| Summary: | This research proposes a highly accurate and novel artificial intelligent (AI) sensor for inferring in-bin grain moisture content (IBGMC) during drying. This work models a real agricultural grain in-bin drying and storage unit. Since it is a batch drying process, it is inherently dynamic. The use of moisture cables is the current practice to obtain IBGMC. However, moisture cables are costly to purchase, install, and maintain. The approach of this work is to install moisture cables in a drying bin to obtain the data necessary to accurately calibrate (i.e., model) its IBGMC AI sensor and then remove the cables and use them to calibrate AI sensors for other drying bins. After calibration and during operation, the AI sensor uses data from air condition sensors positioned near the fan inlet and in the bin's headspace. This AI sensor methodology is a combined theoretically-based dynamic physiological and empirical (i.e., data driven) modeling approach termed Physically-Informed Neural Network (PINN). However, this work uses a new type of PINN approach. It is evaluated using one bin of six fills over a three-year period. One bin-fill is used to obtain (i.e., trained) the virtual sensor, one to supervise model development to guard against over-fitting (i.e., validate), and four to evaluate (i.e., test) its accuracy and efficacy. The following measures of performance compare moisture cable data to AI sensor data: 1. the fitted correlation (rfit); 2. the average absolute difference (AAD) and; 3. the average difference (AD). For the four test fills, all these results are excellent, with little room for improvement, and thus, greatly support this approach as a very promising alternative to monitoring grain dying using moisture cables.
|
|---|
