis la mise au point de l’Acti’Sow. Il s’agit d’un accéléromètre autonome, positionné à l’oreille des truies, qui permet de connaître le temps quotidien passé par les truies à rester couchée, debout ou encore à marcher. Par ailleurs, cette étude offre également une meilleure connaissance du comportement des truies grâce aux nourrisseurs et aux abreuvoirs connectés, à la station de pesée identifiée et aux capteurs d’activité. En moyenne, la consommation d'eau quotidienne d’une truie est de 8,2 l / jour mais ce résultat cache une variabilité conséquente de près de 50% lorsque l'on compare une truie par rapport à une autre et de 38% pour la même truie d'un jour à l'autre. À propos de leur activité, une truie « normale » passe 67% de son temps en position couchée, un peu plus de 28% en position debout sans bouger et moins de 5% en marche. Sachant cela, entre la plus fainéante et la plus active, la dépense énergétique liée à cette activité représente plus de 500 g d’aliment. Le système d'alerte précoce des problèmes de boiterie est l'autre principal objectif de cette étude. En utilisant les comportements alimentaire et hydrique (nombre de visites par jour, heure de chaque visite, quantité d’eau / d’aliments consommée, rang d’accès au nourrisseur / abreuvoirs), le poids individuel et le niveau d’activité, nous avons construit un premier modèle capable de prédire individuellement les boiteries 24 heures avant que l’agriculteur puisse l’observer. La précision est proche de 77%

Pig livestock farming systems encounter several economic and environmental challenges, connected with meat price decrease, sanitary norms, emissions etc. To deal with these issues, methods and models to assess the performance of a pig production system have been developed. For instance, Thermipig model represents the pig fattening room and simulates performances of pigs at the batch level, taking into account interactions between the individual variability of pigs, farmer's practices, room characteristics and outdoor climate conditions. The model requires some static basic inputs fulfilled in several spreadsheets (such as rooms, pigs, and dietary characteristics) but also data files for voluminous variable inputs (such as outdoor temperature or climate control box parameters) for further modelling and outcome producing. This leads to challenges in data providing by the farmers and have to be improved. This paper deals with the implementation of the separate modules of the developed data warehouse system for usability improvements of the Thermipig model. The idea is to substitute input from the data files with online data input and automated variable processing by the model using the python script for connection to the remote data warehouse. The data warehouse system is extended with ‘Property Sets’ section dealing with all the operations that can be performed to a set of input variables. This approach demonstrates the ability of the data warehouse to act as data supplier for the remote model. As well the outcome of the model is also transferable back to the data warehouse for evaluation. This work is done within the Era-Net SuSan PigSys project - Improving pig system performance through a whole system approach.

Modelling growth performance of pigs and within-room thermal balance in different local conditions. 70. Annual Meeting of the European Federation of Animal Science (EAAP)

National audience; Precision feeding is a promising strategy to improve the efficiency of resource use by improving the adequacy between nutrient supplies and animal requirements. Technologies that identify each pig within a group (RFID ear tags), weighs it automatically and mixes different diets to adapt the quality of the feed ration on an individual and daily basis (precision feeders) were combined in a decision support system developed in the H2020 Feed-a-Gene project. It also includes a conceptual model to estimate nutritional requirements that relies on prediction of body weight (BW) and BW gain. These day D+1 criteria must be predicted from daily and individual measurements of BW performed up to day D on growing pigs group-housed in a pen equipped with an automatic weighing scale. The BW predicted with the Holt-Winters’ double exponential smoothing model (HWα, with the smoothing parameter α set at 0.6) were compared to measurements performed over at least 97 days on two groups of 96 pigs. From the 85 and 83 individual growth curves available, the slope of the regression between mean measured and predicted BW averaged 0.98 (R² = 0.99). Based on 9080 and 7662 measured BW available (respectively in each group), mean daily RMSEP regularly varies over time around 4-5% (without any particular event during the trial). Consequently, the accuracy of the prediction method was considered to meet expectations.

