Tunnel Boring Machine (TBM) main bearings often operate in harsh conditions, and their low - speed operation makes fault diagnosis challenging. The LCD-FastICA and MCNN (LFM) method is proposed in this study. The local characteristic - scale decomposition (LCD) decomposes complex non - stationary signals, and Fast Independent Component Analysis (FastICA) extracts independent components to highlight fault - related features. The multi - scale Convolutional Neural Network (MCNN) automatically learns features at different scales. The LFM method has three main advantages: it applies the multi - scale convolution mechanism to TBM main bearing diagnosis for the first time, improving diagnosis accuracy; it performs well with a small sample size, solving the problem of difficult sample collection in practical engineering; and it shows higher diagnostic accuracy and stronger adaptability when dealing with high - noise and low - frequency weak signals compared with traditional methods. Experimental results verify the effectiveness of the proposed method, providing a new solution for TBM main bearing fault diagnosis.