Artificial Intelligence-Based Fault Detection of Hydraulic Pump According to Missing Value Handling

Excavators are construction equipment that operate on rough terrain, and their breakdowns can occur unexpectedly. Sensor data collected from real industrial sites contain large amount of noise and missing values. Therefore, we propose a missing value imputation method and network that are robust to missing values and noise for fault detection of the swash plate axial piston pump, which is a hydraulic pump for excavators. The data are pump outlet pressure signals with missing values for normal and three faults (hose breakage, hose blockage, and one piston blockage) extracted through simulation. Additionally, various missing rates of 1 similar to 50% were considered. Finally, we confirmed that the mean imputation has stable high performance, and the network combining LSTM and Bahdanau attention mechanism is more robust to noise and values to other networks.