Abstract: The detection of winding deformation faults in the main transformer of a gymnasium faces challenges such as coupled interference from multiple fault features, difficulty in synchronously identifying other fault types driven by a single feature, and insufficient generalization capability of diagnostic models, resulting in low overall detection accuracy. To address this, a study on the detection technology of winding deformation faults in the main transformer of a gymnasium was conducted by integrating DSP technology. This method first utilizes DSP to capture and process basic abnormal signals, extracts frequency-domain features of winding deformation faults through fast Fourier transform; then combines support vector machines to establish a fault diagnosis model, ensuring synchronous identification of multiple fault targets such as axial deformation, radial deformation, local deformation, and overall displacement, outputting initial diagnostic results; finally, an adaptive kernel verification mechanism is introduced, employing the information entropy method to weight and fuse feature vectors, further analyzing and adjusting the diagnostic results to achieve effective detection. Experimental results show that under DSP, the misclassification rate of this method is controlled within 1.2% to 1.4%, significantly outperforming the control method with good application effectiveness.