Improvement of feed efficiency in growing pigs is a key issue for the economic and environmental sustainability oflivestock production. This can be achieved with novel techniques such as precision feeding (PF). Within the Horizon 2020 EU Feed-a-Gene program (grant agreement n°633531), a decision support system (DSS) was developed to implement precision feeding in commercial pig farms. This study aimed to test the functioning of the DSS in practical conditions and the consequences on performance and nutrient use of growing pigs fed ad libitum. Sixty-four pigs were reared from 77 to 161 days of age (34 to 109 kg body weight, BW) in a single pen equipped with an automatic weighing-sorting system and eight automatic feeders allowing to register feed intake and deliver a tailored blend of two diets (A and B with respectively 1.0 and 0.4 g SID Lysine (Lys)/MJ NE, and 9.7 MJ NE/kg) to individual pigs. Pigs of the control group received a blend providing 0.9 g SID Lysine (Lys)/MJ NE until the average group BW was 65 kg (growing phase) and 0.7 g SID Lys/MJ NE thereafter (finishing phase). For the PF group, the assessment of the SID Lys requirement was performed individually and on a daily basis, based on up to 20 previous records of BW and feed intake. Feed composition was changed accordingly by blending diets A and B in appropriate proportions. Daily feed intake, average daily gain, and feed conversion ratio did not differ between treatments for the overall period or per period. During the growing period, the SID Lys intake and the nitrogen intake and excretion were respectively 10.8, 8.8, and 14.4% lower in the PF group compared to the control group (P0.68). This could result from a slightly higher feed intake (+100 g/d, P=0.24) in PF group combined with a SID Lys supply already low in control group. A second experiment will be performed in the same conditions to confirm the potential of the PF using the developed DSS.

International audience; Nutritionists, feed companies and equipment manufacturers look for solutions that help farmers to improve sustainability of pig production. Based on experimental results obtained in silico or in vivo, a better adequacy between amino acid supplies and requirements increases feed efficiency and farmer’s income and reduces the environmental impact of growing pigs, highlighting the interest for precision feeding. Data are collected to characterize daily animal traits (e.g. body weight, BW) and their variation from one day to another (e.g. growth rate, ΔBW). They are used to determine the requirement for maintenance and growth on the next day, respectively. Therefore, adequacy between requirements and supplies depends on these predicted BW and ΔBW. The double exponential smoothing (Holt-Winters) method with a smoothing parameter α=0.6 (HW0.6), presents a low sensitivity to the number of latest values used to forecast BW. It seems to allow for a secured prediction of BW soon after the beginning of the growing phase (at least after 4 days). A group of pigs was used in restricted feeding conditions to compare results obtained either with a 2-phase feeding strategy, considered as the control treatment, or a precision feeding strategy based on BW forecasting with the HW0.6 method. Pigs allocated to both treatments were group-housed in the same pen, equipped with the decision support system built in the Feed-a-Gene project to manage the data, to determine in real-time the corresponding nutritional requirements, and to adapt the feed characteristics provided to each pig through the blend of two diets (9.75 MJ net energy/kg, 0.5 or 1.0 g of digestible lysine per MJ). Available results from 24 pigs per treatment indicate that overall average growth performance were not influenced by the feeding strategy (P>0.58 for both average daily gain and feed conversion ratio) but digestible lysine intake was reduced by 6% (1,774 vs 1,879 g, P

