Monitoring and blockage diagnosis in axial flow threshing and separation device under variable feed conditions

To prevent blockage in axial flow threshing and separation devices caused by varying material moisture and feeding rates while simplifying monitoring and diagnostic system, a test bench was used to collect vibration signals from four monitoring points of devices, analyse blockage tendencies under different conditions. Signals were denoised and reconstructed with the Slime Mould Algorithm and Variational Mode Decomposition, and segmented with overlapping moving time windows. Time, frequency, and time-frequency domain features were extracted to assess device operating status and sensitivity of signal changes at different monitoring points. Findings revealed that the duration of a slight blockage tendency was long under normal moisture content and small increments of feeding rate. With high moisture content and large increments of feeding rate, the duration of slight blockage tendency will decrease and quickly enter a severe blockage tendency state, with continued feeding resulting in immediate blockage. The monitoring point directly below the concave grate exhibited the most sensitive signal changes, with the largest waveform variations and standard deviation deviations. Feature dimensionality reduction was performed using Relief-F algorithm, and Bayesian-optimised machine learning models were trained for state identification. The diagnostic model of a monitoring point directly below the concave grate demonstrated high diagnostic accuracy, recall, and reliability, indicative of an effective monitoring point. The Bayesian-optimised Support Vector Machine model achieved the best performance, with 85.1 % and 93.6 % accuracy under different conditions and rapid prediction speeds (53000 and 40000 obs s⁻¹). This met the requirements for a simplified, accurate, and fast online monitoring system.