Source: CORNELL UNIVERSITY submitted to
IMPROVING DAIRY COW HEALTH MONITORING AND MANAGEMENT USING AUTOMATED SENSORS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
TERMINATED
Funding Source
AFRI COMPETITIVE GRANT
Reporting Frequency
Annual
Accession No.
1012489
Grant No.
2017-67015-26772
Project No.
NYC-127566
Proposal No.
2016-09382
Multistate No.
(N/A)
Program Code
A1221
Project Start Date
Jul 1, 2017
Project End Date
Jun 30, 2023
Grant Year
2017
Project Director
Giordano, J.
Recipient Organization
CORNELL UNIVERSITY
(N/A)
ITHACA,NY 14853
Performing Department
CALS - Animal Science
Non Technical Summary
Health disorders, especially in the postpartum period, are a major cause of poor dairy cow welfare and reduced farm profitability. Consequences for cow well-being and performance vary with disorder and severity, but all reduce cow productivity, reproductive performance, and survivability to some extent. Despite recent advances in disease prevention through improved herd management and facilities, a high proportion of dairy cows present at least one event of a health disorder during lactation, with highest incidence during the first 30-60 d after parturition. Therefore, dairy farmers spend substantial time and resources to identify, treat, and care for sick cows. Unfortunately, clinical exams for health disorder detection are time-consuming, labor intensive, a burdensome complication for dairy farmers, and inherently subjective. Some of these issues are exacerbated for large farms in which evaluating all cows that need attention may not be possible. Consequently, the overarching goal of this research is to improve dairy cow health, productivity, and longevity while reducing labor costs of dairy farms through the adoption of automated health-monitoring technologies. To achieve these goals, the proposed research, will (1) gather a large, first-of-its-kind set of automatically collected sensor data and routinely collected cow data on a wide range of behavioral, physiological, and productivity parameters, (2) determine patterns of association between the measured parameters and states of cow health, (3) use data from multiple parameters to develop an accurate, user-friendly index for identifying dairy cows with health disorders, and (4) compare the real-world efficacy of the multiparameter index to that of indices based on individual separate parameters. Our central hypothesis is that (1) cows with health disorders display alterations in multiple behavioral, physiological, and productivity parameters that can be sensed using commercially available devices, (2) these data can be combined with routinely collected information to accurately determine cow health status, (3) disease identification accuracy can be improved by combining multiple parameters (vs. single-parameter identification), and (4) the type of the disorder affecting each cow can be estimated. The specific science objectives of this research are: (1) Determine characteristic patterns of behavioral, physiological, and productivity parameters during periods of disease and non-disease in lactating dairy cows, (2) Use machine learning algorithms to develop maximally accurate, one-dimensional Health Status Indices for identifying dairy cows suffering from health disorders, and (3) Evaluate the performance of the Health Status Indices under commercial farm conditions. We predict that accurate automated cow health monitoring can revolutionize herd health management through: (1) earlier, more accurate disease detection, (2) improved cow well-being, (3) reduced labor needs and cost, (4) indication of cow overall health without blood or bodily fluids collection, and (5) improved public perception of animal care and welfare on dairy farms. Collectively, this research will favor industry-wide adoption of automated health monitoring contributing to greater food security for the US and to a globally sustainable dairy industry through increased cow health, farm profitability, and quality of life for dairy producers. Increased farm profitability via consumer-acceptable strategies will enable net gains to US economy and society.
Animal Health Component
100%
Research Effort Categories
Basic
50%
Applied
30%
Developmental
20%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
3113410102060%
4023499209040%
Knowledge Area
311 - Animal Diseases; 402 - Engineering Systems and Equipment;

Subject Of Investigation
3499 - Dairy cattle, general/other; 3410 - Dairy cattle, live animal;

