Lubrication state monitoring in rolling bearings plays a vital role in preventing premature bearing failure and ensuring reliable machinery operation. Traditional lubrication-related failures continue to be a significant concern, but existing monitoring methods struggle due to the unique properties of grease and the limited availability of relevant datasets. Several methods for lubrication state monitoring, including acoustic emission sensors, accelerometers, and temperature sensors, are reviewed, highlighting their advantages and limitations. To overcome these challenges, a method based on prior knowledge-embedded convolutional autoencoders (PKECA) is proposed. This method extracts unsupervised features from vibration data and establishes monitoring thresholds based on reconstruction errors, thus improving the accuracy of anomaly detection. The method is validated through a custom-designed bearing lubrication state monitoring test bench, which demonstrates its effectiveness in accurately monitoring lubrication states under different conditions. The results confirm that the method can effectively detect lubrication-related anomalies, contributing to the extended lifespan and enhanced reliability of rolling bearings.