Regulating valves are widely used in process industries. Harsh operating conditions can cause failures, and lead to safety hazards. It is important to develop effective fault detection methods for regulating valves. Whereas, in real-world industrial scenarios, obtaining fault data is challenging and requires a substantial allocation of resources for annotation. To address this issue, a residual-based fault detection method is proposed for regulating valves, which does not require fault data in modeling. Firstly, a flow prediction model is developed using least-squares support vector regression (LSSVR) optimization, by the normal operation data of the regulating valve. Comparing the predicted flow with the actual flow, the flow residuals are then obtained. They are further analyzed by virtue of the cumulative sum (CUSUM) control chart in order to detect the valve fault. The proposed method offers a promising alternative for fault detection design, which does not rely on the extensive fault data of regulating valves. It has been validated using the simulated fault data on a testbed, and the results demonstrate its effectiveness for regulating valves.