Field Of Science
2090 - Statistics, econometrics, and biometrics; 1020 - Physiology;
Goals / Objectives
Our long-term goal is to improve dairy cow health, productivity, and longevity while reducing labor costs of dairy farms through the adoption of automated health-monitoring technologies. In the proposed research, we will (1) gather a large, first-of-its-kind set of automatically collected sensor data and routinely collected cow data on a wide range of behavioral, physiological, and productivity parameters, (2) determine patterns of association between the measured parameters and states of cow health, (3) use data from multiple parameters to develop an accurate, user-friendly index for identifying dairy cows with health disorders, and (4) compare the real-world efficacy of the multiparameter index to that of indices based on individual separate parameters. Our central hypothesis is that (1) cows with health disorders display alterations in multiple behavioral, physiological, and productivity parameters that can be sensed using commercially available devices, (2) these data can be combined with routinely collected information to accurately determine cow health status, (3) disease identification accuracy can be improved by combining multiple parameters (vs. single-parameter identification), and (4) the type of the disorder affecting each cow can be estimated.Objectives of this research are as follows:(1) Determine characteristic patterns of behavioral, physiological, and productivity parameters during periods of disease and non-disease in lactating dairy cows.Under Specific Objective 1-1 we will collect behavioral, physiological, and productivity data during periods of disease and non-disease in the early postpartum period of dairy cows using (a)attached and non-attached sensors and (b) routinely logged cow data. Under Specific Objective 1-2 we will determine parameter change corresponding to changes in cow health. Under Specific Objective 1-3 we will determine patterns of behavioral and physiological parameters collected during prepartum period; evaluate relationship of data to postpartum health disorders.(2)Use machine learning algorithms (MLAs) to develop maximally accurate, one-dimensional Health Status Indices (HSIs) for identifying dairy cows suffering from health disorders. Under Specific Objective 2-1 we will use raw data and MLAs to develop easy-to-use multiparameter and single parameter HSIs to identify cows suffering health disorders.(3)Evaluate the performance of the HSIs under commercial farm conditions.Under Specific Objective 3-1 we will (1) Evaluate and compare the performance of the HSIs as tools for the identification of cows with health disorders. (2) Evaluate timing of identification of cows by HSIs vs. by clinical diagnosis by research personnel.
Project Methods
AIM 1: Determine characteristic patterns of behavioral, physiological, and productivity parameters during periods of disease and non-disease in lactating dairy cows.Specific Objective 1-1. Collect behavioral, physiological, and productivity data during periods of disease and non-disease in early lactation of dairy cows using (a)attached and non-attached sensors and (b) routinely logged cow data. Research Methods. In this observational prospective cohort study, cows (n= 1,250) will be enrolled from -30 to 60 days in milk (DIM). At enrollment, cows will be fitted with multiple automated attached sensors and data will be collected from multiple non-attached sensors. Cow features and performance parameters available from dairy management software and environmental data will also be collected. Clinical Health Monitoring. From 1-30 DIM, all cows in the study will be examined daily. Sensor data collected will not be used to diagnose health disorders during the study. Blood Sample Collection. Blood samples will be collected from a subgroup of cows (n = 500) at various and on the day of diagnosis of disease to determine circulating concentrations of markers of metabolic, mineral, and systemic inflammation status. Expected Outcomes. Upon completion of SO 1-1, pattern analysis of this data will be conducted under SO 1-2. In Aim 2, the data will be employed in creating user-friendly health status indices (HIS's) based on single or multiple parameters.Specific Objective 1-2. Determine parameter change corresponding to changes in cow health. Outcomes of Interest. The pattern of each parameter of interest relative to postpartum episodes of clinical health disorders and during periods of non-disease. Environmental, management, and non-sensor cow data will be used to determine the association of these data with cow sensor data during disease and non-disease. Data from these factors will also be integrated into the indices created in Aim 2. Data Analysis. At the group level, two types of analyses will be conducted, (1)comparisons of cows affected vs. not affected by disorders and (2)subgroups of cows affected by the same disorders. Expected Outcomes. Measurements from multiple sensors and detailed and consistent evaluations of cow health will enable an in-depth analysis of patterns in multiple markers of health during periods of disease and non-disease. Understanding these patterns is critical to developing tools and strategies for identifying cows with health disorders, characterizing patterns for healthy cows, and develop group management strategies directly applicable to dairy management.Specific Objective 1-3: Determine patterns of behavioral and physiological parameters collected during prepartum period; evaluate relationship of data to postpartum health disorders. Research Methods. We will use data collected during the prepartum period and up to 48 h after calving for the study of SO 1-1. Outcomes of Interest. Patterns of parameters monitored via sensors from 30 d before calving up to the end of the day of calving. Data analysis. Same as for SO 1-2, except that the period of interest will be from enrollment until 24 h before calving, or the 24 h before and 48 h after calving. Cows that develop or do not develop a health disorder in the postpartum period will be compared.AIM 2: Use machine learning algorithms (MLAs) to develop maximally accurate, one-dimensional Health Status Indices (HSIs) for identifying dairy cows suffering from health disorders. Specific Objective 2-1: Use raw data and MLAs to develop easy-to-use multiparameter and single parameter HSIs to identify cows suffering health disorders. Methods. We will use data collected under SO 1-1, Aim 1 to create HSIs through machine learning algorithms. Specifically, each HSI will represent the predicted probability that a cow is suffering a health disorder. Outcomes of Interest. The main outcome of interest will be creation and accuracy of the indices created. Data Analysis. Validation of HSIs. Once the HSIs are developed, we will identify cutoff points to trigger the alarm indicating that a cow should be examined. Overall Performance. Specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy will be calculated as an overall test rather than for individual disorders. Expected Outcomes. We expect Se and Sp of the indices to be ≥ 95%. Another relevant outcome of this research will be development of a novel, open-source methodology to create indices. This methodology can be used by developers, e.g., companies that provide sensors to favor the adoption and facilitate the use of AHMS in dairy farms. This research will also be useful to researchers working on development and testing of new and existing sensors.AIM 3: Evaluate the performance of the HSIs under commercial farm conditions. Demonstrate the superiority of the multiparameter HSI. Specific Objective 3-1. (1) Evaluate and compare the performance of the indices created as tools for the identification of cows with health disorders. (2) Evaluate timing of identification of cows with health disorders. Methods. This study will follow an observational cohort design and use the same procedures as for the study conducted for SO 1-1. Cows (n = 500) will be enrolled from - 30 to 60 DIM. The same sensor and non-sensor data will be collected as were used to create the various indices. Outcomes of Interest. The ability of the various indices to correctly identify cows with and without disorders. The timing of the first HSIs positive alarm in relationship to the time of clinical health disorder detection will also be of interest. Data Analysis. Estimation of the Sp, PPV, NPV, and accuracy of the indices will be calculated. The difference in days between the timing of the first alert fromthe indices and the day of clinical health disorder detection will be compared. Expected Results. We expect that combining multiple sensor-generated parameters will improve accuracy because of how different disorders affect different parameters or the same parameters but to a different extent.

Progress 07/01/17 to 06/30/23

Outputs
Target Audience:Our target audience has been large and diverse. We reached out to and interacted directly or indirectly with thousands of stakeholders through presentations in scientific, producer, veterinarian, and consultant meetings. Data generated through this project was published in scientific Journals and conference publications. The PD worked directly with commercial farms through visits and consultations. Information and ideas for implementation of novel management strategies based on automated health monitoring systems were shared with farm personnel, producers, veterinarians, and several other multipliers of information. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Graduate and undergraduate students: Through this project the PD and Co-PD's worked directly on training eight graduate students (6 PhD and 2 MS) on multiple aspects of experimental design and research with automated health, behavior, and performance monitoring systems as well as with machine learning and data integration. Undergraduate students (n = 5) have also been involved with project implementation providing support to the lead graduate students. Animal science graduate students gained experience working with multiple sensors systems and their software for data collection for the experiments conducted as part of this project. Undergraduate students in Animal Science have been trained on laboratory techniques for estimation of markers of cow metabolic, physiological, and inflammation status and have also been trained in large dataset processing and management. Students from Computer Science worked on the development of machine learning algorithms using data generated for animals enrolled in this study and others that provide similar types of data and outcomes. Through another collaboration established with Faculty in Computer Science, graduate students (2 PhD) have been trained on thedevelopment of software infrastructure for data integration in real time from multiple sources of dairy farm sensor and nonsensor data. This interdisciplinary project resulted in cross training of animal science students on aspects of computer science and engineering and training of computer science students on aspects of animal research. Students developed and strengthened their professional skills through presentations and participation in scientific, industry, and extension conferences and events. Students engaged in regular meetings and discussions with dairy industry stakeholders including collaborating industry partners (i.e., pharmaceutical companies, precision livestock providers, dairy herd management software providers, others), veterinarians, extension personnel, dairy producers, and scientists from Cornell and other institutions. These activities expanded students and other personnel's professional networks. Technicians and other professionals: Two laboratory technicians and eight research associates with expertise in animal science and veterinary medicine were trained through this project. This personnel participated of planning and execution of on-farm and laboratory activities and experiments. Three computer programmers from the Center for Advanced Computing (CAC) at Cornell University participated of this project through the creation of a data integration software infrastructure and on-farm deployment of machine learning algorithms. Research collaborations: research collaborations with three different data science groups have been developed through this project. These collaborations led to several collaborative publications and presentations at scientific conferences. These collaborations also led to additional federal grants and development of course curricula in digital agriculture. How have the results been disseminated to communities of interest?Findings and results of this research project have been delivered to the scientific community through peer-reviewed manuscripts, abstracts, and presentations at research conferences. Findings from this research project have also been presented to dairy industry stakeholders including dairy producers, dairy consultants, and allied industry personnel at conferences, workshops, and informal meetings. Undergraduate students in Animal Science, Veterinary Medicine students, and students in Computer Science have received instruction and worked in digital agriculture related projects with the help of infrastructure and data generated by this project. Dairy producers and farm personnel: presentations were delivered in formal and informal settings to dairy producers nationally and internationally. The PD and graduate students presented results and discussed implementation of herd health automated monitoring and management strategies and technologies at several state, national, international conferences, and meetings. The PD worked directly with 20+ dairy farms of all sizes to help them establish herd health monitoring and management programs and with adoption of automated monitoring technologies based on the results of this project. Scientific community: results of this project were presented to thousands of scientists through more than 50 abstracts, proceedings, posters, and oral presentations in local, national, and international conferences, meetings, and workshops. Novel information generated through this project activities have been used by other scientists to formulate novel hypotheses, design experiments, design new management strategies for dairy cattle, and improve understanding of some of their own research findings. Industry consultants and extension educators: veterinarians, consultants from allied industries, and extension educators from Cornell Cooperative Extension and PRO-Dairy gained new knowledge through +50 presentations at conferences, workshops, and extension meetings for consultants and educators. This is an important audience because they are multipliers of information working on a daily basis with numerous producers representing farms of all sizes and types. This stakeholder group also benefited directly from this project as the information generated can be used for improving their services to their clientele. Dairy industry: businesses such as veterinary clinics, dairy farm consulting companies, AI companies, providers of technology for dairy farms, and businesses in rural areas benefited directly and indirectly. These stakeholders use the knowledge and management strategies generated through this research to advise their dairy farm clientele. Moreover, results of this research contribute to increase the use of services, products, and technologies provided by these stakeholders and businesses to commercial dairy farms. General public: the general public benefited from this project's activities in multiple ways. The most direct benefits for consumers of dairy foods were through improved dairy herd performance and farm economics as both factors increase availability and affordability of dairy products. Increases in dairy production efficiencies also benefit the general population by reducing the use of natural resources and environmental impacts of food production. What do you plan to do during the next reporting period to accomplish the goals?We will continue to work on publishing data in peer reviewed journals and disseminate our results to our target audiences.

