A plug-and-play and physics-informed wavelet weight initialization is proposed to construct an interpretable convolution kernel, which makes the diagnosis interpretable. Balanced dynamic adaptive ...