Boiler tube wall temperature anomaly situational awareness algorithm based on infrared thermal imaging technology

Abstract In thermal power plant boilers, only directly collected boiler tube wall temperature data is available, resulting in poor real-time perception. Therefore, an algorithm for detecting abnormal boiler tube wall temperature situations based on infrared thermal imaging technology is proposed. The local heating surface is divided according to the maximum silhouette coefficient criterion. Data is collected using an infrared thermal imager and transmitted to a central processing unit. The collected data is denoised using a mean filter algorithm. A 1D-LSTM local feature extraction module is employed, combined with convolution kernels of varying scales to extract and fuse features, thereby constructing an abnormal situation feature extraction model. Abnormal wall temperature situations are classified into five categories. Using an improved fuzzy C-means clustering algorithm, the training data is divided, and the cluster centers are recorded. The Mahalanobis distance is calculated as the detection threshold. Test data is classified and compared with the threshold. Any data exceeding the threshold is considered abnormal, thus enabling abnormal boiler tube wall temperature situation awareness. Experimental results show that when tested in the risk value range of 0.12–0.22, the tube surface temperature rises from 315°C to 368°C within two hours, with a rate of increase of approximately 0.45°C/min. The designed algorithm accurately captures temperature changes. In the special experiment, the situation value fluctuated from 0.15 to 0.42 and then fell back. In the comparative experiment, the designed algorithm accurately located all four abnormal areas. Reference [1] missed two, reference [2] misjudged one, and reference [3][4] was correct in only one. This proves that the algorithm performs best in real-time perception of abnormal boiler tube wall temperature.