Intelligent Recognition of Abnormal Operating Conditions in Cement Grate Cooler Based on CNN-LSTM

The grate cooler plays a critical role in cement clinker cooling and waste heat recovery. However, under high-temperature dust environments and fluctuating loads, it is prone to abnormal conditions such as “snowman formation” and “kiln skin shedding,” which are difficult to detect in real time using empirical thresholds. This study constructs a multi-sensor dataset covering five operating conditions: normal, early/late stages of kiln skin shedding, and early/late stages of snowman formation. An end-to-end CNN-LSTM model is proposed, where the Convolutional Neural Network (CNN) extracts spatial local features and the Long Short-Term Memory (LSTM) network captures cross-window temporal dependencies. Experimental results show that the model achieves an overall accuracy of 94.43% and an F1-score of 94.35%, outperforming Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) by 11.2 and 13.8 percentage points, respectively. The model also achieves a recall of 91% for early-stage anomalies, demonstrating its capability for early warning and robustness. This research offers a novel approach for intelligent monitoring and energy-efficient, low-carbon control of cement grate coolers.