Even though animals are from the same genetic line, farmers have to cope with variability both in sows and piglets. In sows, variability is observed in traits such as parity, prolificacy, appetite, body weight (BW) and back fat thickness (BF). For instance, at the beginning of gestation, variability in body condition among sows can be high due to parity and age. In addition, at a given age, variability in litter size, milk potential, and appetite results in different nutrient requirements and consequently in variability of changes in maternal body reserves. Variability in BF can be a problem as several studies have indicated that too high or too low BF values are to be avoided at farrowing as well as at weaning. In both cases, the longevity of the sow is impaired, and farmers are advised to manage the sows toward a target BF depending on the physiological stage, associated with an age-dependant BW, increasing with age up to mature BW. In addition, variation in sow’s body condition at farrowing and in prolificacy influences the new born and weaning piglet traits. Compared to less prolific sows, high-prolific sows farrow more piglets, which are both lighter on average and more heterogeneous. Compared to normal birthweight piglets, the survival rate of low birthweight piglets is lower. Providing additional care around birth helps these piglets to survive, but subsequent housing and feeding management have to be adapted to deal with the variability in their growth potential. Nutritional strategies (based on modelling approaches that take into account criteria that influence requirements) are suggested to optimise the expression of the animals’ potential, but most often without an intention to reduce inter-individual variability in growth performance. In order to control or reduce variability, other solutions have been evaluated in experimental studies that focus on the level and the dynamic of the feeding plan and the quality of the diet. The challenge is now to validate these solutions in production units, which will be more or less easy depending on the existing housing and feeding systems, and the economic, welfare and environmental context.

Les retardateurs de flamme bromés (RFB) sont des additifs destinés à limiter l’inflammabilité de certains matériaux, notamment les polystyrènes (PS). Parmi eux, les polybromodiphényléthers (PBDE) et l’hexabromocyclododécane (HBCD) peuvent s’accumuler dans les tissus animaux. Leur concentration dans les denrées alimentaires d’origine animale (DAOA) est généralement faible, mais elle atteint de l’ordre 100 à 1000 fois la médiane dans certains rares échantillons d’œuf et de viande de volaille ou de porc. Des prélèvements et des enquêtes de terrain menées dans le cadre du projet BrAviPorc (CASDAR, n°1256) ont permis (1) de confirmer la faible concentration de PBDE et d’HBCD dans les produits avicoles et porcins français, (2) d’établir que les isolants contenant des PS sont des sources d’HBCD en élevage qui, accidentellement ingérés par les animaux, conduisent à des DAOA très contaminées. Pour les trois productions, des expérimentations in vivo montrent que l’ -HBCD s’accumule préférentiellement dans les tissus les plus riches en lipides neutres. Ainsi, la cuisse de poulet est 10 fois plus concentrée en -HBCD que le filet. Pour la première fois, un modèle mathématique, commun aux trois productions et prenant en compte leurs dynamiques lipidiques respectives, a été calibré. Il pourra être adapté à d’autres polluants lipophiles et apolaires., Brominated flame retardants (BFRs) are synthetic substances primarily used as additive in insulating materials, including polystyrenes (PS). Among them, polybrominated diphenyl ethers (PBDEs) and hexabromocyclododecane (HBCD) can accumulate in animal tissues. Overall, their concentrations in animal-derived foodstuffs are low, but some rare samples of eggs, poultry meat and pork display concentrations exceeding 100 to 1000 times the median value. Sampling and field surveys carried ou[br/] within the BrAviPorc project (CASDAR, n° 1256) allowed i) to confirm the low concentration of PBDE and HBCD in French poultry and pork products, ii) to establish that insulating foams containing PS may be sources of HBCD in livestock buildings, which lead to highly contaminated DAOAs if animals accidentally ingest them. In vivo experiments carried out with laying hens, broilers and growing pigs showed that -HBCD accumulates preferentially in the tissues richest in neutral lipids. Thus, chicken leg is 10 times more concentrated in -HBCD than breast. For the first time, a mathematical model, common to the three animal models and taking into account their respective lipid dynamics, was calibrated. It can be adapted to other lipophilic and nonpolar pollutants

