A universal hydraulic-mechanical diagnostic framework based on feature extraction of abnormal on-field measurements: Application in micro pumped storage system

How to extract the running feature information and realize multi-type faults diagnosis is the key to carry out intelligent operation and maintenance of energy conversion machinery. The pumped storage unit (PSU) has various operating conditions, both energy storage and power generation. It may lead to diversified types of failures under the joint influence of hydraulic and mechanical factors. The existing data-driven models often show excellent diagnostic performance with laboratory-specific fault types or single subsystem but are not satisfactory when applying real operating scenarios or migrating to other devices. Therefore, a universal hydraulic-mechanical diagnostic framework integrating signal acquisition, feature extraction and fault recognition is proposed in this paper. In Stage 1, eight types of datasets caused by the hydraulic-mechanical coupling characteristics via abnormal on-field measurements form an on-site benchmark; for Stages 2 and 3, a novel refined composite multiscale cosine similarity Lempel-Ziv complexity method is proposed to quantify the various fault features based on multiscale signal processing and the nonlinear dynamics methodology, and random forests model is introduced to realize the efficient recognition of different status signals. Its core advantage is versatility, which is not limited to specific components but can be applied to different subsystems of pumped storage, such as hydraulic system (identification of vortex conditions, detection of hydraulic imbalances) and mechanical system (wear of shaft, bearing and runner). This framework is applied in the micro PSUs, the comprehensive experiments show that all evaluation indexes are above 92%. Various comparative analysis indicates that the framework is not only applicable to the detection and analysis of hydraulic anomalies but also has a competitive advantage in the diagnosis of hydraulic-mechanical faults. It is suitable for fault detection of different subsystems in real power stations, and also could be flexibly extended to other rotating energy storage systems, as a helpful tool to improve energy conversion efficiency and reduce maintenance cost.