Rotating machinery is extensively used in various industrial applications. However, its complex operating conditions pose significant challenges for accurate fault diagnosis. This study proposes a multi-task fault diagnosis framework that leverages multi-sensor information fusion to mitigate the impact of complex operating environments on industrial equipment. The framework integrates acoustic-based and vibration-based fault diagnosis tasks through a multi-input, multi-task diagnostic model, where the two tasks remain independent at both the input and evaluation stages. During the training phase, task-specific expert networks and a shared expert network jointly extract features from the input signals of each task. These features are weighted by a gating mechanism and then processed by the task-specific encoder module to generate the final diagnostic output. In the inference phase, the model independently performs accurate diagnoses using either acoustic or vibration signals. This design not only enhances the generalization and robustness of fault diagnosis through multi-sensor feature fusion but also ensures reliable operation even when one sensor modality fails. Experimental results demonstrate that the proposed method consistently achieves high diagnostic accuracy across extensive test samples for both vibration and sound signals.