A Self-Learning Multi-Sensing Selection Process: Measuring Objects One by One

The paper presents a smart approach for a real time inspection and selection of granular objects in continuous flow. In the proposed approach, parallel channels are carved on a planar substrate to contain object flow. Each channel operates independently by processing and selecting grains one by one in real-time using multiple sensing units. A 3D conformational characterization of single objects is achieved by means of simultaneous cross-combined optical and impedimetric sensing technique. The sorting process is based on a 2 phase operative methodology defined by software control: 1) a self-learning step where the apparatus "learns" to identify objects by inputting a-priori selected classes of objects so that decision thresholds are adjusted accordingly; 2) an operative selection process where objects are detected, classified using a decisional algorithm and selected in real time by electromechanical actuators. As working example, cereal grain selection is presented.