Lubricating oil physical and chemical parameter detection method based on multi-model fusion
The multi-model fusion method for detecting the physicochemical parameters of lubricating oil solves the problems of long detection time and low prediction accuracy of traditional detection methods, and realizes rapid, non-destructive testing and high-precision monitoring of lubricating oil. It is suitable for real-time monitoring and oil quality assessment in industrial sites.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING INST OF RADIO METROLOGY & MEASUREMENT
- Filing Date
- 2025-12-08
- Publication Date
- 2026-05-01
AI Technical Summary
Traditional methods for monitoring the physicochemical parameters of lubricating oils are time-consuming and costly, making it difficult to achieve real-time monitoring. Furthermore, single infrared spectral models have limited predictive accuracy when dealing with complex oil compositions and nonlinear relationships.
A multi-model fusion method for detecting the physicochemical parameters of lubricating oil is adopted, including acquiring mid-infrared spectral data for preprocessing and feature extraction, selecting pre-trained multivariate regression, support vector regression and neural network models, and constructing an integrated prediction system to adapt to the challenges of nonlinear and high-dimensional modeling under different oil types and operating conditions.
It enables rapid and non-destructive testing of lubricating oil physicochemical parameters, improves prediction accuracy and stability, and is suitable for real-time monitoring and oil quality assessment in industrial settings. It has strong versatility and scalability.
Smart Images
Figure CN121963931A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of lubricating oil quality testing and condition monitoring technology. More specifically, it relates to a method for detecting the physicochemical parameters of lubricating oil based on multi-model fusion. Background Technology
[0002] During use, lubricating oils gradually oxidize, sulfide, increase water content, and rise in acid value, leading to a decline in performance. Traditional methods for monitoring physicochemical parameters (such as titration and chromatography) are time-consuming, costly, require specialized operation, and are difficult to implement in real time. Infrared spectroscopy technology has advantages such as speed, non-destructive nature, and rich information, but single models have limited predictive accuracy when dealing with complex oil compositions and nonlinear relationships. Summary of the Invention
[0003] The purpose of this invention is to provide a method for detecting the physicochemical parameters of lubricating oil based on multi-model fusion, so as to solve at least one of the problems existing in the prior art.
[0004] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for detecting the physicochemical parameters of lubricating oil based on multi-model fusion, comprising: Acquire multi-channel mid-infrared spectral data of the oil sample to be tested; The mid-infrared spectral data are preprocessed and feature extracted to obtain feature data; Select the pre-trained model based on the target parameters and the number of channels; The feature data is input into a pre-trained model to obtain the physicochemical parameters of the oil sample to be tested.
[0005] Optionally, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: The mid-infrared spectral data is filtered to obtain filtered mid-infrared spectral data.
[0006] Optionally, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: Baseline correction is performed on the filtered mid-infrared spectral data to obtain baseline-corrected mid-infrared spectral data.
[0007] Optionally, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: The baseline-corrected mid-infrared spectral data are subjected to a standard normal variable transformation to obtain standardized mid-infrared spectral data.
[0008] Optionally, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: Characteristic data are obtained by performing principal component analysis on the standardized mid-infrared spectral data.
[0009] Optionally, the pre-trained model includes a pre-trained multivariate regression model, a pre-trained support vector regression model, and a pre-trained neural network model.
[0010] Optionally, the physicochemical parameters of the oil sample to be tested include oxidation value, sulfidation value, moisture content, acid value, and alkalinity value.
[0011] Optionally, the step of selecting the pre-trained model based on the target parameters and the number of channels includes: if the number of channels of the photodetector is N1 or N2 and the target parameters are oxidation value and sulfidation value, then a pre-trained multiple regression model is selected.
[0012] Optionally, the step of selecting the pre-trained model based on the target parameters and the number of channels includes: if the number of channels of the photodetector is N3 and the target parameters are acid value and alkalinity value, then a pre-trained support vector regression model is selected.
[0013] Optionally, the step of selecting the pre-trained model based on the target parameter and the number of channels includes: if the number of channels of the photodetector is N3 and the target parameter is moisture content, then a pre-trained neural network model is selected.
