Performing Department
Animal and Poultry Science
Non Technical Summary
Precision field crop agriculture is seen as one of a handful of key, early-adoption, commercial opportunities for drones and drone-based technologies because of the demonstrated ability to survey large areas from an elevated vantage point to measure and map crop needs. The time scales for these types of interventions, while highly profitable, are often measured in days or weeks. Precision animal agriculture (PAA), on the other hand, presents a much richer environment for cyber-physical systems (CPS) research. Animals, individually, are complex organisms that require constant nutritional adjustment, yet they are social beasts with herd behavior that emerges from the collective. For issues of nutrition, health, productivity and efficiency, animal agriculture must treat both the individuals and the collective, making it ideal for the science and application of CPS principles. With growing global awareness of the negative effects of livestock production on non-renewable and renewable resources, concurrent with the negative effects of global population growth and the need to feed more mouths, the transformational impact of CPS on the largely unexplored realm of precision animal agriculture is enormous. This is particularly critical as a significant portion of the projected increases in global food production is anticipated to come from ruminants. This proposal presents basic science exploring the complex relationship between individual animal and herd behaviors on agriculture systems efficiency, while demonstrating its potential on the specific area of dairy farm management with the goal of improving sustainability and efficiency. The long-term objective of this team's research is to develop the foundations for CPS that apply, generically, to intensified management of the individual animal and herd in areas of nutrition, health, productivity, and efficiency that underpin this new area of precision animal agriculture.
Animal Health Component
10%
Research Effort Categories
Basic
75%
Applied
25%
Developmental
0%
Goals / Objectives
A set of 4 specific goals have been identified for this project:Develop the decision layer of the generic precision animal agriculture CPS that incorporates flexible animal and herd models, guided by real data examples. Utilize existing data to test model parameterization strategies for feeding and health to reformulate supplement mixes based on real-time performance inputs representing both the individuals and the herd.Develop the network layer of the generic precision animal agriculture CPS, including body networking to extract embedded sensor data from individual animals; herd networking and tracking to consolidate and report individual and collective herd data with appropriate edge analytic capabilities; cloud computing capacity and software algorithms to receive, log, and interpret data from the edge to the core; and on-farm production interfaces to evaluate the production environment of the entire system.Develop the physical layer artifacts for generic animal and herd networking as well as the purpose-built sensors and actuators for the dairy farm example CPS. Generic elements include an enhanced animal collar/mobility sensor and field-ready wireless access points. The purpose-built elements include an in-dwelling rumen sensor (developed separately, but this project is not dependent on its availability), automated feed delivery device, cow weight sensor, and in-line milk analysis equipment.Develop new knowledge based on data networks that link animal and herd data with increased efficiency, profitability and animal well-being. Link the physical, network, and decision layers of the CPS and deploy, test and validate on-farm in two separate, networked installations.
Project Methods
Goal 1.1: (White and Daniels) Develop a profile of naturally occurring fluctuations in efficiency of production for cattle using remotely sensed behavioral and physiological observations collected daily. In developing CPS to enhance precision of animal management and interventions to enhance productivity it is first essential to determine the natural variation that exists in the system and to assess the potential for altering animal efficiency based on sensor input. During the first year of the study, cohorts of cattle at Virginia Tech will be subjected to efficiency screening to determine their responsiveness to real-time sensed indicators. Cattle (n=96) will be split into 4 cohorts of 24 animals, transitioned to Calan gates over a 14 day period and fed a standard lactation total mixed ration for a 35 d data collection period. Sensors will be placed in the rumen (for temperature and pH), on the neck (for rumination time, duration, and interval), in the feed bunk (feeding frequency, duration and amount), on the ankle (for speed frequency, and duration of movement) and in the milking equipment (for amount, composition and quality of milk). Using noted sensors, cows will be continuously monitored over the experimental period and data analyzed to determine the composite predictors of overall animal productive efficiency as milk production per unit of feed consumed. The emphasis for on-body sensors will be on incorporating wireless capability to avoid intrusion. White and Daniels will work closely together with Priya and Voyles to identify COTS sensor nodes that can be modified to include low-power wireless transceivers and with Sundaram to perform system identification and observability assessment to modify data gathered, as needed.Goal 1.2 (Voyles, Priya, Chiu, Sundaram). The objective of this sub-goal is to develop the structure of the generic