A Multi-Source Data-Driven Fault Detection and Diagnosis Method for Pumps

Fault detection and diagnosis (FDD) in pumps is crucial for building energy management by detecting the abnormal operation status, increasing the service life of equipment, and enhancing the energy performance of buildings. Most FDD methods predominantly rely on single-source data, such as building automation (BA) data or vibration data. However, sensors in BA systems are prone to inaccuracies, which consequently impedes the performance of FDD algorithms. This paper proposes a novel FDD method for pumps based on multi-source data, which integrates traditional BA electrical power data with non-intrusive measurements, including audio data, vibration data, and infrared thermal images. The method includes two stages: (1) multi-source data anomaly detection and (2) pump fault diagnosis. Various fault scenarios were tested on an experimental platform. The results demonstrate that the proposed method can effectively diagnosis pump faults in detail, such as voltage fluctuations, shaft or bearing wear and tear, inadequate ventilation, and foundation vibration. With intrusive and non-intrusive data, the proposed FDD method is more robust and could provide more detailed diagnosis of pump faults.