[0014] The beneficial effects of this invention are as follows: The technical solution described in this invention enables rapid and non-destructive testing of the physicochemical parameters of lubricating oil; improves prediction accuracy and stability through multi-model fusion; is suitable for real-time monitoring and oil quality assessment in industrial settings; has strong versatility and scalability, and can be adapted to different oil types; and aims to solve the problems of low detection efficiency and poor model adaptability in existing technologies. Attached Figure Description
[0015] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.
[0016] Figure 1 The flowchart illustrates a method for detecting the physicochemical parameters of lubricating oil based on multi-model fusion, as provided in an embodiment of the present invention. Detailed Implementation
[0017] To more clearly illustrate the present invention, the following description, in conjunction with embodiments and accompanying drawings, further explains the invention. Similar components in the drawings are indicated by the same reference numerals. Those skilled in the art should understand that the specific description below is illustrative rather than restrictive and should not be construed as limiting the scope of protection of the present invention.
[0018] During use, lubricating oils gradually oxidize, sulfide, increase water content, and rise in acid value, leading to a decline in performance. Traditional methods for monitoring physicochemical parameters (such as titration and chromatography) are time-consuming, costly, require specialized operation, and are difficult to implement in real time. Infrared spectroscopy technology has advantages such as speed, non-destructive nature, and rich information, but single models have limited predictive accuracy when dealing with complex oil compositions and nonlinear relationships.
[0019] In view of this, such as Figure 1 As shown, one embodiment of the present invention provides a method for detecting the physicochemical parameters of lubricating oil based on multi-model fusion, comprising: acquiring multi-channel mid-infrared spectral data of the oil sample to be tested; preprocessing and extracting features from the mid-infrared spectral data to obtain feature data; selecting a pre-trained model according to the target parameters and the number of channels; and inputting the feature data into the pre-trained model to obtain the physicochemical parameters of the oil sample to be tested.
[0020] In a specific example, by preprocessing and extracting features from mid-infrared spectral data, and constructing a modeling system that integrates multiple regression, support vector regression, and neural networks, rapid and high-precision monitoring of key physicochemical parameters such as oxidation value, sulfidation value, moisture content, acid value, and alkalinity value of lubricating oil can be achieved.
[0021] In a specific example, infrared spectroscopy analysis is based on the characteristic absorption peaks generated by molecular vibrational energy level transitions. Different chemical bonds and functional groups exhibit unique absorption characteristics at specific beams. Changes in lubricating oil during operation, such as oxidation, sulfidation, and contamination, will cause changes in its infrared spectral characteristics, thus reflecting the evolution of its physicochemical state.
[0022] In a specific example, a multi-channel infrared detector is used to acquire spectral data, covering the mid-infrared characteristic band (e.g., 400 cm⁻¹). -1 ~4000cm -1 By establishing a quantitative relationship model between spectral characteristics and physicochemical parameters, rapid inversion of oil composition is achieved. To further improve model robustness and prediction accuracy, a multi-model fusion strategy is introduced, combining multiple regression, support vector regression, and neural networks to construct an integrated prediction system that effectively addresses the challenges of nonlinear and high-dimensional modeling under different oil types and operating conditions.
[0023] The technical solution described in this invention enables rapid and non-destructive testing of the physicochemical parameters of lubricating oil; improves prediction accuracy and stability through multi-model fusion; is suitable for real-time monitoring and oil quality assessment in industrial settings; has strong versatility and scalability, and can be adapted to different oil types; and aims to solve the problems of low detection efficiency and poor model adaptability in existing technologies.
[0024] In a specific example, the infrared spectral data of lubricating oil is susceptible to interference from factors such as instrument noise, baseline drift, and scattering effects during the acquisition process, directly affecting the modeling accuracy and prediction reliability. To address this, a complete data processing workflow was designed, including three main stages: spectral preprocessing, feature extraction, and model building. This aims to improve data quality, reduce dimensional redundancy, and establish a stable prediction model.
[0025] In a specific example, because the spectral data output by the infrared detector has a high dimensionality, with each sample corresponding to a high-dimensional spectral vector, and some bands exhibiting high correlation or redundant information, directly using the raw data for modeling not only increases computational complexity but may also introduce noise that affects model accuracy. Therefore, feature extraction processing is performed on the spectral data before modeling.
