Intelligent Online Inspection of the Paste Quality of Prebaked Carbon Anodes Using an Anomaly Detection Algorithm

Abstract

Prebaked carbon anodes are a critical consumable in the aluminum electrolysis industry. Prebaked carbon anode paste is the intermediate product of the prebaked carbon anode, and its quality significantly impacts the prebaked carbon anode. Therefore, inspecting the quality of the prebaked carbon anode paste is essential. Currently, the quality inspection of the paste still relies on laboratory analysis or manual experience. A laboratory inspection cannot obtain results in real time, while manual inspection poses potential risks. To address these issues, an online intelligent inspection method for prebaked carbon anode paste based on an anomaly detection algorithm was proposed. Firstly, we acquired the temperature of the paste and the power of the kneading motor. Secondly, we transformed these time-series data into images using the Gramian Angular Field (GAF) technique and joined them to create the paste anomaly detection dataset. Thirdly, we trained a matched anomaly detection model based on the PatchCore algorithm. Finally, we compared two advanced models: HaloAE and TSRD. PatchCore performs best on our dataset with an AUC-ROC score of 0.9943, followed by HaloAE (0.9906) and TSRD (0.9811). Our proposed method enables on-time intelligent inspection of prebaked carbon anode paste quality. This eliminates the need for manual inspection, reduces labor requirements, and ensures worker safety.