基于FDD模型的掘进机截割减速器油液状态评估研究

Oil condition evaluation for cutting reducer of roadheader based on FDD model

掘进机截割减速器的可靠运行与润滑油状态息息相关,为合理评估油液状态,依据粘度、水分、颗粒数三种油液指标,提出了一种基于模糊深度学习模型(FDD)的油液状态评估方法。首先,按照单个指标将油液状态划分为四个等级,根据模糊综合评估法进行模糊评估;其次,将各指标数据进行归一化处理,作为深度神经网络的输入,再运用ReLU激活函数对网络进行激活,得到一个过拟合的神经网络;然后利用Dropout层特性,降低网络拟合程度,同时使用遗传算法对模型中的超参进行优化。最后,使用仿真数据对模型进行训练,并利用实际数据对模型进行验证。结果表明,该方法对油液状态的平均预测精度达到97%,数据损失0.0018,解决了由于多指标信息不一致导致油液状态表征困难及数据较少情况下神经网络训练困难的问题。

Reliable operation of the cutting reducer of the roadheader is closely related to the lubricating oil state.In order to reasonably evaluate the oil state,an oil state evaluation method based on fuzzy deep learning model(FDD)was proposed according to the viscosity,moisture and particle number.Firstly,the oil state was divided into four grades according to a single index,and the fuzzy evaluation was carried out according to the fuzzy comprehensive evaluation method;Secondly,the index data was normalized as the input of the deep neural network,and then the ReLU activation function was used to activate the network to obtain an overfitting neural network;Then,the degree of network fitting was reduced using the Dropout layer feature,and the hyper parameters in the model was optimized using the genetic algorithm;Finally,the model was trained with the simulation data,and the the model was validated with the actual data.The results show that the average prediction accuracy of the proposed method is 97%,and the data loss is 0.0018,which solves the difficulty in characterizing oil state due to inconsistent multi-index information and the dfficulty in training neural network with less data.