Research on oil fluid condition assessment of cutting reducer of boom-type roadheader based on FDD model

Subject to dynamic shock, the lubricating oil in the boom-type roadheader cutting reducer is prone to premature failure, which severely impacts equipment operation. This paper proposes a fuzzy deep learning model (FDD) that incorporates a Dropout layer for real-time oil condition monitoring, utilizing viscosity, moisture, and particle count. First, the oil fluid's state is evaluated fuzzily for each indicator and classified into four distinct levels. Next, the data for each indicator is normalized and processed as input for the deep neural network, where the ReLU activation function is applied to activate the network, resulting in an overfitted neural network. The Dropout layer is then employed to mitigate overfitting, and a genetic algorithm is used to optimize the model's hyperparameters. Finally, the model is trained using simulation data and validated with real data. The results demonstrate that the FDD model can efficiently monitor the state of the oil fluid, achieving an average prediction accuracy of 97% and a data loss of only 0.0018.