The long-term exposure of mechanical components and their peripheral equipment to self-excited and forced vibrations may affect operational safety, potentially impacting both production and daily life. Ensuring the normal operation of critical mechanical components throughout their life cycle is essential. To address the issue of fault diagnosis for vibrating metal mechanical components, this paper proposes an improved diagnostic algorithm called FDANet, based on the Domain-Adversarial Neural Network (DANN) domain adaptation algorithm. First, to address the problem of insufficient datasets and label scarcity, the DANN algorithm is used. This algorithm leverages adversarial learning between the generator and the discriminator to achieve system stability at its optimum, effectively solving the small sample problem encountered in real-world engineering applications, while also efficiently utilizing the optimal parameters obtained from previous tasks. Second, to tackle the training challenge of gradient explosion, which may lead to convergence difficulties, the FDANet model employs ShuffleNet V2. This enables stable and fast feature extraction, improving system processing speed and enhancing the effectiveness of domain adaptation. Finally, this paper compares the proposed algorithm with other current diagnostic classification algorithms, further validating its superiority.