[Objective] Under deep and rapid load changes driven by power dispatch and coal quality fluctuations, pulverizing systems of coal-fired units frequently operate under wide-range variable loads, easily inducing coal blockage and coal interruption faults. These faults are concealed and evolve gradually with hysteretic observable parameters, so traditional fixed-threshold DCS alarms cannot provide accurate early warning, and diagnosis and closed-loop disposal lag behind. To address these problems, an intelligent diagnosis and self-healing control method for coal mill blockage and interruption abnormal conditions is proposed. 

[Methods] For a medium-speed mill positive-pressure direct-fired pulverizing system, a key state parameter model is built on material balance, air-coal transport and motor load characteristics, describing the dynamic coupling among internal material inventories, differential pressure and mill current. A nonlinear state observer with prediction-correction feedback is introduced to estimate the unmeasurable internal inventories online via the measured mill current residual, and model parameters are identified from healthy operating data by the particle swarm optimization algorithm. Diagnostic criteria integrating virtual material states, model residuals and parameter directionality are then constructed based on differential pressure, current, outlet temperature and instantaneous coal feed rate. Four feature quantities are fused through fuzzy quantification into a continuous comprehensive fault index, and a hierarchical warning mechanism comprising early warning and fault warning is established. Self-healing sequential control procedures for the two faults are designed, with maintenance personnel notification and manual intervention links introduced; a cloud-edge coordinated closed-loop diagnosis and self-healing control system is developed, with modules deployed on a hyper-converged server and linked to the DCS via OPC UA and Modbus TCP. 

[Results] Validation on a 4 h independent data set shows that the reconstructed current and differential pressure track DCS measurements well, with RMSE of 1.12 A and 0.04 kPa and mean relative errors of 1.9% and 0.72%. Field application on a 1 000 MW ultra-supercritical unit shows that the system identifies blockage and interruption trends at the early fault stage and achieves self-healing by coordinating coal feed adjustment, hot/cold air regulation and vibration device activation. The blockage and interruption self-healing processes take about 10 min and 5 min; the disturbance to main unit parameters is small, and key mill parameters are smoothly restored to normal ranges. A model degradation monitoring and parameter online updating mechanism based on sliding residual windows further ensures long-term model effectiveness. 

[Conclusion] The proposed method supports the transformation from post-fault alarm to early trend prompting and immediate disposal, providing a reference for intelligent diagnosis, risk suppression and reduced-manpower operation of pulverizing systems under bidirectional disturbances from power dispatch and coal quality.