Design and Implementations of Open-Source AG IoT Devices for Farm Machinery Data Acquisition and Integrated Analytics

Yang Wang, Purdue University

Abstract

Agricultural machinery is critical in modern farming. With continuous technological advancements in farm machinery, farm machines have evolved from simple mechanical machines to cyberphysical systems that contain rich sources of multimodal sensor data. Effective acquisition and analyses of these data have become essential but challenging tasks in revealing machine-centric and logistical insights to researchers and farmers.In this dissertation, theses challenge are addressed in two parts. The first part demonstrates successful development and deployment of two open-source telematic devices for collecting machine network, geospatial, and video data. The first, ISOBlue 2.0, was designed to be a logger of both GPS and CAN data with wireless data streaming capabilities. The second, ISOBlue HD, an extension of ISOBlue 2.0, was configured to behave as a network server that interfaced with external cameras for automatic video recording of machine operation contexts. These devices were deployed in a variety of machines in different farming activities. A total of over 1 TB of multimodal machinery data were collected.The second part presents three problems that focus on analyzing primarily GPS track data collected from past wheat harvests. The first poses an activity classification problem. It involved clustering a 3D feature set generated from both GPS and CAN data from a combine using the Density-Based Spatial Clustering of Applications with Noise algorithm. The resultant clusters between on-road and in-field data samples as well as normal and anomalous activities. The second problem concentrates on combine unloading event detections using GPS tracks of multiple combines in 16 harvest sessions. The identified events from a novel algorithm that couples Interacting Multiple Models filtering and composite rules were utilized to estimate the total yield for each session. The estimated yields had an overall accuracy of over 90% when comparing to the actual weight ticket records. Lastly, two instantaneous metrics, instantaneous area capacity and swath utilization, were proposed and estimated using GPS tracks of multiple combines in 7 different fields during various harvest years. A novel algorithm was created for estimating instantaneous actual harvested area and swath utilization. This enabled exact computations of instantaneous metrics as oppose to conventional rough estimates of area capacity. Harvest performances were evaluated both temporally and geospatially by machines and years. It was discovered that three contributing factors that lead to high area capacity were wide header attachments, high harvesting speed, and uniform harvesting patterns. Moreover, it was found that the benefit of a wider header might diminish if the harvesting speed was low.

Degree

Ph.D.

Advisors

Krogmeier, Purdue University.

Subject Area

Computer science|Geographic information science|Agricultural engineering

COinS