[0026] In a specific example, before feature extraction, in order to reduce the impact of factors such as instrument noise, scattering effects and optical path instability during the spectral acquisition process, the original spectral data is smoothed, denoised, baseline corrected and normalized to improve data quality and signal-to-noise ratio.
[0027] In one possible implementation, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: filtering the mid-infrared spectral data to obtain filtered mid-infrared spectral data.
[0028] In a specific example, Savitzky-Golay filtering is a smoothing method based on local polynomial fitting. Its core idea is to fit a polynomial function using the least squares method within a sliding window, thereby achieving noise reduction while preserving the peak structure of the spectrum.
[0029] Furthermore, let the original spectral data signal be... Window length is Smoothed values Represented as:
[0030] In the formula, This is the offset index relative to the center point within the sliding window; The width is half the width of the window; For the order of the polynomial The smoothing coefficient, which is determined together with the window length; For the first in the window The original spectral signal values at each location.
[0031] Furthermore, SG filtering can effectively suppress high-frequency random noise while maintaining the shape and width of the absorption peak.
[0032] In one possible implementation, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: performing baseline correction on the filtered mid-infrared spectral data to obtain baseline-corrected mid-infrared spectral data.
[0033] In a specific example, baseline correction (polynomial fitting method) addresses the potential baseline shift in the spectrum due to scattering from the optical system, transmittance drift, or environmental factors. To eliminate background trends, this invention employs a polynomial fitting method for baseline correction. Assume we have selected M baseline points. ,in Fitting a model using these points. Polynomial function of order:
[0034] In the formula, The baseline signal to be fitted; Wavelength; Let be the coefficient of the constant term to be determined; The order of the polynomial; Obtain the optimal fitting polynomial Then, the entire spectrum is corrected:
[0035] In the formula, The corrected spectral sequence after baseline removal; This is the original spectral sequence.
[0036] Furthermore, this method can effectively compensate for background deviations caused by spectral drift, making the spectral intensity more stable.
[0037] In one possible implementation, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: performing a standard normal variable transformation on the baseline-corrected mid-infrared spectral data to obtain standardized mid-infrared spectral data.
[0038] In a specific example, the Standard Normal Variable Transform (SNV) is used to eliminate scale inconsistencies caused by differences in optical path length, particle scattering, or changes in sample concentration. This invention performs SNV normalization on each spectrum.
[0039] Furthermore, let the absorbance sequence of a single spectrum be... ,in Indicates the first The spectrum in the first The absorbance values at each wavelength. The standardized result is:
[0040] In the formula, For the first The mean of the spectrum, Standard deviation; This is the standardized spectral vector.
[0041] Furthermore, SNV processing can eliminate the scattering effect on the sample surface and the difference in detection optical path, making the spectra of different samples comparable.
[0042] In one possible implementation, the preprocessing and feature extraction of the mid-infrared spectral data to obtain feature data includes: performing principal component analysis on the standardized mid-infrared spectral data to obtain feature data.
[0043] In a specific example, principal component analysis (PCA), after preprocessing, yields the spectral feature matrix. It still contains a lot of redundant information. Let be the set of real numbers, indicating that the elements in the matrix are real numbers; This represents the number of samples (i.e., the number of spectra). The number of variables (i.e., the number of wavelength points in each spectrum). Principal component analysis (PCA) was used to reduce dimensionality and extract key features.
[0044] Furthermore, the basic idea of PCA is to transform the original correlated variables into a new set of linearly independent variables (principal components) through linear transformation, while preserving the main variances in the data as much as possible. Its calculation process is as follows: Furthermore, the spectral feature matrix is centered as follows:
[0045] In the formula, The centered spectral data matrix; This is the preprocessed spectral data matrix; It is the mean matrix; Furthermore, the covariance matrix is calculated.
[0046] In the formula, This is the covariance matrix calculated from the centered spectral data; This represents the number of samples (i.e., the number of spectra). The centered spectral data matrix; Furthermore, eigenvalue decomposition is performed on the covariance matrix:
[0047] In the formula, , is the eigenvalue matrix, arranged in descending order; This is the corresponding eigenvector matrix; where is the eigenvalue of the covariance matrix.