Two trials were conducted at experimental stations of IFIP, located in Romillé (France, Trial 1), and INRA, located in Saint Gilles (France, Trial 2), on pregnant sows equipped with individual ear tag accelerometers to record their activity level: duration of lying, standing and moving sequences. The first trial involved 72 sows penned on a slatted floor in a dynamic group with connected drinkers and automatic feeders, whereas the second trial was carried out on 4 small groups of 6 sows penned on a concrete floor with straw and fed in individual stalls. Firstly, an algorithm was built from video recordings of 24 sows on the slatted floor (2 x 2 h sequences per sow, 96 h). Secondly, the accuracy of the algorithm was assessed by recording and sequencing 96 h and 109 h, respectively, on the slatted floor and concrete floor with straw. The respective sensitivities of the lying, standing and moving behaviours on the slatted floor were 94.4%, 66.9% and 68.4%. With straw, lower sensitivity values were found: 93.65% for lying, 68.35% for standing and 38.83% for moving, linked to more investigative behaviours using the head. The final step was to use these data to improve the feeding practices of pregnant sows, taking their activity level into account. The strong inter- and intra-individual variability shown in the physical activity is a limiting factor for detection of health problems, such as lameness, through the accelerometers. Thus we need additional information, especially the behaviour data generated by identified drinkers and automatic feeders.

National audience; In pig farms, direct consumption of energy depends on thermal exchanges between indoor and outdoor, which are related to the rules of climate management in the rooms. Besides, the impact of indoor climate conditions on pig growth and efficiency is well known through an extensive literature over the last twenty years. Therefore, it appeared possible and consistent to support field advice by designing a tool based on a dynamic model (called MEDIBATE). This model represents and predicts at barn room level the energy balance impacted by ‘action levers’ such as room characteristics (insulation) and equipment (heating, ventilation), climate control management (temperature driving) and animals (number, status). The general design and the equations involved in MEDIBATE are described in this paper. The temperature is predicted every 30 seconds. The results are evaluated and compared to values measured in an experimental farm. This comparison shows an average daily gap of only 0.32°C (±0.19°C), lower than the accuracy of the monitoring device. The model also simulates the ventilation rate and heating required to reach the target temperature allowing the calculation of direct energy consumption. Indirect energy is calculated through the prediction of daily feed intake, considering the effect of both the live weight and ambient temperature. Finally, economic calculations will be added to the model to allow the comparison of different housing design and management strategies.; Dans les élevages de porcs, les consommations d’énergie directe sont, par nature, dépendantes des échanges thermiques entre les bâtiments et leur environnement, eux-mêmes conditionnés par la gestion de l’ambiance des salles. L’impact de ces conditions d’ambiance sur les performances des porcs en croissance est par ailleurs bien connu au travers de nombreuses publications au cours des vingt dernières années. Il est donc apparu possible et pertinent de réaliser un outil d’aide au conseil basé sur un modèle dynamique (MEDIBATE). L’objectif de ce modèle est de prédire les consommations d’énergie occasionnées par le chauffage et la ventilation des salles d'élevage, en relation avec des variables d’action comme les caractéristiques de la salle (isolation…), les équipements (chauffage, ventilation), la gestion de l’ambiance (températures de consigne…) et les animaux (nombre, stade). La structure du modèle et les équations utilisées sont décrites dans ce papier. La température est prédite toutes les 30 secondes. In fine, le modèle obtenu est en mesure de simuler la température d’une salle avec un écart journalier moyen de 0,32°C (±0,19°C) par rapport à des enregistrements réalisés in situ en stations expérimentales, soit un écart inférieur à la précision des sondes de température utilisées pour les mesures. La température prédite permet dans le même temps de simuler le taux de ventilation et le niveau de chauffage nécessaires pour respecter les consignes choisies par l’utilisateur. Une prédiction des consommations d’énergie directes peut alors être réalisée. Les flux d’énergie indirecte sont estimés à partir de la prédiction de la consommation moyenne journalière d’aliment, en tenant compte du poids vif de l’animal et de la température ambiante. L’association future d’un bilan économique à ce modèle permettra d’évaluer le résultat de la mise en oeuvre de différents choix techniques.

