To ensure continued profitability in today's increasingly competitive market, industries need to work at near-maximum yields while maintaining high reliability and operational safety. However, production equipment can rarely be maintained at its maximum capacity because it inevitably deteriorates with use and ageing. Without proper maintenance, equipment can fail in unexpected ways, leading to the possible termination of services, loss of production and, in the worst case, human casualties. Hence, it is important to conduct proper fault diagnosis and prognosis via condition-based maintenance (CBM) so that the remaining useful life (RUL) of equipment can be predicted as it degrades. Slurry pumps are commonly used in mining and heavy industry, such as in the extraction of synthetic crude oil from oil sands. This kind of slurry pump is referred to here as an oil-sand pump. The pump impellers and other components are prone to significant degradation as they are under continuous impact from the sand and small stones embedded in the oil. Oil-sand pumps are expensive to replace due to the high cost of manufacturing the pump components, which must be more durable than those of pure water or liquid pumps. Moreover, the unexpected shut-down of an oil-sand pump can yield a loss of over a million dollars per day in the oil production industry. To reduce the time required to repair the pump components and the associated financial loss, effective fault diagnosis and prognosis methods that are tailor-made for the field operation of oilsand pumps are urgently needed. Although numerous CBM methods have been developed, most have been verified only by simulations or experiments conducted in well-controlled laboratory environments, and few have been tested on equipment that is actually operating in the field. This thesis presents some fault diagnostic and prognostic methods that are tailor-made to ensure the effective operation of oil-sand pumps in the field. Three techniques contributed to the effectiveness of the methods. The first was the application of exponentially weighted moving average (EWMA) control charts to identify the obvious shift point from the signal trend for confirming the occurrence of an anomaly in the monitored equipment. Four types of EWMA chart were studied and compared and charts yielding superior performance with respect to the detection of incipient anomalies were identified. Second, two unsupervised clustering ensemble methods were developed to assess the wear status of a monitored pump and then classify the wear severity. Once the obvious shift point for confirming the occurrence of an anomaly had been successfully identified and the fault had been diagnosed, the third technique involved the design of a relevance vector machine (RVM)-based model to predict the RUL of the damaged pump impeller. All three techniques were verified using operational data collected from running oil-sand pumps. The findings indicate that the oil industry could minimize production downtime and thus achieve a dramatic decrease in maintenance costs and human casualties by applying the three techniques identified in this thesis. These methods could also be modified to suit other types of machines exhibiting similar degradation trends, such as deep hole drillers, coal milling machines and rock crushers used in mines. Like oil-sand pumps, these machines suffer from continuously generated severe impacts when handling hard materials.