[0048] Furthermore, by selecting the first t principal components whose cumulative variance contribution rate reaches 90% or more, the dimensionality-reduced feature matrix is obtained:
[0049] In the formula, For the score matrix, It is the load matrix composed of the first t eigenvectors.
[0050] Furthermore, PCA dimensionality reduction not only preserves the most representative feature information in the spectrum but also significantly reduces the feature dimension, thereby reducing noise interference and providing stable input data for subsequent regression modeling (multivariate regression, support vector regression, neural networks, etc.).
[0051] In one possible implementation, the physicochemical parameters of the oil sample to be tested include oxidation value, sulfidation value, moisture content, acid value, and alkalinity value.
[0052] In one possible implementation, the pre-trained model includes a pre-trained multivariate regression model, a pre-trained support vector regression model, and a pre-trained neural network model.
[0053] In a specific example, after extracting features from spectral data, it is necessary to establish a quantitative mapping model between spectral features and the physicochemical parameters of oil products to predict unknown samples. This invention uses oil samples with known components as samples, maps their spectral feature vectors to experimentally measured physicochemical parameters (such as oxidation value, sulfide value, moisture content, acid value, and alkalinity value), and constructs a prediction model.
[0054] In a specific example, to verify the reliability of the model, the sample dataset was randomly divided into a training set (80%) and a validation set (20%) in an 8:2 ratio.
[0055] In a specific example, the training set is used to learn the model parameters, and the validation set is used to evaluate the model's predictive performance on unknown samples. After the model is trained, the model's predictive accuracy is evaluated using metrics such as mean squared error (MSE), correlation coefficient (R²), and residual analysis to determine the optimal modeling parameters.
[0056] In one possible implementation, the selection of the pre-trained model based on the target parameters and the number of channels includes: if the number of channels of the photodetector is N1 to N2 and the target parameters are oxidation value and sulfidation value, then a pre-trained multiple regression model is selected, where N1 to N2 is 4 to 8.
[0057] In a specific example, this method establishes a latent structural model between high-dimensional spectral variables and target variables, making it suitable for scenarios with strong variable correlations.
[0058] Furthermore, let the spectral feature matrix be...
[0059] In the formula, n is the number of samples, and m is the feature dimension; The spectral feature matrix has dimensions n×m; This is the first feature value of the first sample; This is the m-th feature value of the first sample; It is the first feature value of the nth sample; It is the m-th feature value of the n-th sample; Furthermore, the corresponding vector of physicochemical indicators for oil products is:
[0060] In the formula, This represents a vector of physicochemical properties of oil products. These correspond to the true values of the physicochemical indicators of the 1st, 2nd, ..., nth samples, respectively. Furthermore, the linear regression model can be expressed as:
[0061] In the formula, w is the regression coefficient vector, and b is the bias term; Let be the predicted value of the physicochemical index for the i-th sample; Let be the feature vector of the i-th sample.
[0062] Furthermore, the model parameters are obtained by minimizing the sum of squared prediction errors:
[0063] In the formula, Let be the true value of the physicochemical index of the i-th sample; The number of samples; Furthermore, in partial least squares regression (PLSR), the spectral feature matrix X and the response matrix Y are simultaneously decomposed into latent variables:
[0064]
[0065] In the formula, For the score matrix, , These are the first load matrix and the second load matrix, respectively, and E and F are the first residual matrix and the second residual matrix, respectively.
[0066] Furthermore, by iteratively extracting the latent variable T, the covariance of X and Y can be maximized, thereby establishing the prediction equation:
[0067] In the formula, X is the regression coefficient matrix, and X is the spectral feature matrix. After training, it can be used to predict the composition of unknown samples.
[0068] Furthermore, after the model training is complete, the validation set samples are input into the model to calculate the predicted values. and the actual value Compare and calculate the coefficient of determination. With root mean square error (RMSE):
[0069]
[0070] In the formula, For the first The true value of each sample; For the first Predicted values for each sample; The average of the true values of all samples; The total number of samples in the validation set; For sample index; when A low RMSE indicates a good model fit.
[0071] In one possible implementation, the selection of the pre-trained model based on the target parameters and the number of channels includes: if the number of channels of the photodetector is N3 and the target parameters are acid value and alkaline value, then a pre-trained support vector regression model is selected, where N3 is 128.
