对船舶柴油机而言, 润滑油常受到冷却液的污染, 引起润滑油劣化变质, 从而导致其功能失效。 冷却液的主要成分是水、 乙二醇及少量的防腐蚀、 抗穴蚀、 消泡沫等添加剂。 将拉曼光谱用于检测润滑油被冷却液污染的浓度, 是一种针对复杂混合物的拉曼光谱检测问题, 单个拉曼峰强度的定量分析方法无法满足浓度的定量检测。 为此, 将拉曼光谱分析和LSTM神经网络数据挖掘方法应用于检测润滑油冷却液污染的浓度。 在实验室条件下, 配制了冷却液污染浓度为2%, 1.5%, 1%, 0.5%, 0.25%和0%的柴油机润滑油油样, 对每个油样取样50次, 并进行拉曼光谱分析, 共获得300个拉曼光谱数据, 随机抽取其中80%的数据作为神经网络训练样本, 剩余20%的数据作为测试样本, 拉曼光谱样本数据的光谱范围为300~2 000 cm^-1; 对数据进行预处理, 包括采样、 拟合、 离散点平均梯度估计等; 构建训练样本集, 将LSTM神经网络和多层全连接层(FC)结合, 建立4种不同的神经网络模型结构; 得到其在训练集和测试集上的平均误差曲线、 测试集上的检测准确率曲线。 分析结果表明, FCs, LSTM-FCs-1, LSTM-FCs-2和LSTM-FCs-3等4种神经网络模型, 检测准确率分别为96.7%, 93.3%, 98.3%和83.3%。 选取任意1%的波数点, 加入幅值随机正负变化1%的噪声之后, 4种神经网络模型的检测准确率分别为88.3%, 90.0%, 96.7%和78.3%。 可见, 相比于其他3种神经网络结构模型, LSTM-FCs-2模型更适用于进行润滑油冷却液污染的定量估计, 加噪后最高准确率仍可以达到96.7%, 鲁棒性优于其他三种模型。 拉曼光谱结合LSTM网络中的LSTM-FCs-2模型, 应用于冷却液污染浓度分别为0.2%和0.4%的实际油样检测, 相对误差分别为5.0%和7.5%, 结果表明该方法可用于在用润滑油冷却液污染浓度的检测。

For marine diesel engines, lubricating oil is often contaminated by the coolant, resulting in the deterioration of lubricating oil, further leading to its functional failure. The main components of the coolant are water, ethylene glycol, and a small number of additives such as anti-corrosion, anti-cavitation, and defoaming. The application of Raman spectrum to detect the concentration of coolant contaminating lubricating oil is a kind of Raman spectrum detection problem for complex mixtures. The quantitative analysis method of single Raman peak strength cannot meet the quantitative detection of concentration. Therefore, Raman spectral analysis and LSTM neural network data mining are applied to lubricant coolant contamination. Under laboratory conditions, diesel oil samples with coolant contamination concentrations of 2%, 1.5%, 1%, 0.5%, 0.25% and 0% were prepared. Each oil sample was analyzed by Raman spectroscopy for 50 times, and a total of 300 Raman spectral data were obtained. 80% of the data were randomly selected as neural network training samples, and the remaining data were taken as test samples. The wavenumber of Raman spectral sample data was 300~2 000 cm^-1. Data preprocessing, including sampling, fitting, discrete point average gradient estimation. The training sample set was constructed, and the LSTM neural network was combined with multi-layer full connection layer (FC) to establish four different neural network model structures, including FCs, LSTM-FCs-1, LSTM-FCs-2, andLSTM-FCs-3. The average error curves and detection accuracy curves of the four networks on the training set and test set are obtained. The results showed that the accuracy of FCs, LSTM-FCs-1, LSTM-FCs-2, and LSTM-FCs-3 neural network models was 96.7%, 93.3%, 98.3% and 83.3%, respectively. In order to study the robustness of the four models, the detection accuracy of the four neural network models was analyzed by selecting any wavenumber of 1% and adding noise whose amplitude changed by 1% randomly. The results were 88.3%, 90.0%, 96.7% and 78.3%, respectively. It can be seen that compared with the other three neural network structural models, LSTM-FCs-2 model is more suitable for quantitative estimation of lubricant coolant contamination, and its highest accuracy can still reach 96.7% after adding noise, and its robustness is better than the other three models. Raman spectroscopy combined with the LSTM-FCs-2 model in the LSTM network was applied to the sample of lubricating oil in use with 0.2% and 0.4% coolant contamination concentrations, respectively, with relative errors of 5.0% and 7.5%. It shows that this method can be used to detect the concentration of used lubricating oil contaminated by the coolant.