Material identification in bales using multichannel capacitive and weighing sensors with data-driven modelling

Materials are usually compressed into dense bales for easy storage and transportation in the recycling industry. A range of techniques, such as X-ray, ultrasound, microwave, and optical sensors, have been developed to identify materials distributed on a conveyor belt for sorting purposes. However, identifying materials, especially mixed materials in bales, is more challenging. Currently, there is no effective non-contact measurement system for identifying materials in bales. In this study, a new method is proposed for the first time, based on a purpose-built multichannel capacitive sensor and a weighing sensor with data-driven modelling. The capacitive sensor with three sets of transceiver units is designed for sensing the permittivity information of bales from three directions (down-up, left-right and back-front). It is optimised using the finite element method considering high sensitivity and uniform sensitivity distribution. A signal conditioning circuit is developed to amplify and demodulate the sensor outputs. The weighing sensor is used to obtain the mixture density information of the bales. Subsequently, the support vector machine (SVM) model is established to identify materials in the bales based on the multimodal sensing data. The proposed method was evaluated using model bales containing different materials, including paper, cardboard, polyvinyl chloride (PVC), and their combinations. Results demonstrate that fusing the capacitive sensor with the weighing sensor significantly outperforms using the capacitive sensor alone, validating the advantage of multimodal sensing. In addition, the SVM algorithm was compared with random forest (RF) and multilayer perceptron (MLP) algorithms, and SVM demonstrated superior performance in data-driven modelling. To further assess general applicability, woodchip was introduced as an additional material to the original bale composition. Despite the increased complexity, the method maintained satisfactory identification performance, confirming its robustness and effectiveness as a novel solution for material identification in bales.