Impacts
What was accomplished under these goals? This project contributed to improving the sustainability of dairy farms through the development and implementation of cow health monitoring strategies that incorporate automated monitoring technologies. Novel herd health monitoring programs based on real time health status alerts that combine data from wearable and non-wearable sensors were developed and tested in dairy farms. This project demonstrated that cows affected with common health disorders that reduce lactation performance, survivability, and well-being of cows can be promptly and accurately identified through changes in behavioral, physiological, and performance parameters monitored by precision technologies. Knowledge, data-driven tools, and management strategies generated by this project have directly benefited major dairy industry stakeholders. Dairy producers, consultants, and other dairy industry professionals adjusted dairy herd management practices through the adoption and improved use of precision cow health monitoring technologies. Novel methods to develop, test, and validate simple and complex health monitoring alerts developed with machine learning and non-machine learning methods using automatically integrated data generated were made available to the scientific community and providers of dairy technology. Through the development and refinement of automated health monitoring strategies and promoting adoption of precision technologies, this project contributed to improve dairy herd management practices that enhance dairy cow productivity and health and can help dairy farms mitigate increasing labor challenges. Under objective (1) a series of observational prospective cohort studies were conducted at commercial farms to determine characteristic patterns of behavioral, physiological, and productivity parameters during periods of disease and non-disease in lactating dairy cows. Cows (n=~4,500) were fitted with neck-, leg-, and ear-attached sensor tags and ruminal sensors to monitor behavioral and physiological parameters including rumination, eating behavior, body temperature, physical activity, and resting behavior. Performance parameters including milk volume, milk components, and milk conductivity were collected automatically at every milking using parlor sensors. Herd management practices such as feeding, pen moves along with environmental conditions were recorded automatically. Previous and current lactation health and reproductive events andseveral other performance records were collected. Cow health status was monitored daily through clinical examination procedures to identify cows with health disorders, determine the day of the onset of clinical signs of disease, define the type of health disorder affecting cows, determine the severity of health disorders diagnosed, and determine time to resolution of clinical signs of disease. Cows affected by health disorders manifested alterations of the trajectory of sensor-monitored parameters around clinical diagnosis of disease of sufficient magnitude to be detected through visual inspection of data or algorithms developed using machine learning and non-machine learning techniques. For example, cows presented different trajectories of the pattern of rumination time, eating time, physical activity, body temperature, lying time, milk volume, milk components, and milk conductivity immediately before, during, and after clinical diagnosis of health disorders. Depending on the type and severity of health disorders, the magnitude, and temporal dynamics of changes for sensor-monitored parameters were of sufficient magnitude to be detectable before clinical manifestation of disease. Detection of these changes enabled automated health monitoring systems to flag cows for clinical examination immediately before or at the time of clinical manifestation of signs of disease. Cows were more easily identified through automated alerts if affected systemically or concomitantly by another health disorder. Strong associations between the degree of alteration of the pattern of parameters monitored by sensors and the clinical status of cows were observed. Specifically, certain behaviors and physiological parameters were affected considerably more in cows with multiple and potentially more severe clinical signs of disease. Therefore, automated health monitoring systems that monitor more than one parameter to generate health alerts might be more effective for identifying cows affected by health disorders that cause more severe alterations to cow behavior, physiology, and performance or cows that manifest multiple and more severe clinical signs of disease. Collectively, the observed temporal shifts of sensor data patterns and the magnitude and timing of changes were useful for automatically screening dairy cow health. Under objective (2) we used data collected in the studies conducted under objective (1) for developing machine learning (ML) algorithms for accurately identifying dairy cows suffering from health disorders. Several ML techniques such as XGBoost (XGB), Multi-Layer Perceptron (MLP), Recurrent Neural Networks (RNN), decision trees, and logistic regression were evaluated for daily prediction of the health status of lactating dairy cows. Models were also built to predict the type of disorder affecting cows. After splitting datasets including +100 predictors from sensor and non-sensor data streams in subsets for training (80%) and testing (20%), hundreds of models were created in Phyton using a wide range of model hyperparameters specific to each MLA technique. XGBoost algorithms had the best predictive ability for identifying cows with health disorders and specific types of disorders such as metabolic-digestive, uterine disease, and mastitis. Overall, the most predictive models for all disorders combined had sensitivity (85-90%), specificity (80-90%), precision (35-45%), negative predictive values (90- 99%), and F1 scores (0.40-0.60) within a range that warrants value for farm implementation. Dairy farms that implement these models could expect to observe good sensitivity to identify cows with health disorders at the expense of more clinical exams in healthy cows due to relatively low precision. Models for predicting specific types of health disorders had acceptable performance depending on the disorder, but overall underperformed compared with models predicting all disorders combined. Under objective (3) evaluate the performance of cow health alerts using health status indexes under commercial farm conditions, a series of observational and randomized controlled experiments were conducted to validate health alerts created with data from objective (1) and (2) and to determine the value of automated health monitoring strategies using precision technologies. In some experiments a custom-built automated real-time data aggregator software infrastructure to integrate data from all sensor and non-sensor data streams for running ML algorithms was used. In other experiments, commercially available automated health alerts were sued. Results from these experiments demonstrated that health monitoring strategies that rely primarily on automated health alerts can be as effective as intensive traditional health monitoring programs that rely heavily on extensive clinical examination of cows or can be more effective and economically beneficial compared with non-intensive health monitoring based on visual observation only. Collectively, results of this first of its kind research project demonstrated that dairy farms can successfully implement health monitoring programs that rely primarily on automated health monitoring systems data. The potential to improve health monitoring and management of dairy cows using automated health monitoring systems was demonstrated.

