Method for monitoring the health of engine oil of a vehicle

Abstract

The present invention relates to a method for monitoring the engine oil health condition of a vehicle, the method comprising the steps of: obtaining input operational data, the input operational data being a subset of engine related operational data; predicting an oil quality grade characterizing the engine oil health condition using a trained model based on the obtained input operational data; and outputting the predicted oil quality grade, wherein the model input of the trained model is filtered from the engine related operational data by performing a correlation analysis of the engine related operational data and associated engine oil periodic oil sample analysis test results.

Description

Technical Field

[0001] This invention relates to a method for monitoring the health status of engine oil in a vehicle. Background Technology

[0002] The engine is the primary power source of a vehicle. Vehicles can be, for example, construction machinery in mines or on construction sites, or motor vehicles on roads, such as trucks and passenger cars. Keeping the engine in good working order is crucial, and engine oil plays a key role in the operation of the internal combustion engine. The main functions of engine oil include lubrication, cooling, cleaning, sealing, protection, noise reduction, and fuel efficiency. However, with increased operating time, the health of engine oil gradually deteriorates. Under extremely harsh operating environments or high vehicle loads, engine oil may even degrade prematurely. Therefore, regularly checking and changing engine oil at predetermined intervals is essential to ensure the effective implementation of these functions and to maintain the overall health and performance of the engine.

[0003] However, in the field of construction machinery, it is often difficult to monitor the health of engine oil in real time. The common practice is to use scheduled oil sampling (SOS). But this method requires first obtaining an oil sample from the engine and then sending it to an SOS laboratory for analysis. This is not only time-consuming (about two weeks), but also very expensive.

[0004] Another approach is to install an oil quality sensor in the vehicle. However, this requires not only adding the sensor but also numerous accessories such as brackets, wiring harnesses, and displays, making it very expensive, and end customers are usually unwilling to purchase it. Summary of the Invention

[0005] Therefore, the object of the present invention is to provide a method for monitoring the health status of engine oil in a vehicle, which overcomes at least some of the aforementioned defects of the prior art.

[0006] This objective is achieved through a method having the following characteristics.

[0007] According to one aspect of the present invention, a method for monitoring the health status of engine oil in a vehicle is provided, the method comprising the steps of: acquiring input operating data, said input operating data being a portion of engine-related operating data; predicting an oil grade characterizing the health status of the engine oil using a trained model based on the acquired input operating data; and outputting the predicted oil grade, wherein the model input of said trained model is selected from the engine-related operating data by performing a correlation analysis on the engine-related operating data and the results of associated periodic oil sample analysis (SOS) tests of the engine oil.

[0008] The method of this invention proposes to establish, calibrate, and standardize a supervised machine learning classification model based on actual sampled data from the Caterpillar SOS platform database. This model is driven to monitor the health status of engine oil. Besides the vehicle's existing sensors, this method requires no additional physical sensors, thus reducing commercialization costs. Furthermore, by combining the classification model with correlation analysis, the most relevant operational data to engine oil health status can be selected from massive amounts of vehicle operational data as input data for the classification model. The most relevant data to the output data can then be further selected from the input data for subsequent feature extraction and classification, saving prediction time and improving the prediction accuracy of the classification model.

[0009] According to one embodiment, the input operating data includes first data and second data. The first data is selected from the following: total vehicle operating hours, engine speed, engine coolant temperature, actual engine torque, engine friction torque, and inlet and outlet temperatures of the exhaust aftertreatment device. The second data is selected from the following: vehicle model, engine model, and engine oil type and specification. The advantage of this approach is that it significantly reduces the number of model input data points, lowers the dimensionality of the model input, and simplifies the model structure.

[0010] According to another embodiment, the input operating data is detected by corresponding sensors installed on the vehicle, and in particular, the input operating data is detected continuously. The advantage here is that the input operating data can be conveniently acquired at any time for monitoring engine oil health.

[0011] According to another embodiment, the correlation analysis is performed using the Pearson correlation coefficient, which is calculated using the following formula:

[0012]

[0013] Where i is the number of observations, X i Y i Let be the i-th observation values ​​of two variables X and Y, respectively. These are the average values ​​of two variables, where variable X represents the engine-related operating data, and variable Y represents the SOS test result. An advantage of this approach is that correlation analysis can be easily performed.

[0014] Here, the engine-related operating data and the associated SOS test results are detected at the same time. The advantage here is that correlation analysis can be correctly performed.

[0015] According to another embodiment, operational data whose absolute value of the Pearson correlation coefficient with the SOS test results is greater than a preset correlation threshold is selected as the model input. The preset correlation threshold is 0.9, 0.8, 0.7, 0.6, 0.5, 0.4, or 0.3. The advantage here is that operational data with high correlation can be identified to accurately predict the health status of the engine oil.

[0016] According to another embodiment, the SOS test results include one or more, especially all, of the following indicators: oil viscosity, contamination level, impurity level, and oxidation level.

[0017] According to another embodiment, the trained model is trained based on selected operational data as training input data and oil quality grades obtained from associated SOS test results as training output data. The advantage here is that a suitable model can be built and trained to predict the health status of engine oil.

[0018] According to another embodiment, training the trained model includes the following steps:

[0019] Obtain engine-related operating data;

[0020] Obtain SOS test results and determine the oil grade;

[0021] A correlation analysis was performed on the acquired runtime data and the acquired SOS test results to select the first runtime data as the model input;

[0022] A classification model is trained using a supervised machine learning algorithm based on the determined first operating data and the oil grade associated with the first operating data to obtain the trained model.

