Wireless sensor networks, actuation, and signal processing for apiculture

Thumbnail Image
FEM_OnlineThesis.pdf(4.86 MB)
Full Text E-thesis
Edwards Murphy, Fiona
Journal Title
Journal ISSN
Volume Title
University College Cork
Published Version
Research Projects
Organizational Units
Journal Issue
Recent United Nations reports have stressed the growing constraint of food supply for Earth's growing human population. Honey bees are a vital part of the food chain as the most important pollinator for a wide range of crops. Protecting the honey bee population worldwide, and enabling them to maximise productivity, are important concerns. This research proposes a framework for addressing these issues by considering an inter-disciplinary approach, combining recent developments in engineering and honey bee science. The primary motivation of the work outlined in this thesis was to use embedded systems technology to improve honey bee health by developing state of the art in-hive monitoring systems to classify the colony status and mechanisms to influence hive conditions. Specific objectives were identified as steps to achieve this goal: to use Wireless Sensor networks (WSN) technology to monitor a honey bee colony in the hive and collect key information; to use collected data and resulting insights to propose mechanisms to influence hive conditions; to use the collected data to inform the design of signal processing and machine learning techniques to characterise and classify the colony status; and to investigate the use of high volume data sensors in understanding specific conditions of the hive, and methods for integration of these sensors into the low-power and low-data rate WSN framework. It was found that automated, unobtrusive measurement of the in-hive conditions could provide valuable insight into the activities and conditions of honey bee colonies. A heterogeneous sensor network was deployed that monitored the conditions within hives. Data were collected periodically, showing changes in colony behaviour over time. The key parameters measured were: CO2, O2, temperature, relative humidity, and acceleration. Weather data (sunshine, rain, and temperature) were collected to provide an additional analysis dimension. Extensive energy improvements reduced the node’s current draw to 150 µA. Combined with an external solar panel, self-sustainable operation was achieved. 3,435 unique data sets were collected from five test-bed hives over 513 days during all four seasons. Temperature was identified as a vital parameter influencing the productivity and health of the colony. It was proposed to develop a method of maintaining the hive temperature in the ideal range through effective ventilation and airflow control which allow the bees involved in the activities above to engage in other tasks. An actuator was designed as part of the hive monitoring WSN to control the airflow within the hive. Using this mechanism, an effective Wireless Sensor and Actuator Network (WSAN) with Proportional Integral Derivative (PID) based temperature control was implemented. This system reached an effective set point temperature within 7 minutes of initialisation, and with steady state being reached by minute 18. There was negligible steady state error (0.0047%) and overshoot of <0.25 °C. It was proposed to develop and evaluate machine learning solutions to use the collected data to classify and describe the hive. The results of these classifications would be far more meaningful to the end user (beekeeper). Using a data set from a field deployed beehive, a biological analysis was undertaken to classify ten important hive states. This classification led to the development of a decision tree based classification algorithm which could describe the beehive using sensor network data with 95.38% accuracy. A correlation between meteorological conditions and beehive data was also observed. This led to the development of an algorithm for predicting short term rain (within 6 hours) based on the parameters within the hive (95.4% accuracy). A Random Forest based classifier was also developed using the entire collected in-hive dataset. This algorithm did not need access to data from outside the network, memory of previous measured data, and used only four inputs, while achieving an accuracy of 93.5%. Sound, weight, and visual inspection were identified as key methods of identifying the health and condition of the colony. Applications of advanced sensor methods in these areas for beekeeping were investigated. A low energy acoustic wake up sensor node for detecting the signs of an imminent swarming event was designed. Over 60 GB of sound data were collected from the test-bed hives, and analysed to provide a sound profile for development of a more advanced acoustic wake up and classification circuit. A weight measuring node was designed using a high precision (24-bit) analogue to digital converter with high sensitivity load cells to measure the weight of a hive to an accuracy of 10g over a 50 kg range. A preliminary investigation of applications for thermal and infrared imaging sensors in beekeeping was also undertaken.
Apiculture , Wireless sensor networks , Machine learning , Honey bee , Actuation , Signal processing
Edwards Murphy, F. 2017. Wireless sensor networks, actuation, and signal processing for apiculture. PhD Thesis, University College Cork.