[0072] In a specific example, the Support Vector Regression (SVR) model, compared to the traditional linear regression method, effectively addresses the nonlinear coupling between spectral features and sample noise, exhibiting better robustness. By appropriately selecting the kernel function and hyperparameters, the model can achieve high-precision predictions for different oil types, providing reliable algorithmic support for oil quality testing.
[0073] Furthermore, assume the sample set is ,in The spectral feature vector is the result of feature extraction or dimensionality reduction. These are the corresponding physicochemical index values of the oil. Support vector regression introduces a nonlinear mapping function. The input data is mapped to a high-dimensional feature space, and a regression function is established in this space:
[0074] In the formula, For the regression coefficient vector, For bias terms, For the prediction function of support vector regression, For feature vectors High-dimensional feature representation after nonlinear mapping This is the input feature vector for a single sample.
[0075] Furthermore, the goal of model training is to ensure that the prediction error does not exceed a given threshold. Under the given conditions, the model should have the highest reliability. The optimization problem can be expressed as:
[0076]
[0077] In the formula, C is the penalty factor, which is used to balance model complexity and training error; The first and second slack variables are used to measure excess. Error in insensitive intervals; For sample index; The number of samples; For model complexity; For the first The true value of each sample; For the first Feature vector of each sample High-dimensional representation after nonlinear mapping; This is an insensitive loss parameter.
[0078] Furthermore, since the relationship between oil spectra and physicochemical parameters is usually nonlinear, kernel functions are used to implicitly map the original input to a high-dimensional feature space. A commonly used radial basis function (RBF) is defined as follows:
[0079] In the formula, The radial basis function kernel value is used to measure the similarity between two samples in a high-dimensional feature space. For the first The feature vector of each sample For the first The feature vector of each sample It is an exponential function. The kernel width parameter determines the model's sensitivity to local features.
[0080] Furthermore, the final regression prediction function can be written as:
[0081] In the formula, These are the first and second Lagrange multipliers, which take non-zero values only when the sample point is a support vector. The kernel function value represents the training sample. Compared with the predicted sample The similarity between them; For the sample Predicted physicochemical index values; For sample index; This represents the number of samples.
[0082] Furthermore, cross-validation (5-fold cross-validation) is used in the training set to select the optimal kernel parameters. With penalty factor Then, the model's predictive performance is evaluated using a validation set.
[0083] Furthermore, the evaluation indicators include , , Furthermore, The calculation formula is:
[0084] In the formula, The total number of samples; For the first The actual physicochemical index values of each sample; For the first Predicted physicochemical index values for each sample; For the first The absolute error of prediction for each sample; Furthermore, when the validation set error converges and there is no overfitting, the model is considered to have good reliability.
[0085] In one possible implementation, selecting the pre-trained model based on the target parameter and the number of channels includes: if the number of channels of the photodetector is N3 and the target parameter is moisture content, then a pre-trained neural network model is selected, where N3 is 128.
[0086] In a specific example, when the prediction results of the two models mentioned above are not ideal, a feedforward multilayer perceptron neural network (BP neural network) is further used to model the nonlinear relationship between the spectrum and oil components. Its structure includes an input layer, several hidden layers, and an output layer. The number of nodes in the input layer corresponds to the number of spectral variables after feature extraction; the hidden layers are used to extract the nonlinear combination relationship of the input features, and the number of layers and nodes is determined based on empirical formulas or cross-validation results; the number of nodes in the output layer corresponds to the number of predicted physicochemical indicators of the oil.
[0087] Furthermore, let the input layer feature vector be... , Let be the 1st, 2nd, ..., mth eigenvalues of the input feature vector; the computation process of the l-th layer of the network is as follows:
[0088] In the formula, Here is the weight matrix for the l-th layer; It is the bias vector; For activation functions; That is, the input feature vector; This is the weighted input for the l-th layer; This is the activation output of neurons in layer 1.
[0089] Furthermore, the output layer provides the prediction results:
[0090] In the formula, is the activation function of the output layer; L is the total number of layers in the neural network.
[0091] Furthermore, model parameters Optimization is achieved by minimizing the mean squared error (MSE).