National audience; In pig farms, direct consumption of energy depends on thermal exchanges between indoor and outdoor, which are related to the rules of climate management in the rooms. Besides, the impact of indoor climate conditions on pig growth and efficiency is well known through an extensive literature over the last twenty years. Therefore, it appeared possible and consistent to support field advice by designing a tool based on a dynamic model (called MEDIBATE). This model represents and predicts at barn room level the energy balance impacted by ‘action levers’ such as room characteristics (insulation) and equipment (heating, ventilation), climate control management (temperature driving) and animals (number, status). The general design and the equations involved in MEDIBATE are described in this paper. The temperature is predicted every 30 seconds. The results are evaluated and compared to values measured in an experimental farm. This comparison shows an average daily gap of only 0.32°C (±0.19°C), lower than the accuracy of the monitoring device. The model also simulates the ventilation rate and heating required to reach the target temperature allowing the calculation of direct energy consumption. Indirect energy is calculated through the prediction of daily feed intake, considering the effect of both the live weight and ambient temperature. Finally, economic calculations will be added to the model to allow the comparison of different housing design and management strategies.; Dans les élevages de porcs, les consommations d’énergie directe sont, par nature, dépendantes des échanges thermiques entre les bâtiments et leur environnement, eux-mêmes conditionnés par la gestion de l’ambiance des salles. L’impact de ces conditions d’ambiance sur les performances des porcs en croissance est par ailleurs bien connu au travers de nombreuses publications au cours des vingt dernières années. Il est donc apparu possible et pertinent de réaliser un outil d’aide au conseil basé sur un modèle dynamique (MEDIBATE). L’objectif de ce modèle est de prédire les consommations d’énergie occasionnées par le chauffage et la ventilation des salles d'élevage, en relation avec des variables d’action comme les caractéristiques de la salle (isolation…), les équipements (chauffage, ventilation), la gestion de l’ambiance (températures de consigne…) et les animaux (nombre, stade). La structure du modèle et les équations utilisées sont décrites dans ce papier. La température est prédite toutes les 30 secondes. In fine, le modèle obtenu est en mesure de simuler la température d’une salle avec un écart journalier moyen de 0,32°C (±0,19°C) par rapport à des enregistrements réalisés in situ en stations expérimentales, soit un écart inférieur à la précision des sondes de température utilisées pour les mesures. La température prédite permet dans le même temps de simuler le taux de ventilation et le niveau de chauffage nécessaires pour respecter les consignes choisies par l’utilisateur. Une prédiction des consommations d’énergie directes peut alors être réalisée. Les flux d’énergie indirecte sont estimés à partir de la prédiction de la consommation moyenne journalière d’aliment, en tenant compte du poids vif de l’animal et de la température ambiante. L’association future d’un bilan économique à ce modèle permettra d’évaluer le résultat de la mise en oeuvre de différents choix techniques.

National audience

A feeder prototype has been developed by IFIP in association with an equipment manufacturer to perform precision feeding for 96 group‐housed growing pigs, each equipped with a RFID (Radio Frequency IDentification) ear tag. It consists in five automatic feeders (AF), one weighing‐sorting station (WS), and three areas. The WS station is placed on force sensors and allows for weighing pigs individually, with a 0.1 kg accuracy, and sorting them towards (1) the AF (pigs whose daily cumulated feed intake is presently below the daily allowance), (2) a pen for specific care or departure for the slaughterhouse (after selection by the farmer), (3) the original pen for the other cases. In case 1, the pig enters an AF, both back and front doors close automatically. Its ear RFID tag is recognized by an antenna, and successive doses of feed fall into the trough. Each AF can deliver (min 30 g) and mix two diets, and is connected to a computer that calculates the daily dietary allowances per pig, based on time, body weight (BW) and feeding strategy. In this first trial, all pigs were restricted and two feeding strategies were designed (2‐ vs 9‐ phase strategy) based on the same evolution of amino acid requirements with BW for all pigs. Taking into account pigs not adapted to the system after 9 days (n = 4), accidents (n = 2), health issues (n = 5) and those which failed to consume their feed allowance on more than 4 days (n = 14), data from 71 pigs were available at the end of the fattening period.

absent