Multimodal Hybrid Aero‐Engine Mechanical Wear Fault Diagnosis Algorithm Based on Two‐Channel Data Input Types

ABSTRACT Aero‐engines are complex and sophisticated systems combining mechanical, thermal, and fluidic domains. Abnormal wear of mechanical components is becoming more prevalent due to severe changes in flight conditions and the external environment, which may lead to drastic performance degradation and accidents. Therefore, the diagnosis of such wear and tear faults is urgent, and based on this need, more researchers and scholars are focusing their attention on it. To address the current shortcomings of fault diagnosis algorithms that only rely on one‐dimensional datasets or two‐dimensional image analysis and the low accuracy of final fault identification, an innovative hybrid algorithm is proposed in this study. The algorithm integrates one‐dimensional time series data and two‐dimensional image data, converts the one‐dimensional dataset into a two‐dimensional image dataset through the Gramian Angle Field technique, and subsequently uses a dual‐channel GRU‐CNN (Convolutional Neural Network‐Gated Recurrent Unit) algorithm model designed for fault diagnosis, which can simultaneously analyze and map the features and fault modes of the one‐dimensional dataset and the two‐dimensional image. In order to extract features with richer semantic information and stronger discriminative ability, a multimodal fusion technique is employed, which successfully addresses the limitations of the wear‐and‐tear feature distributions of the two datasets using the cross‐extraction fusion method and combines the advantages of both in terms of trend distributions of the time series and edge feature distributions of the image sequences, respectively. The best fault diagnosis results were achieved by using the strong mapping relationship between the saliency feature expression and the fault modes. The final analysis shows that the recognition rate of typical mechanical wear of aero‐engines exceeds 97%, thus achieving the desired goal.