[0092]
[0093] In the formula, The value of the loss function; The total number of samples; For the first The true value of each sample; For the first Predicted values for each sample; Furthermore, the gradient is calculated using the backpropagation (BP) algorithm:
[0094]
[0095] In the formula, This represents the Hadamard product (element-by-element multiplication). The derivative of the activation function. The loss function is applied to the weight matrix of the l-th layer. gradient, For the error term of the l-th layer, This is the weight matrix for the (l+1)th layer. This is the error term for the (l+1)th layer.
[0096] Furthermore, the parameter updates employ an adaptive optimization algorithm:
[0097] In the formula, This is the learning rate.
[0098] Furthermore, after training, the spectral characteristics of the unknown oil sample are input. The predicted output is calculated through forward propagation:
[0099] In the formula, The predicted values of the physicochemical properties of the oil are as follows: This is the mapping function for the entire forward propagation process. Let be the input feature vector of the sample to be predicted. This is the set of weight parameters for a neural network. This is the set of bias parameters for the neural network.
[0100] Furthermore, the estimated values of the physicochemical parameters of the oil can be obtained.
[0101] Furthermore, during the training phase, 80% of the data is used for learning, and 20% is used for validation. If the validation set loss shows an upward trend after several iterations, an early stopping strategy is implemented to prevent overfitting. Finally, regression analysis is performed on the test results. If the predicted values are highly linearly correlated with the true values, it indicates that the neural network modeling is accurate and reliable.
[0102] In a specific example, after completing spectral preprocessing, feature extraction, and model building, the trained model can be used to quickly and accurately predict the physicochemical parameters of unknown lubricating oil samples. The system ultimately outputs the prediction results to the host computer interface, enabling real-time monitoring and evaluation of lubricating oil products. To ensure long-term prediction accuracy, the system supports regular model updates and maintenance. When the type of lubricating oil changes or new degradation products appear, the model can be incrementally trained or retrained by adding new calibration samples, ensuring the system maintains optimal performance.
[0103] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.
Claims
1. A method for detecting the physicochemical parameters of lubricating oil based on multi-model fusion, characterized in that, include: Acquire multi-channel mid-infrared spectral data of the oil sample to be tested; The mid-infrared spectral data are preprocessed and feature extracted to obtain feature data; Select the pre-trained model based on the target parameters and the number of channels; The feature data is input into a pre-trained model to obtain the physicochemical parameters of the oil sample to be tested.
2. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 1, characterized in that, The feature data obtained by preprocessing and feature extraction of the mid-infrared spectral data includes: The mid-infrared spectral data is filtered to obtain filtered mid-infrared spectral data.
3. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 2, characterized in that, The feature data obtained by preprocessing and feature extraction of the mid-infrared spectral data includes: Baseline correction is performed on the filtered mid-infrared spectral data to obtain baseline-corrected mid-infrared spectral data.
4. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 3, characterized in that, The feature data obtained by preprocessing and feature extraction of the mid-infrared spectral data includes: The baseline-corrected mid-infrared spectral data are subjected to a standard normal variable transformation to obtain standardized mid-infrared spectral data.
5. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 4, characterized in that, The feature data obtained by preprocessing and feature extraction of the mid-infrared spectral data includes: Characteristic data are obtained by performing principal component analysis on the standardized mid-infrared spectral data.
6. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 5, characterized in that, The pre-trained models include pre-trained multivariate regression models, pre-trained support vector regression models, and pre-trained neural network models.
7. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 6, characterized in that, The physicochemical parameters of the oil sample to be tested include oxidation value, sulfidation value, moisture content, acid value, and alkalinity value.
8. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 7, characterized in that, The selection of the pre-trained model based on the target parameters and the number of channels includes: If the number of channels in the photodetector is N1 to N2 and the target parameters are oxidation value and sulfidation value, then a pre-trained multiple regression model is selected.
9. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 7, characterized in that, The selection of the pre-trained model based on the target parameters and the number of channels includes: If the number of channels in the photodetector is N3 and the target parameters are acid value and alkalinity value, then a pre-trained support vector regression model is selected.
10. The method for detecting physicochemical parameters of lubricating oil based on multi-model fusion according to claim 8, characterized in that, The selection of the pre-trained model based on the target parameters and the number of channels includes: If the number of channels in the photodetector is N3 and the target parameter is moisture content, then a pre-trained neural network model is selected.