To address the issues of incomplete condition information in single-sensor monitoring, the disconnection between fault diagnosis and remaining useful life (RUL) prediction, and the insufficient utilization of multisource data in rotating machinery under complex operating conditions, this paper proposes a multisensor data fusion method for fault diagnosis and RUL prediction of rotating machinery. First, multisource monitoring signals are uniformly preprocessed and organized into samples, and sensor-specific feature extraction modules are employed to learn local condition features from different sensor channels. Then, an adaptive multisensor fusion module is designed to dynamically weight and effectively integrate multisource information, thereby constructing a shared health representation that characterizes the machinery health evolution process. Based on this representation, a dual-task joint learning framework for fault diagnosis and RUL prediction is further established to collaboratively optimize current fault-state recognition and future degradation trend modeling. Experimental results demonstrate that the proposed method outperforms single-sensor methods, simple feature concatenation methods, and representative deep learning baselines in terms of diagnostic accuracy, RUL prediction error, and robustness under noisy environments. Ablation studies further verify the effectiveness of multisensor input, adaptive fusion strategy, and dual-task joint learning in improving the overall model performance. The results indicate that the proposed method can provide a more comprehensive characterization of rotating machinery health conditions and offers effective technical support for predictive maintenance and intelligent operation.

rotating machinery, multisensor data fusion, fault diagnosis, remaining useful life prediction