Modified Grammian Angular Field for spatially informed induction motor fault diagnosis using feature fusion for fault tolerant control applications

Early detection of bearing faults is crucial for ensuring the operational safety and fault-tolerant control (FTC) of motors. Traditional 2D representations of 1D time series data often fail to preserve temporal variations within a spatial context, limiting their effectiveness in fault diagnosis and control adaptation. To address this limitation, Grammian Angular Field (GAF) provides a structured encoding of time series data. This paper introduces a Modified Grammian Angular Field (M. GAF), a refined feature representation that enhances fault detection and enables adaptive control strategies. Our approach employs a weighted overlapping technique that fuses the Grammian Angular Difference Field (GADF) and logarithmic Grammian Angular Summation Field (Log. GASF), effectively capturing intricate temporal patterns and spectral features. The enhanced M. GAF representation serves as the foundation for our proposed Involution Convolution Feature Concatenation (I. C. FC) framework, which extracts both channel agnostic and spatial-specific, as well as spatial agnostic and channel-specific features, facilitating robust fault diagnosis and real-time control adaptation. Experimental results demonstrate the robustness of our approach, achieving 99.74% accuracy on the University of Ottawa dataset and 100% accuracy on the Case Western Reserve University dataset. Furthermore, our methodology contributes to the advancement of FTC strategies by enabling intelligent fault detection.