The global demand for ultrapure water(UPW) has been rapidly increasing, driven by the expansion of high-tech industries such as semiconductors, display panels, and secondary batteries. UPW is a critical resource in these industries, particularly in semiconductor manufacturing, where stringent water purity standards are essential for processes such as wafer cleaning and etching. As manufacturing technologies continue to scale down, the quality and stability of UPW directly impact production yield and process reliability. The UPW market is expected to grow significantly, with increasing investment in advanced water treatment systems and smart plant operation technologies to meet rising purity requirements and operational efficiency.
In semiconductor manufacturing processes, ultrapure water removed contaminants such as particles, organic matter, ions and dissolved gases is essential during wafer cleaning. More than 20 unit processes are typically involved in UPW production, and its stable and efficient operation requires various advanced technologies. Recently, there has been growing interest in anomaly detection technologies utilizing virtual sensors, which estimate physical conditions without physical sensors through mathematical or data-driven approaches. These technologies can reduce the number of installed sensors, lower CAPEX, and improve operational efficiency by predicting difficult-to-measure variables or monitoring system status.
This study aims to diagnose anomalies in the return line of UPW plant using virtual sensor technology by accurately detecting abnormal events in flow, pressure and valve opening. Through validation of the virtual sensor-based anomaly detection algorithm, the method demonstrated high accuracy in detecting flow and pressure valve anomalies. However, pressure measurements showed relatively lower accuracy, indicating the need for further calibration of the model. It is expected to enhance the operational stability of UPW plants by provi