Publications

  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Perez M. M., E. M. Cabrera, J. O. Giordano*. 2023. Effect of targeted clinical examination based on alerts from automated health monitoring systems on herd health and performance of lactating dairy cows. J. Dairy Sci. 106(12):9474-9493. https://doi:10.3168/jds.2023-23477
  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Rial C., A. Laplacette, L. Caixeta, C. Florentino, F. Pe�a-Mosca, J. O. Giordano*. 2023. Metabolic-digestive clinical disorders of lactating dairy cows were associated with alterations of rumination, physical activity, and lying behavior monitored by an ear-attached sensor. J. Dairy Sci. 106(12):9323-9344 https://doi:10.3168/jds.2022-23156
  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Rial C., A. Laplacette, L. Caixeta, C. Florentino, F. Pe�a-Mosca, J. O. Giordano*. 2023. Metritis and mastitis events in lactating dairy cows were associated with alterations of the pattern of rumination, physical activity, and lying behavior monitored by an ear-attached sensor. J. Dairy Sci. 106(12):9345-936 https://doi:10.3168/jds.2022-23157
  • Type: Journal Articles Status: Awaiting Publication Year Published: 2023 Citation: Garcia, N. L., Rodrigues-Motta, M., Migon, H. S., Petkova E., Tarpey, T., Todd Ogden, R., Giordano J.O. and M. M. Perez.2023. Unsupervised Bayesian classification for models with scalar and functional covariates. Journal of the Royal Statistical Society: Series C. (Accepted in proof)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Rubambiza, G., Chin, S. W., Rehman, M., Atapattu, S., Mart�nez, J. F., & Weatherspoon, H. (2023, July). Comosum: An Extensible, Reconfigurable, and Fault-Tolerant IoT Platform for Digital Agriculture. In USENIX Annual Technical Conference.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Prediction of pregnancy in lactating dairy cows with machine learning algorithms using behavioral, physiological, and performance sensor data and other cow, herd, and environmental data. G. E. Granados*, M. M. Perez, and J. O. Giordano. J. Dairy Sci. Volume 105, Suppl. 1 (Abstract)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J. O. Integration of automated monitoring for improving health monitoring and management of dairy cattle. International Congress on Herd Health and Management. Antalya, Turkey. October 23rd, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Giordano J.O. Using sensor technologies to monitor dairy cattle health. Operations Managers Conference. Syracuse, NY. January 31st, 2023
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Giordano J.O. Integrating automated monitoring in herd health and reproductive monitoring and management. Technology Tuesdays. . Cornell PRO-Dairy. (Delivered Virtually) January 10th, 2023.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J.O. Automated monitoring sensor systems and their integration in dairy herd management. Penn State 2022 Dairy Nutrition Workshop. Hershey, PA. November 2nd, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J.O. Harnessing automated monitoring systems data for herd health and reproductive management. Northeast ADSA Annual Meeting. Syracuse, NY. September 19th, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J.O. Integrating automated monitoring technologies for improving dairy reproductive management. The Dairy Signal TM. Professional Dairy Producers of Wisconsin. (Delivered Virtually) August 11th, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J.O. 1)Automated health monitoring of dairy cattle. 2) Reproductive management of lactating dairy cattle. 3) Monitoring reproductive programs. Alta Dairy School. Garden City, KS. May 12th, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J.O. Integrating automated monitoring technologies in dairy herd management. Phibro dairy skills academy. Auburn, NY. April 19th, 2022. Workshop and training session.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J.O. Finding the value of automated monitoring systems for dairy herd management. Northeast Dairy Production Medicine Symposium. Syracuse, NY. March 13th, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Giordano J.O. Challenges and opportunities with the use of automated systems for monitoring estrus and health of dairy cattle. Event for veterinary practitioners and dairy producers. Organized by the National University of La Pampa. Villa Maria, Cordoba, Argentina. August 10th, 2023.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Giordano J.O. 1) Integrating cow biology and technology for designing effective reproductive management programs. 2) Improving herd health monitoring and management using data-driven automated technologies. ICARE for Cows Conference. Delivered to Canadian Bovine Veterinarians. Organized by Solvet, Canada. Buenos Aires, Argentina. August 11th, 2023.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Giordano J.O. Using the power of automated monitoring technologies for improving herd reproductive management and performance. Future proofed dairy conference. Organized by Smaxtec, Austria. Graz, Austria. July 19th, 2023.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Giordano J. O. Integrating automated monitoring technologies in dairy herd management. Polo Formazzione Maccarese. (Delivered Virtually) September 19th, 2022.
  • Type: Theses/Dissertations Status: Published Year Published: 2023 Citation: Rial Clara. Improving dairy herd health and reproduction with data-driven tools and management strategies. Thesis. Cornell University. December, 23rd 2023
  • Type: Theses/Dissertations Status: Under Review Year Published: 2024 Citation: Perez Martin. Improving Dairy Cow Health Monitoring and Management Using Automated Sensors. Thesis. Cornell University. March, 2024
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Rial C., Stangaferro M. L., Thomas M. J. and Giordano J. O.* 2023. Association between vaginal discharge scores with rumination time, activity time, a health index score, and milk yield in lactating dairy cows. J. Dairy Sci. 106 (E-Suppl.1)(Abstract).
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Giordano J. O. Improving dairy herd monitoring and management using automated monitoring technologies. Italian Association of Animal Science Annual Meeting. Bari, Italy. June, 14th 2023.