animal models that will form the basis of decisions for action. As part of the decision layer, an animal-model-based structure will provide the abstraction that allows the architecture to apply to multiple farms and farm types (cow, sheep, poultry, etc). To estimate the state of each animal, discrete-time Kalman filters are employed with separate models for metabolic health and nutrition. These will be investigated below, with application to the cow model based on data from the above sub-goal.Goal 2.1 (Voyles, Priya, Min, Sundaram). The objective of this sub-goal is to construct a wireless networking infrastructure to support real-time precision animal agriculture. The goal of the wireless networking system is to adapt to varying demands of urgency for both data and executable programs (analytics) as they are routed between the core and the edge nodes. The ultimate goal seems simple: to move data from various layers in the network from the edge to the core of a cloud-based data storage platform to enable real-time decisions regarding the productivity, quality, safety, and efficiency of the farm and its products. Yet, because of the complexity of living entities, it is not so simple and will involve a novel collection of hardware layers, routing protocols, and distributed decision support across the system to support both top-down and bottom-up decisions and control.Goal 2.2 (Voyles, Priya, Sundaram, Chiu). The objective of this sub-goal is to integrate data obtained with the active sensing nodes into the network layer of the Generic Precision Animal Agriculture CPS.Co-PI Priya is exploring various active sensors that will test the CPS network interface protocol for edge analytics embedded in a purpose-built active sensor. Active sensing is not limited to robots or in-animal devices, but represents purpose-built sensors that actively gather data of specific interest, rather than passive sensors, such as security cameras. These sensors require computational capability and reside at the edge of the network, hence a protocol for "edge analytics" is necessary to integrate them modularly and robustly into the Precision Animal Agriculture reference architecture. In Year 1 and 2, the focus will be on collecting data from the passive sensor node with pH, temperature, and oxygen sensing capability. He will rely on commercially available wireless micro-sensors for pH, temperature, O2, and liquid density as a long term goal of the project is to map heterogeneity of microbe populations and fermentation parameters.?Goal 3.1 (Priya, Voyles, Donkin, White, and Daniels). The objective of this sub-goal is to build the new feed dispenser robot that will be used in the dairies.The new feed dispenser machine will be an integral part of the sustainable dairy CPS for both Virginia Tech and Purdue. The prototype will be developed at Virginia Tech as a stationary machine located in the freestall barn but will reside in campus laboratory space for initial testing. Figure 5 shows the current setup at Virginia Tech's milking station that provides the baseline platform. Upon entry into the station, a gate will close behind the cow preventing her egress for a minimum of 1 min. The cow's RFID ear tag and/or RFID leg tag will be read by an RFID reader. This will initiate autonomous action by the feed dispenser system. In addition to RFID, proximity sensors such as (LV-MaxSonar®-EZTM series) will be utilized for animal detection in order to improve accuracy. The LV-MaxSonar-EZ detects objects from 0-inches to 254-inches and provides sonar range information from 6-inches out to 254-inches with 1-inch resolution. Depending on individual animal needs, the cow will be delivered a specified quantity of feed supplement to support lactation needs or remedy acidosis, or both. The feed dispenser pan will contain load cells for automatic weighing of any feed refusals; weighbacks of uneaten feed will be weighed and discarded as the animal leaves the station. The machine will be modeled after construction of automatic dairy cow feed stations (e.g., Lely Cosmix; Lely, The Netherlands). Further, to entice cows to enter the station, a grooming roller will be installed over the cow chute (e.g. Lely Luna; Lely, The Netherlands). These rollers are popular and objectively enjoyed by cows.Goal 3.2 (Priya, Voyles, Donkin). The objective of this sub-goal is to replicate the new feed dispenser robot at Purdue.This is primarily an integration task, but deserves a separate line item as the prototype will be evaluated for ease of re-application.Goal 4. (White, Daniels, and Donkin) Evaluate the CPS research by deploying on two heterogeneous test farms During the third year of the study, two cohorts of animals (n=36 each; one at Virginia Tech and one at Purdue) will be fed in Calan gates to measure individual feed efficiency. Cows will undergo a 7 d gate training period, a 35 d screening period, and two experimental periods (14 d diet adaptation followed by 35 d treatment). Cattle will be partitioned into two groups. During the first experimental period, group 1 will be fed a standard lactation total mixed ration and group 2 will be fed a basal TMR supplemented according to the CPS algorithm for her response category. During the second experimental period, the treatments applied to the two groups will be switched. This will be replicated at VT and Purdue to evaluate the stability of cow sorting algorithms and feeding strategies defined in Obj 1.1, across institutions. In addition to providing greater numbers of animals the use of two locations will help to eliminate any potential bias based on the effects of the system, environment, or animals at a single location.