Performing Department
(N/A)
Non Technical Summary
There is growing interest in the use of precision livestock technologies (PLT) for dairy calf management. However, there are few sensors commercially available and validated for use in this population. The long-term goal of this research is to optimize pre-weaned dairy calf management using PLT to improve calf health through early disease detection, customized treatment recommendations, and monitoring for recovery. The interdisciplinary project described in this seed grant proposal aims to address the first stage of this long-term goal through design and development of a novel multimodal sensor technology embedded into the nipple from which calves receive milk at automated milk feeding (AMF) stations. We propose to integrate three sensor modalities (pressure, vibration, and contact thermography) within the nipple from which calves feed. The objectives of this project will be to develop this novel PLT for calves fed milk via nipple feeder; to measure and characterize suckling behavior over time at multiple levels (individual feeding, day, 60 d pre-weaning period); and, to describe deviations in normal suckle behavior associated with health and disease in dairy calves prior to weaning. Preliminary activities will focus on sensor development, testing, and validation. Preliminary data will characterize normal suckle behavior and inform sensor placement around the nipple. The anticipated impact of this novel PLT development for calves is to leverage existing behaviors and activities (i.e., feeding milk to pre-weaned calves) to assist dairy producers with calf health management and thus improve calf health, welfare, and make more efficient use of farm staffing resources.
Animal Health Component
100%
Research Effort Categories
Basic
70%
Applied
30%
Developmental
(N/A)
Goals / Objectives
There is growing interest in the use of precision livestock technologies (PLT) for dairy calf management. However, there are few sensors commercially available and validated for use in this population. The long-term goal of this research is to optimize pre-weaned dairy calf management using PLT to improve calf health through early disease detection, customized treatment recommendations, and monitoring for recovery. The interdisciplinary project described in this seed grant proposal aims to address early disease detection, the first stage of the long-term goal, by developing a novel multimodal sensor technology embedded into the nipple from which calves receive milk. The objectives of this project will be 1) to develop this novel PLT for calves fed milk via nipple feeder; 2) to measure and characterize suckling behavior over time at multiple levels (individual feeding, day, 60 d pre-weaning period); and 3) to describe deviations in suckle behavior associated with health and disease.
Project Methods
Objectives. The research objectives of this seed grant will be to: R1) test and validate sensor modalities for integration with nipples at automated milk feeding (AMF) stations; R2) validate, measure, and characterize suckle behavior at multiple levels (within and between calves, individual feeding visit, across feeding visits in a day, and across days in the pre-weaning period); and, R3) describe deviations in suckle behavior for health and disease.Methods: R1. All sensors need to be designed to withstand a variety of environmental conditions. Sensors embedded into nipples need to withstand environmental conditions, and continue to be thin, flexible, and stretchable to not be obtrusive to the usual suckle behavior while conforming to the shape of the deformable nipple. The Co-PD will leverage their previous work on multimodal fabric-based sensor technologyand modify it to make it more robust to environmental fluctuations in temperature, moisture, negative pressure from sucking, as well as compression and shear forces from chewing. The pressure sensor prototype is nearing readiness for field deployment and testing. Funding from this proposal would support the addition of thermal sensing integration and acoustic vibration for testing and validation on farm. A data-acquisition module will be developed for all sensor data collection and storage at high enough frequency rates to capture and characterize suckle behaviors. Each data source will be synchronized and collected at the same frequency with timestamps for data post-processing. Individual calf identification will be recorded at the automated milk feeding station using existing RFID ear tags along with individual feeding data from the AMF station. Testing and validation with be performed at a single cooperating commercial dairy farm in upstate New York (letter of support provided). A convenience sample of dairy calves will be enrolled in the testing of each prototype. Prototypes will be tested at a single feeding station available to one group of 20 calves. It is estimated that up to 5 prototypes will be tested in the field sequentially as modifications and sensor integration occur. R2 & R3. Validation will involve intensive daily health monitoring performed by trained personnel in the Dr. von Königslöw laboratory to collect daily health scores. Targetted collection of serum on d 2 with be used to estimate transfer of passive immunity, blood gas analysiswill be performed at neonatal calf diarrhea(NCD) diagnosis, thoracic auscultationand thoracic ultrasoundwill be performed bi-weekly from d 7 - 60. Fecal samples and nasal swabs will be collected and stored at first sign of NCD or bovine respiratory disease, respectively. A subset of samples will be analyzed to identify relevant pathogens. A convenience sample will be enrolled representing 50 calves at the participating farm in upstate New York. Calves will be followed from the time they are introduced to the AMF (d 4) until weaning at approximately d 60 of life. It is estimated that 15,000 (50 calves x 5 visits/d x 60 d) time series data records of pressure, temperature, and sound/vibration will be collected. As such, we can use temporal convolution networks (TCNs) or multimodal transformers such as CMXby treating the distributed tactile data as a tactile image. Model inputs will include all synchronized sensor channels concatenated as a vector along with health outcome labels (healthy/diseased/recovered) as determined through the on-farm daily health monitoring described above. If required, a modality-specific encoder can be explored to create and concatenate low-dimensional vector representation of each modality (pressure, temperature, sound/vibration) to pass as inputs to the prediction network. These research questions are open for exploration and validation. This novel data will be used to establish a baseline and within and between calf variation in sensor data at multiple levels (feeding, day, 60 d pre-weaning period). Deviations in suckle behavior will be described and characterized for healthy and diseased calves.Plans for communicating results. Data will be made available through publications and presentations at scientific meetings and seminars throughout the project timeline. We expect to publish our research findings in peer-reviewed scientific journals. Any thesis and/or dissertations resultant from the proposed study will be deposited online at eCommons (https://ecommons.cornell.edu). It is anticipated that future projects will include a substantial extension component.Evaluating Outcomes. Data records will be monitored to prevent interruptions in data collection and will be backed up both on and off site. Model validation using one-calf-out cross validation as well as k-fold cross-validation will be used to see if the performance of our data-driven model can generalize to unseen calves or unseen data. Success will be the equivalence or better performanace than the manual performance of objective health scoringand point of care testing.