Traditional deep learning-based fault diagnosis assumes that the training set and test set are independently and identically distributed with consistent label spaces. However, in actual industrial scenarios, new fault types and "distribution shifts" in data frequently occur, necessitating open-set cross-domain recognition, which results in poor model generalization and reduced diagnostic accuracy. To address the issue of open-set cross-domain fault diagnosis, we propose an weighted adversarial domain adaptation method based on a deep one-dimensional convolutional neural network (1D-CNN). Firstly, the 1D-CNN is employed to extract representative features from input samples across multiple domains. Then, a fine-grained feature classifier is utilized to distinguish between shared and outlier features across these domains. Subsequently, a weighting module guides the domain discriminator to achieve domain-invariant feature learning. Finally, the domain discriminator uses these domain-invariant features to guide the feature classifier in minimizing global loss, thereby achieving open-set cross-domain fault diagnosis. Experimental results demonstrate that the proposed model achieves higher cross-domain fault diagnosis accuracy compared to other methods on the Case Western Reserve University (CWRU) dataset and the gearbox dataset, particularly showing significant advantages in open-set data scenarios.