Anomaly Detection of Industrial Smelting Furnace Incorporated With Accelerated Sampling Denoising Diffusion Probability Model and Conv-Transformer

Industrial smelting furnace is the core device for various metal preparations, and its abnormality detection is essential for ensuring the safety, stability, and high quality of smelting production. Traditional manual inspections often struggle to detect them promptly and accurately. This difficulty primarily arises from the harsh environments characterized by high temperatures, intense lighting, substantial dust, and dense water mist, all of which significantly compromise the safety and stability of the melting process. To tackle this challenge, a novel intelligent detection approach incorporated with the accelerated sampling denoising diffusion probability model (DDPM) and the Conv-Transformer model is proposed, which can simultaneously enhance the accuracy of semi-molten condition detection and diminish the model complexity. First, in order to solve the limited availability of samples representing abnormal working conditions in the actual melting process, the DDPM is utilized for model training. Nonetheless, the standard sampling process of DDPM is time-consuming, and this article implements a step-by-step sampling strategy within the DDPM framework, aimed at accelerating the overall sampling process while simultaneously augmenting the detection stability under abnormal condition with small samples. Then, to overcome the challenge of incomplete feature extraction, this study introduces the Conv-Transformer model, which combines the local and global features of the images to improve the detection rate of abnormal condition. Finally, experimental results using industrial smelting data of fused magnesium furnace (FMF) indicate that the proposed detection method demonstrates superior performance, offering higher accuracy and demanding fewer computational resources.