Research on a blockage detection method for the suction pipeline of the fan system in a pneumatic jujube picker

To address the problems of insufficient detection accuracy, poor real-time performance, and lack of long-term adaptability in blockage detection for the suction pipeline of the fan system in pneumatic jujube pickers, this study proposes an intelligent detection and operation–maintenance method that integrates multi-task learning and digital twin technology. A JujubePipe-BlockMTL-LightGBM multi-task learning model is constructed to simultaneously identify the blockage location (front, middle, rear) and determine the blockage area ratio (six levels) within a unified framework, thereby overcoming the limitation of traditional single-task models that ignore the physical correlation between tasks. Furthermore, the proposed model is embedded into a digital twin-based real-time monitoring system. Through cyber–physical mapping and closed-loop control, online fault diagnosis is achieved, and a model self-evolution mechanism is introduced to cope with data distribution drift, ensuring long-term stability and accuracy of the system. Experimental results show that, on the test set, the proposed model achieves 100% recognition accuracy for front and middle blockages and 98.18% for rear blockage, significantly outperforming traditional machine learning and deep learning baseline models. In a 72 h continuous test, the overall diagnostic accuracy of the digital twin system reaches 98.67%, with no false alarms for severe blockages, thereby verifying the comprehensive advantages of the proposed method in terms of high accuracy, strong robustness and continuous self-adaptation. This work provides an effective technical pathway for intelligent operation and maintenance of pneumatic conveying agricultural equipment.