Progress 07/01/21 to 06/30/22

Outputs
Target Audience:Members of the scientific community interested on this area of research. These include researchers at Universities, graduate students and postdocs, and researchers in industry. Dairy cattle producers and allied industry personnel because the results of this research have direct implications for dairy operations. Multiple industry providers of precision livestock technology. The PI and Co-PIs work regularly on disseminating research results to different stakeholder groups. Changes/Problems:Nothing major to report other than some continued delays to conduct the field experiment under Specific Objective 3 due to the complexity and tedious work required to complete the data integration and data analytics component of this project. We continue to make significant progress but at a slower pace than expected. What opportunities for training and professional development has the project provided?During this reporting period, five graduate students (4 PhD and 1 MS) have been trained in multiple aspects of experimental design and research with automated health, behavior, and performance monitoring systems as well as with machine learning and data integration. Other graduate (n = 1) and undergraduate students (n = 2) have been involved with project implementation providing support to the lead graduate students. Animal science graduate students gained experience working with multiple sensors systems and their software for data collection for the experiments conducted as part of this project. Students also learned to manage and work with the multiple software systems from the sensor and non-sensor technologies used for the study. Undergraduate students in Animal Science have been trained on laboratory techniques for estimation of markers of cow metabolic, physiological, and inflammation status and have also been trained in large dataset processing and management. Students from Computer Science worked on the development of machine learning algorithms using data generated for animals enrolled in this study and others that provide similar type of data and outcomes. Through another collaboration established with Faculty in Computer Science, graduate students (3 PhD) have been trained on the development of software infrastructure for data integration in real time from multiple sources of dairy farm sensor and nonsensor data. Therefore, this interdisciplinary project resulted in cross training of animal science students on aspects of computer science and engineering and training of computer science students on aspects of animal research. One laboratory technician has also been involved with project planning and execution. The PI and both Co- PI's worked directly with students on their research and training. Research collaborations - research collaborations with three different data science groups have been developed through this project. Our groups meet on a regular basis to evaluate progress and plan research activities for the short and long-term. These collaborations led to several abstract publications and presentation at scientific conferences. We are currently working on submission of full manuscripts for peer reviewed Journals and peer reviewed conference papers (Computer Science). How have the results been disseminated to communities of interest?Findings of this research project have been delivered to the scientific community through peer-reviewed abstracts and presentations at research conferences. Findings from this research project have also been presented to dairy industry stakeholders including dairy producers, dairy consultants, and allied industry personnel at conferences, workshops, and informal meetings. Undergraduate students in Animal Science and Computer Science have received instruction and worked in digital agriculture related projects with the help of infrastructure and data generated by this project. What do you plan to do during the next reporting period to accomplish the goals?We expect to conduct the last experiment under Specific Objective 3. We expect the field experiment will require 8 to 9 months to be completed. Thereafter, we will need approximately 6 more months to complete data analysis, submit our final report, and draft manuscripts for publication.

Impacts
What was accomplished under these goals? During this reporting period, activities continued to focusprimarily on data analytics, data integration, and exploration of associations between biological outcomes and sensor-generated data. Data analytics: We continued to explore different methods to generate Health Status Indices (HIS) based on Machine Learning Algorithms (MLA) and non-machine learning algorithms. We explored many different approaches to pre-process data for MLA model training and optimization. Different MLA techniques were used to generate and compare the performance of HIS. In this reporting period, we also explored development of algorithms for prediction of specific types of health disorders. As expected, different performance levels were observed for different algorithms and different types of disorders affecting dairy cows in early lactation. Data integration: The hardware and software infrastructure developed for real-time integration of data streams from multiple sources of sensor and non-sensor generated data was expanded, refined, and tested. A centralized database stored at Cornell University servers (i.e., RedCloud) accumulates in real-time data collected at a commercial dairy farm from in-line milk sensors, cow wearable sensors, walk-in scales, infrared cameras, environmental conditions sensors systems, dairy herd management software, and feed management software. Data from this centralized database has been integrated with software tools containing the ML algorithms and non-ML algorithms for calculation of HIS. Alerts generated in real time through this infrastructure are now being generated and will be used to conduct the research under Objective 3 of this project. Biological outcomes and sensor data: We continued to explore and characterize patterns of behavioral, physiological, and performance parameters before and after the occurrence of health disorders and calving. Due to unique nature and quantity of data generated for this project, we have been exploring the use of these data for data analytics in other areas of dairy farm management. For example, we have explored the use of the sensor and non-sensor data collected for development of HSI for prediction of the reproductive performance of cows. We have explored the association among the multiple sensor and non-sensor data collected with the probability of pregnancy at first insemination postpartum and initiated the development of MLA for prediction of pregnancy before insemination. Ultimately, the infrastructure created for health monitoring could be used to improve decision-making for reproductive management.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Using automated health monitoring systems to improve health monitoring and management of dairy cows. Giordano, J.O. Academy of Dairy Veterinary Consultants. Albuquerque, NM. September 25th, 2021.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Using automated health monitoring systems to improve health monitoring and management of transition dairy cows. Giordano, J.O., M.M. Perez, C. Rial, M.L. Stangaferro, and A.L. Laplacette. Joint ADSA and NMC Transition cow symposium. American Dairy Science Association Annual Meeting. July 12th, 2021.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Finding the value of automated monitoring systems for dairy herd management. Northeast Dairy Production Medicine Symposium. Giordano J.O. Syracuse, NY. March 13th, 2022.
  • Type: Journal Articles Status: Submitted Year Published: 2021 Citation: Bayesian latent class model with scalar and functional covariates. Garcia, N.L, M. Rodrigues-Motta, H.S. Migon, E. Petkova, T. Tarpey, R. Todd Ogden, J.O. Giordano, and M.M. Perez. Journal of Royal Statistical Society: Series C. (Submitted)


