There are many rotating parts in the crane, such as rolling bearings, gears, etc., whose failure may lead to serious safety hazards and economic losses, so effective fault diagnosis of these rotating parts is very important. Addressing the challenge of poor model generalization in cross-device fault diagnosis tasks due to significant data distribution differences between the source domain and target domain, and limited labeled fault data in the target domain, a domain-adaptive cross-device fault diagnosis method is proposed based on pre-training and multi-loss strategies. Initially, normalized spectra of labeled data from the source domain and a small amount of labeled data from the target domain are used as training samples for pre-training, enhancing the model's feature extraction and recognition capabilities for target device data. Subsequently, the pre-trained feature extraction network is frozen, and unlabeled data from the target domain are incorporated. Multi-loss, including classification loss, Maximum Mean Discrepancy (MMD) loss, and domain adversarial loss, is employed to minimize the distance between source domain data and target domain data in the feature space, improving the model's generalization performance. Experimental validation using the CWRU and SRBF rolling bearing datasets demonstrates the superior performance of the proposed method in cross-device fault diagnosis tasks.