Palm tree icon

AI-powered technology for early detection of poultry diseases on a farm

With the help of IoT sensors, the AI-powered system scans the air and detects poultry infections. The action principle of the solution can be likened to the fire alarm system. The platform can be scaled up to thousands of poultry farms across the globe.

End-to-end Solution for Date Palm Growth Image End-to-end Solution for Date Palm Growth Image End-to-end Solution for Date Palm Growth Image

Results

Resources

About the project

Bacterial infections cause economic damage worth billions of US dollars every year. These diseases also affect people’s health. The time required to identify an infection turns into a critical factor. It can take from 4 to 10 days to obtain a result of a bacterial culture test. This duration is enough for the disease to put a poultry farm into unsafe conditions and jeopardize the health of workers.

Our approach became a game changer!

IoT sensors analyze air 24/7 and signal a dangerous concentration of strains in the air. The monitoring process is fully remote. The hardware-software complex was implemented through a partnership with a biotechnological company that promotes sustainability and biological security in the poultry industry.

The portfolio of our RnD department has already included several proprietary technologies developed in this direction. When we started our work, we relied on our engineering expertise accumulated at the juncture of artificial intelligence, machine learning, and the Internet of Things.

Requirements for the Solution

Project Challenges

Building an AI system that will be able to detect infectious diseases based on air quality alone was an unmatched challenge in its complexity
Our solution is remarkable with an original approach that utilizes regular sensors for analyzing air contamination

Fast data mining and processing are crucial to keeping the sustainability of poultry production
When data processing involves massive flows (terabytes of data), there are good chances that the end result will be generated with delays. Such lags lead to losses and damages for a poultry farm. One of the main requirements for our solution was to ensure a prompt analysis of data along with the accuracy of findings.

Research and Development

A pursuit of non-typical tasks challenged us to come up with an intense approach to requirement gathering and shaping a final solution. Consultations with experts were conducted on both ends — by the partnering company and our team. We conducted several consultations with scientists and made sure that we are on the same page regarding the future of the platform. The assistance of microbiologists accompanied us throughout all development stages.

Our Work

  • 1. Planning

    A clear action plan was established first. We were able to achieve this clarity and specify general goals for the first 2 - 3 iterations.

  • 2. The choice of sensor devices and how they worked

    To start with, our data scientists began collecting and expanding the list of the air parameters that will be analyzed. Based on specified parameters, we set up sensors and made necessary adjustments to this construction depending on climate conditions.

    We used regular air quality sensor devices, which measured the following:
    • Temperature and humidity
    • Speed and direction of air movement. To ensure high sensitivity, we chose thin-filmed sensors produced by EPulse
    • Also, our set included CO and CO2 gas sensors

    Every single sensor must be tuned up to the optimal mode so that pathogenic agents in the air could be detected early on. Step-by-step we adjusted sensors and aligned them with changing conditions. Obtaining precision data was crucial for the correct detection of infectious agents.

    We’d like to explain the choice of sensor devices. To apply the air analysis methods that we devised, we purchased smell and gas sensors from a famous manufacturer. The price of one unit doesn’t exceed $100. The chosen device can detect Volatile Organic Compounds (VOCs), volatile sulfur compounds (VSCs), and other gases such as carbon monoxide and hydrogen.

    Sensors are placed checker wise on the ceiling at a distance of 8 - 10 meters and by 0.3 - 1 meter above the floor because gas is heavier than air and accumulates below.

  • 3. The test stage — checking how sensors respond to the air contamination

    As part of test labs, we set up various degrees of air microbial contamination (contamination degrees) or E-coli in the sealed sterile box. It was due to the necessity to account for the nonlinear temperature-affected variation of results. This way, we took the readings of devices and shaped a clear understanding of how devices respond to contamination of the air.

  • 4. Data processing

    With the air parameters’ data delivered by sensors, we could process the information about the source of the disease. This part of the work was central to our development. The data mining allowed for discovering patterns, anomalies, and other useful insights.

    On the one hand, the task was to classify the data transmitted by devices. On the other hand, it took us to apply the methods of multivariate statistics, probabilistic programming, and fuzzy logic.

  • 5. Machine Learning

    The data volume was massive and amounted to terabytes, which demanded the creation of specialized Machine Learning pipelines. For this purpose, we utilized Mlflow and Azure platforms. Our engineers worked on the data-driven ML model training. New data insights were considered for regular adjustment of sensors’ parameters.

  • 6. Interface design

    To let farm operators act on visualized data, ready charts, and analytical insights, we devised a light interface. It connects farmers to the real-time monitoring of air quality on sites. Once sensors have registered the deviation from regular air parameters, the system immediately sends notifications to the board.

Results