Progress 07/01/20 to 06/30/21

Outputs
Target Audience:Members of the scientific community interested on this area of research. These include researchers at Universities, graduate students and postdocs, and researchers in industry. Dairy cattle producers and allied industry personnel because the results of this research have direct implications for dairy operations. Multiple indsutry providers of precision livestock technology. The PI and Co-PIs work regularly on disseminating research results to different stakeholder groups. Changes/Problems:The project continues to progress as expected, however, progress has been slower than initially planned for some areas. We have faced challenges due to the COVID pandemic, supply chain disruptions, changes in hardware and software for the commercially available technologies being used, and some disruptions due to management issues beyond our control at the commercial farm where the field component of this research is being conducted. We have worked diligently on addressing these challenges and continued to make steady progress toward project completion. Ultimately, the greatest consequence has been a delay in accomplishing some of our objectives. Another contributor to our delay has been the exploration of novel methods for data analytics and additional ways to analyze the vast data collected. What opportunities for training and professional development has the project provided?During this reporting period, five graduate students (4PhD and 1 MS) have been trained in multiple aspects of experimental design and research with automated health, behavior, and performance monitoring systems as well as with machine learning and data integration. Other graduate (n = 2) and undergraduate students (n = 3) have been involved with project implementation providing support to the lead graduate students. Animal science graduate students gained experience working with multiple sensors systems and their software for data collection for the experiments conducted as part of this project. Students also learned to manage and work with the multiple software systems from the sensor and non-sensor technologies used for the study. Undergraduate students in Animal Science have been trained onlaboratory techniques for estimation of markers of cowmetabolic, physiological, and inflammation status and have also been trained in large dataset processing and management.Students fromComputer Scienceworkedon the development ofmachine learning algorithms using data generated for animals enrolled in this study and others that provide similar type ofdata and outcomes. Through another collaboration established with Faculty in Computer Science, graduate students (3 PhD) have been trained onthe development of software infrastructure for data integration in real time from multiple sources of dairy farm sensor and non-sensor data. Therefore, this interdisciplinary project resulted in cross training of animal science students on aspects of computerscience and engineering and training of computer science students on aspects of animal research. One laboratory technicianhas also been involved with project planning and execution. The PI and both Co- PI's worked directly with students on theirresearch and training. Research collaborations - researchcollaborations with three different data science groupshave been developed through this project.Our groups meet on a regular basis to evaluate progress and plan research activities for the short and long-term. These collaborations led to several abstract publications and presentation at scientific conferences. We are currently working on submission of full manuscripts for peer reviewed Journals and peer reviewed conference papers (Computer Science). How have the results been disseminated to communities of interest?Findings of this research project have been delivered to the scientific community through peer-reviewed abstracts and presentations at research conferences. Findings from this research project have also been presented to dairy industry stakeholders including dairy producers, dairy consultants, and allied industry personnel at conferences, workshops, and informal meetings. Undergraduate students in Animal Science and Computer Science have received instruction and worked in digital agriculture related projects with the help of infrastructure and data generated by this project. What do you plan to do during the next reporting period to accomplish the goals?During the next reporting period we will focus on completing the activities of Objective 3of the project. The hardware and software infrastructure for real time generation of health alerts will be implemented, tested under commercial farm conditions, and refined as needed. A prospective cohort study (describedunder Objective 3 of the project) will be conducted to test at a commercial farm the health alerts generated. Other studies will be conducted to further characterize the patterns of behavioral, physiological, and performance parameters during states of health and disease during different stages of the lactation cycle in cows. We plan on submitting research results for publication, disseminating our findings through multiple channels to reach all interested stakeholders, and plan additional research and educational activities that may stem from this project. We will also continue to build and strengthen our research collaborations.

Impacts
What was accomplished under these goals? During this reporting period, activities focused primarily on data analytics, data integration, and exploration of the association between biological outcomes and sensor-generated parameter data. Data analytics: We continued to exploredifferent methods to generate Health Status Indices (HIS) based on Machine Learning Algorithms (MLA) and non-machine learning algorithms. We explored many different approaches to pre-process data for MLA model training and optimization. Different MLA techniques were used to generate and compare the performance of HIS. In this reporting period, we also explored development of algorithms for prediction of specific types of health disorders. As expected, different performance levels were observed for different algorithms and different types of disorders affecting dairy cows in early lactation. Data integration: Thehardware and software infrastructure developed for real-time integration of data streams from multiple sources of sensor and non-sensor generated data was expanded, refined, and tested. A centralized database stored at Cornell University servers (i.e., RedCloud) accumulates in real-time data collected at a commercial dairy farm from in-line milk sensors, cow wearable sensors, walk-in scales, infrared cameras, environmental conditions sensors systems, dairy herd management software, and feed management software. Data from this centralized database has been integrated with software tools containing the ML algorithms and non-MLalgorithms for calculation of HIS. Alerts generated in real time through this infrastructure are now being generated and will be used to conduct the research under Objective 3 of this project. Biological outcomes and sensor data: We continued to exploreand characterizepatterns of behavioral, physiological, and performance parameters before and after the occurrence of health disorders and calving. Due to unique nature and quantity of data generated for this project, we have been exploring the use of these data for data analytics in other areas of dairy farm management. For example, we have explored the use of the sensor and non-sensor data collected for development of HSI for prediction of the reproductive performance of cows. We have explored the association among the multiple sensor and non-sensor data collected with the probability of pregnancy at first insemination postpartum and initiated the development of MLA for prediction of pregnancy before insemination. Ultimately, the infrastructure created for health monitoring could be used toimprove decision-making for reproductive management.

Publications

  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Perez M.M., C. Rial, and J. O. Giordano. Reticulo-ruminal temperature in the peripartum period was associated with the occurrence of health disorders after calving in dairy cows. J. Dairy Sci. Vol. 104, Suppl. 1. (Abstract)
  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Perez M. M., W. Song, Y. Yang, M. Liu, K. P. Birman, and J. O. Giordano. Cloud-based artificial intelligence infrastructure for automated real-time integration of dairy data in support of predictive analytics. J. Dairy Sci. Vol. 104, Suppl. 1. (Abstract)
  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Giordano J.O., M. M. Perez, C. Rial, M. L. Stangaferro, and A. L. Laplacette. Using automated health monitoring systems to improve health monitoring and management of transition dairy cows. J. Dairy Sci. Vol. 104, Suppl. 1. (Abstract)
  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Giordano J.O. Use of multiple biological, management, and performance data for the design of targeted reproductive management strategies for dairy cows. J. Dairy Sci. Vol. 104, Suppl. 1. (Abstract)
  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Rial C., A. L. Laplacette, M. M. Perez, C. C. Florentino, F. Pena-Mosca, L. Caixeta, and J. O. Giordano. Pattern of rumination time and physical activity captured by an ear-attached sensor around the time of clinical diagnosis of metritis and mastitis in dairy cows. J. Dairy Sci. Vol. 104, Suppl. 1. (Abstract)
  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Rial R., A. L. Laplacette, M. M. Perez, C. C. Florentino, F. Pena- Mosca, L. Caixeta, and J. O. Giordano. Pattern of rumination time and physical activity captured by an ear-attached sensor before, during, and after the clinical diagnosis of metabolic-digestive disorders in lactating dairy cows. J. Dairy Sci. Vol. 104, Suppl. 1. (Abstract)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Using automated health monitoring systems to improve health monitoring and management of transition dairy cows. Julio Giordano (presenter), Martin Perez, Clara Rial, Matias Stangaferro, Ana Laplacette. Presentation at the Joint Animal Health & National Mastitis Council Symposium: Management Strategies to Enhance Health of Dairy Cows during the Transition Period. Annual Meeting of the American Dairy Science Association. Delivered online.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Use of multiple biological, management, and performance data for the design of targeted reproductive management strategies for dairy cows. Julio Giordano (presenter), Emily Sitko, Clara Rial, German Granados, Martin Perez. Invited presentation at the Reproduction Symposium II: Prebreeding Predictors of Fertility. 2021 Annual Meeting of the American Dairy Science Association. Delivered online.


