In machinery condition monitoring, it is often vital to consider information from multiple sources due to possible sensor failure or signal distortion, which may result in misclassification of the health status. An issue with multiple sensor data fusion, however, is that the classification can be affected by conflicting results between sensor signals. The proposed method uses a novel three-module approach to information fusion in order to address the problem. Features corresponding to signal integrity are extracted and employed for training a one-class support vector machine to detect unwanted distortions or sensor failures. Different classifiers are trained for the different sensor types available and each signal recorded is used to determine machine health. Decision-level fusion is conducted through a majority voting system using the integrity scores derived from the OCSVMs and the separate classification results. From this, a dynamically weighted fault diagnosis based on sensor signal quality is obtained. Experimental verification using vibration and acoustic emission signals show that the framework is viable and allows for an increased reliability in machinery health diagnosis.