Abstract: Mechanical assets are vital for process industries business operations. Failure of these assets leads to business interruption and significant losses. Early detection of anomalies can help industries to formulate a maintenance plan and avoid unplanned downtimes. This also enhances assets’ reliability, maintainability, and availability. The aim of this study is to leverage different data-driven multivariate statistical techniques to detect a heat exchanger leak at a local refinery in real-time. The data is collected from the refinery’s historian for a heat exchanger in a distillation unit preheat train network. A set of offline models are developed using static Principal Component Analysis (PCA), dynamic PCA, static Projection to Latent Structures (PLS), and dynamic PLS to detect a heat exchanger leak that occurred in the past. The models then are deployed online to monitor the heat exchanger’s performance and detect leaks in real-time. The results of the study demonstrated that static PCA, dynamic PCA, and dynamic PLS succeeded to detect the leak 50 days (with 95% confidence limit) and 19 days (with 99% confidence limit) before it was found in the facility by operators. The online models also, without generating any false alarms, performed well in monitoring the heat exchanger’s performance in real-time.