Progress 07/01/19 to 06/30/20

Outputs
Target Audience:Members of the scientific community interested on this area of research. These include researchers at Universities, graduate students and postdocs, and researchers in industry. Dairy cattle producers and allied industry personnel becausethe results of this research have direct implications for dairy operations. Multiple companies providers ofprecision livestock technology. The PI and Co-PIs work regularly on disseminating research results to different stakeholder groups. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?During this reporting period, fivegraduate students (4PhD and 1 MS) have been trained in multiple aspects of experimental design and research with automated health, behavior, and performance monitoring systems as well as with machine learning and data integration. Other graduate (n = 2) and undergraduate students (n = 3) have been involved with project implementation providing support to the lead graduate students. Graduate students and postdoc gained experience working withmultiple sensors systems and their software for data collection for the experiments conducted as part of this project. Students also learned to manage and work with the multiple software systems from the sensorand non-sensor technologies used for the study. Students from the laboratory of the Computer Science Co-PI have been working on the development of machine learning algorithms using data generated for animals enrolled in this study and others that provide similar type of data and outcomes. This interdisciplinary project resulted in cross training of animal science students on aspects of computer science and engineering and training of computer science students on aspects of animal research. One laboratory technician hasalso been involved with project planning and execution. The PI and both Co- PI's worked directly with students on their research and training. Research collaborations - A robust collaboration between the PI and Co-PI's (Dr. Weinberger and Dr. Nydam) has been developed. Our groups meet on a regular basis to evaluate progress and plan research activities for the short and long-term. How have the results been disseminated to communities of interest?Findings of thisresearch project have been delivered to the scientific community through peer-reviewed abstracts and presentations at research conferences. Findings fromthis research project have also been presented to dairy industry stakeholders including dairy producers, dairy consultants, and allied industry personnel at conferences, workshops, and informal meetings. Undergraduate students in Animal Science and Computer Science have received instruction and worked in digital agriculture related projects with the help of infrastructure and data generated by this project. What do you plan to do during the next reporting period to accomplish the goals?During the next reporting period we will focus on completing the activities of Objective 2 and 3 of the project. The hardware and software infrastructure for real time generation of health alerts will be finalized. Once this is finalized, a prospective cohort study (describe underObjective 3 of the project) will be conducted to test at a commercial farm the health alerts generated. We will also be able to test the robustness and resilience of the data collection and processing infrastructure under the conditions of a commercial dairy farm. During the next reporting period we also plan on submitting research results for publication, disseminating our findings through multiple channels to reach all interested stakeholders, and plan additional research and educational activities that may stem from this project. We will also continue to build and strengthen our research collaborations.

Impacts
What was accomplished under these goals? During this reporting period, activities focused primarily on data analytics, data integration, and exploration of the association between biological outcomes and sensor-generated parameter data. Data analytics: Data collected under Objective 1 was processed in preparation for development and initial evaluation of Health Status Indices (HIS) based on Machine Learning Algorithms (MLA) and non-machine learning algorithms. We explored many different approaches to pre-process data for MLA model training and optimization. Different MLA techniques were used to generate and compare the performance of HIS. Data integration: A hardware and software infrastructure was developed for real-time integration of data streams from multiple sources of sensor and non-sensor generated data. A centralized database stored at Cornell University servers (i.e., RedCloud) accumulates in real-time data collected at a commercial dairy farm from in-line milk sensors, cow wearable sensors, walk-in scales, infrared cameras, environmental conditions sensors systems, dairy herd management software, and feed management software. This centralized database is being integrated with software tools containing the MLA and non-MLA algorithms for calculation of HIS. Alerts generated in real time through this infrastructure will be used to conduct the research under Objective 3 of this project. Biological outcomes and sensor data: We explored and characterized in detail patterns of behavioral, physiological, and performance parameters before and after the occurrence of health disorders and calving. Due to unique nature and quantity of data generated for this project, we have been exploring the use of these data for data analytics in other areas of dairy farm management. For example, we have explored the use of the sensor and non-sensor data collected for development of HSI for prediction of the reproductive performance of cows. We have explored the association among the multiple sensor and non-sensor data collected with the probability of pregnancy at first insemination postpartum and initiated the development of MLA for prediction of pregnancy before insemination. Ultimately, the infrastructure created for health monitoring could be used to improve decision-making for reproductive management.

Publications

  • Type: Journal Articles Status: Published Year Published: 2020 Citation: M. M. P�rez, E. M. Cabrera, C. Rial, I. Foddanu, and J. O. Giordano. 2020. Effect of metritis on the pattern of behavioral, physiological, and performance parameters monitored by sensors in dairy cows. J. Dairy Sci. Volume 103, E-Supplement 1.(Abstract)
  • Type: Journal Articles Status: Published Year Published: 2020 Citation: M. M. Perez, E. M. Cabrera, C. Rial, D. V. Nydam, and J. O. Giordano. 2020. Effect of metabolic and digestive disorders on patterns of behavioral, physiological, and performance parameters of lactating dairy cows. J. Dairy Sci. Volume 103, E-Supplement 1.(Abstract)
  • Type: Journal Articles Status: Published Year Published: 2020 Citation: M. M. P�rez, Y. You, Y. Wang, K. Q. Weinberger, D. V. Nydam, and J. O. Giordano. 2020. Performance of the machine learning method XGBoost for prediction of clinical health disorders in lactating dairy cows. J. Dairy Sci. Volume 103, E-Supplement 1.(Abstract)
  • Type: Journal Articles Status: Published Year Published: 2020 Citation: M. M. P�rez, Y. You, Y. Wang, K. Q. Weinberger, D. V. Nydam, and J. O. Giordano. 2020. Performance of different machine learning methods for prediction of the health status of lactating dairy cows. J. Dairy Sci. Volume 103, E-Supplement 1.(Abstract)
  • Type: Journal Articles Status: Published Year Published: 2020 Citation: M. M. P�rez, G. Rubambiza, B. Barker, H. Weatherspoon, and J. O. Giordano. 2020. Automated real-time integration of data from multiple sensors and nonsensor systems for prediction of dairy cow and herd status and performance. J. Dairy Sci. Volume 103, E-Supplement 1.(Abstract)
  • Type: Journal Articles Status: Published Year Published: 2020 Citation: M. M. P�rez, E. M. Cabrera, C. Rial, D. V. Nydam, and J. O. Giordano. 2020. Pattern of behavioral, physiological, and performance parameters before and after clinical diagnosis of mastitis. J. Dairy Sci. Volume 103, E-Supplement 1.(Abstract)
  • Type: Journal Articles Status: Published Year Published: 2020 Citation: G. E. Granados, M.M P�rez, and, J. O. Giordano. 2020. Pattern of Behavioral, Physiological, and Performance Parameters before Insemination in Dairy Cows that Become Pregnant or Not to First Service. J. Dairy Sci. Volume 103. E-Supplement 1.(Abstract)


