{"id":216972,"date":"2026-05-04T11:39:29","date_gmt":"2026-05-04T15:39:29","guid":{"rendered":"https:\/\/www.hajim.rochester.edu\/senior-design-day\/?p=216972"},"modified":"2026-05-04T11:39:29","modified_gmt":"2026-05-04T15:39:29","slug":"zalliant-lightweight-heat-stress-detection-for-cow-health-monitoring","status":"publish","type":"post","link":"https:\/\/www.hajim.rochester.edu\/senior-design-day\/zalliant-lightweight-heat-stress-detection-for-cow-health-monitoring\/","title":{"rendered":"Zalliant – Lightweight Heat Stress Detection for Cow Health Monitoring"},"content":{"rendered":"
\n
\"\"<\/figure>\n<\/div>\n\n\n

DSCC 383W Team 7: <\/strong>Aakanksha Dutta, Elvis Imamura, Eva Salmone, Sreejato Chatterjee, Duole Hong<\/p>\n\n\n\n

Sponsors: <\/strong>Zalliant \u2014John Balbian, Emmanuel Mounier, Utsav<\/p>\n\n\n\n

Advisor:<\/strong> Professor Cantay Caliskan<\/p>\n\n\n\n

Background: <\/strong>Cattle health monitoring is vital for agricultural productivity and economic stability. Traditional manual observation often detects illness only after symptoms progress, leading to high mortality and costly, late-stage interventions. Zalliant addresses this by using rumen boluses (sensors) to continuously monitor internal temperature, shifting herd management from reactive to proactive.<\/p>\n\n\n

\n
\"\"
An image of Zalliant\u2019s bolus. The device is swallowed by the cow and remains in the rumen (stomach) for the lifetime of the device<\/figcaption><\/figure>\n<\/div>\n\n\n

Goal:<\/strong> Improve & modify Zalliant’s existing temperature-based alert system by evaluating whether more efficient, data-driven methods can enhance performance while reducing cost. Assess the feasibility of machine learning approaches within Zalliant’s workflow.<\/p>\n\n\n\n

Our Solution: <\/strong>A lightweight, context-aware statistical model that transforms raw rumen temperature streams into actionable health alerts without relying on costly machine learning methods. Instead of using fixed thresholds, our approach models each cow’s normal temperature behavior using rolling baselines and measures deviations relative to that individual context. By applying robust techniques such as median-based smoothing and variability-aware thresholds, our model identifies sustained, abnormal patterns rather than isolated noisy spikes. These consecutive anomalies are grouped into meaningful events, while mechanisms like gap suppression reduce false positives from missing data. This results in a system that is interpretable, resilient to real-world sensor noise, and capable of detecting potential health issues hours earlier, while enabling lower sampling rates to reduce hardware and operational costs.<\/p>\n\n\n\n

Economic & Operational Impact<\/h3>\n\n\n\n

Reduced Hardware Costs<\/strong>: The model remains effective at lower sampling rates, doubling or tripling battery life and reducing bolus replacement costs by up to 50%<\/strong>.<\/span><\/p>\n\n\n\n

Farmer Profitability:<\/strong> Early intervention enabled by the 9-hour lead time saves an estimated $10,000 per 100 cows<\/strong> annually in health expenses.<\/span><\/p>\n\n\n\n

Operational Stability:<\/strong> By focusing on sustained elevations (minimum 3-hour duration), the model reduces false alarms caused by transient behavioral spikes.<\/strong><\/span>

\u00a0<\/div>
<\/p>\n\n\n\n