The rotating machinery is widely used to transmit power form the prime mover to the load, such as gearboxes, automotive engines, centrifugal pumps, generators, and so on. Any abnormal situations of the rotating machinery may hazard personnel as well as interrupt normal machine operation, causing enormous economic loss. Therefore, it is of great significance to develop a reliable and accurate intelligent system for fault diagnosis of the rotating machinery. However, rotating machinery diagnosis has been a challenging problem because the existence of simultaneous-faults (i.e. several single-faults appear concurrently) and no unique sensor can detect all kinds of machine faults. By studying the literature, it is learnt that signal-based fault diagnostic systems have the best potential to deal with the challenging problem, but those available systems cannot give reliable diagnostic results and cannot detect many kinds of faults. Aiming to enhance the reliability and diagnose more faults in the rotating machinery, this thesis proposes a new diagnostic framework called probabilistic committee machine (PCM). The proposed PCM consists of three stages: 1) feature extraction for processing the fault signals, 2) fault classification using multiple classifiers, and 3) fault diagnosis by combining the classification results through a probabilistic ensemble method. In the first stage, ensemble empirical mode decomposition combined with feature selection (sample entropy or singular value decomposition) is used for extracting the useful information from the raw fault signals. In the second stage, multiple pairwisecoupled sparse Bayesian extreme learning machines (PCSBELM) are built. Every committee member in PCSBELM is trained by using different kind of signal in order to detect more kinds of faults. In the last stage, a new probabilistic ensemble method is proposed, in which each committee member is assigned with an optimal weight in iii accordance with their reliability and accuracy so that a reliable and widelycovered fault diagnostic results can be obtained from the weighted combination of the members. To verify the effectiveness and generalization of the proposed fault diagnostic framework, it is applied to two typical rotating machines, automotive engine and gearbox. The evaluation results show the proposed framework is superior to the existing single probabilistic classifier. Moreover, the proposed framework can successfully diagnose both single- and simultaneous-faults of the two rotating machines effectively, while the diagnostic systems are trained by single-fault patterns only.