Progress 07/01/18 to 06/30/19

Outputs
Target Audience:Initially other members of the scientific community interested on this area of research. Once this research is completed our target audience will also include dairy cattle producers and allied industry as the results of this research may impact their operations and or businesses. Multiple companies that provide precision livestock technology will also be interested on this work. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Graduate student and postdoc training - Through this project, sixgraduate students (5 PhD and 1 MS) and one Postdoctoral fellow from Animal and Computer Science have been trained in multiple aspects of experimental design and research with automated health, behavior, and performance monitoring systems as well as with machine learning. Other graduate (n = 2) and undergraduate students (n = 6) have been involved with project implementation providing support to the lead graduate students and postdoc. Graduate students and postdoc gained experience installing and setting up multiple sensors systems and their software for data collection for the experiments conducted as part of this project. Students also learned to manage and work with the multiple software systems from the sensors and non-sensor technologies used for the study. Students from the laboratory of the Computer Science Co-PI have been working on the development of databases and initial testing of machine learning algorithms using data generated for animals enrolled in this study and others that provide similar type of data and outcomes. This interdisciplinary project resulted in cross training of animal science students on aspects of computer science and engineering and training of computer science students on aspects of animal research. One field technician and a laboratory technician have also been involved with project planning and execution. The PI and both Co-PI's worked directly with students on their research and training. Research collaborations - A robust collaboration between the PI and Co-PI's (Dr. Weinberger and Dr. Nydam) has been developed. Our groups meet on a regular basis to evaluate progress and plan research activities for the short and long-term. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?During the next reporting period we will continue working processing and analyzing data collected under SO 1-1 to accomplish the goals of SO 1-2 and 1-3. Using all behavioral, physiological, productivity and environmental parameters collected we will continue our work to accomplish the goals of Objective 2 to develop the health status indices using machine learning and non-machine learning techniques. Students and personnel currently engaged in the project will continue with their training and research. Additional studies using data collected under Objective 1-3 are being planned and additional experiments to evaluate the value and on-farm implementation of automated health monitoring technologies are underway. As data become available manuscripts will be drafted and submitted for publication in peer reviewed Journals.

Impacts
What was accomplished under these goals? Objective1: Under Specific Objective 1-1 we completed data collection for this phase of the project using 1,250 dairy cows from a commercial farm. We collected data with multiple automated health monitoring systems including cow attached and non-attached sensors to monitor multiple behavioral, physiological, and productivity parameters. In addition, we collected environmental data from multiple sensors installed throughout barns and surrounding areas to monitor environmental conditions around cows. During the study period research personnel followed detailed standard operating procedures to determine the health status of cows. Data for occurrence, type, and timing of health disorders was recorded and will be used to characterize the shifts in parameters of interest around events of all health disorders of interest. We also develop new methods to download and merge sensor and non-sensor data generated by different automated monitoring systems, environmental sensors, and other non-sensor sources of data such as the dairy herd management software. A centralized database integrating all different types of data was created and evaluated for data quality assurance. We have initiated activities to accomplish the goals of Specific Objectives 1-2 and 1-3. All data collected under SO 1-1 was used for initial validation of sensor data and for exploration of associations between sensor data and health events. Work conducted under SO 1-2 and 1-3 will serve as the basis for the development of Health Status Indices with machine learning and non-machine learning methods under Objective 2. Initial programing steps and testing of machine learning algorithms was conducted using partial data collected under SO 1-1. Activities to accomplish the research proposed under Objective 3 will be conducted once the activities under Objective 2 are completed.

Publications


    Progress 07/01/17 to 06/30/18

    Outputs
    Target Audience: Nothing Reported Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?Through this project, four graduate students (3 PhD and 1 MS) and one Postdoctoral fellow from Animal and Computer Science have been trained in multiple aspects of experimental design and research with automated health, behavior, and performance monitoring systems as well as with machine learning. Other graduate (n = 2) and undergraduate students (n = 6) have been involved with project implementation providing support to the lead graduate students and postdoc. Graduate students and postdoc gained experience installing and setting up multiple sensors systems and their software for data collection for the experiments conducted as part of this project. Students also learned to manage and work with the multiple software systems from the sensors and non-sensor technologies used for the study. Students from the laboratory of the Computer Science Co-PI have been working on the development of databases and initial testing of machine learning algorithms using data generated for animals enrolled in this study and others that provide similar type of data and outcomes. This interdisciplinary project resulted in cross training of animal science students on aspects of computer science and engineering and training of computer science students on aspects of animal research. One field technician and a laboratory technician have also been involved with project planning and execution. The PI and both Co-PI's worked directly with students on their research and training. Research collaborations - A robust collaboration between the PI and Co-PI's (Dr. Weinberger and Dr. Nydam) has been developed. Our groups meet on a regular basis to evaluate progress and plan research activities for the short and long-term. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?During the next reporting period we will collect data to complete the first phase of Objective 1. Once all behavioral, physiological, productivity and environmental parameters are collected we will develop the health status indices using machine learning and non-machine learning techniques under Objective 2. Students and personnel currently engaged in the project will continue with their training and research. Additional studies using data collected under Objective 1-3 are being planned and additional experiments to evaluate the value and on-farm implementation of automated health monitoring technologies are underway. As data become available manuscripts will be drafted and submitted for publication in peer reviewed Journals.

    Impacts
    What was accomplished under these goals? Objective1: Under Specific Objective 1-1 we completed the installation of multiple automated health monitoring systems including cow attached and non-attached sensors to monitor multiple behavioral, physiological, and productivity parameters. In addition, environmental sensors have been installed throughout barns and surrounding areas to monitor environmental conditions around cows. Cows are being enrolled in the study before calving and monitored until the period of maximum likelihood of disease occurrence as planned. Standard operating procedures for health monitoring and treatment decisions have been develop to ensure proper disease detection conducted by research personnel. Data for occurrence, type, and timing of health disorders will be critical to characterize the shifts in parameters of interest around health disorders events. Methods to download data generated by different automated monitoring systems, environmental sensors, and other non-sensor data have been developed. A centralized database for integration of different types of data is under development. As more cows are enrolled in the study abundant data will become available to accomplish the goals of Specific Objectives 1-2 and 1-3 and serve as the basis for the development of Health Status Indices with machine learning and non-machine learning methods under Objective 2. Activities to accomplish the research proposed under Objective 3 will be conducted once the activities under Objective 2 are completed.

    Publications