[0023] Here, the SOS test result used to label the oil grade associated with the first operating data was detected at the same time as the first operating data.

[0024] According to another embodiment, multiple classification models are simultaneously trained based on determined first operating data and oil grades associated with the first operating data. The performance of the multiple classification models is evaluated on a test vehicle, and the best-performing classification model is selected as the trained model. The advantage here is that an optimal predictive model can be trained to monitor engine oil health.

[0025] Advantageously, the classification model is one or more of decision trees, random forests, support vector machines, convolutional neural networks, and recurrent neural networks, and / or the classification model with the minimum loss function is selected as the trained model. For example, mean squared error or mean absolute error can be used as the loss function.

[0026] According to another embodiment, the oil grade includes green, yellow and red grades; or the oil grade includes "normal", "requires maintenance" and "requires replacement".

[0027] According to another embodiment, the vehicle is construction machinery, such as a mining truck or a heavy-duty truck.

[0028] According to another aspect of the invention, a computer program product is provided, comprising instructions that are implemented to perform the method of the invention.

[0029] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit the scope of protection of the present invention. Attached Figure Description

[0030] The invention will now be described in detail with reference to the accompanying drawings. In the drawings,

[0031] Figure 1 A flowchart illustrating a method for training an engine oil health status prediction model according to the present invention is shown.

[0032] Figure 2 A flowchart of a method for monitoring engine oil health according to the present invention is shown. Detailed Implementation

[0033] Figure 1 A flowchart of a model training method according to the present invention is shown, which is used to train a model according to the present invention for predicting engine oil health status based on engine operating data. Predicting engine oil health status using a model based on engine operating data is an equipment health monitoring technology. It is well known that the health status of engine oil largely depends on the engine's operating conditions. However, there is a large amount of engine-related operating data (up to 20 or more). If all of this operating data were used as model input to predict engine oil health status, the prediction process would be very complex. Furthermore, some of this operating data is not closely related to engine oil health status; considering this data would increase data noise, thus adversely affecting the accuracy of engine oil health status prediction. Therefore, it is desirable to have an engine oil health status prediction model that includes fewer operating data as model input while maintaining high prediction accuracy. To this end, the present invention proposes a method for determining and training such an engine oil health status prediction model.

[0034] First, in step S101, engine-related operating data is acquired. This operating data can be obtained from an existing database, such as Caterpillar's measurement database. This measurement database can be stored, for example, on a server external to the vehicle, such as in a backend system. For example, in the case of mining construction machinery, to establish this measurement database, engine-related operating data can be periodically collected from mining trucks, loading vehicles, unloading vehicles, auxiliary vehicles, and other construction machinery, and this operating data can be stored in the measurement database. For example, this operating data can be automatically or transmitted from the corresponding vehicle to the server and stored in the measurement database thereby via a wired or wireless communication link by the vehicle operator. Operating data may include parameters such as vehicle model, engine model, engine oil type and specification, service meter hours (SMH), idling time, engine speed, vehicle speed, fuel pressure, intake pressure or vacuum, throttle opening, engine coolant temperature, actual engine torque, engine friction torque, exhaust gas temperature, inlet and outlet temperatures of the SCR catalytic converter or exhaust aftertreatment device, maximum engine power, and maximum engine torque. The values ​​of these operational data can be obtained from the corresponding sensors that are already installed on the respective vehicles.

[0035] In step S102, the SOS test results are obtained, and the oil grade is determined. Here, the oil grade is determined manually based on the SOS test results and a preset value range, by labeling the health condition of the engine oil. The SOS test results can be obtained from an existing database, such as Caterpillar's SOS database. To establish this SOS database, oil sampling ports are provided in the engines of multiple test vehicles. Oil samples are periodically extracted from these ports, sent to the SOS laboratory for analysis, and the results are stored in the SOS database. SOS test results may include indicators such as oil viscosity, contamination level, impurity level, and oxidation level. Viscosity refers to the flow resistance of the engine oil at a specific temperature. Contamination level refers to the number and size of particles contained in the oil, also known as particle size. Impurity level refers to whether the oil contains impurities that cause abnormal wear. Oxidation level refers to the phenomenon where the oil molecular structure changes under the influence of high temperature, oxygen, and moisture, leading to a decrease in lubrication performance. Oil grades may be classified, for example, as green, yellow, and red grades. A green rating indicates that the oil is in perfect condition, free from contamination and wear impurities. A yellow rating indicates that some monitored indicators are abnormal, such as increased contamination levels, minor impurities, or decreased oil quality. A red rating indicates that some monitored indicators exceed specifications, requiring action. Oil quality can also be classified in other ways, such as "normal," "requires maintenance," and "requires replacement." "Normal" means the oil is in good condition and requires no action. "Requires maintenance" means the oil quality is deteriorating but still within specifications, requiring close monitoring or maintenance such as cleaning or filtering. "Requires replacement" means the oil quality is very poor and needs to be replaced. Accordingly, each indicator in the SOS test results has a preset value range corresponding to the oil quality level. In other words, each indicator is divided into three value ranges, corresponding to green, yellow, and red levels, or "normal," "requires maintenance," and "requires replacement" levels, respectively. The division between the value ranges of each indicator can be determined according to various international or national standards, such as the ISO 4406 standard used for contamination levels. Of course, the boundary values ​​for each range can also be predetermined by the labelers, for example, based on experience or requirements. Obviously, the number of oil grades can be two, four, five, or more, instead of three. Accordingly, a value range equal to the number of oil grades is preset for each indicator.

[0036] Labelling personnel determine and label the oil grade of each oil sample by comparing the values ​​of various indicators in the SOS test results with corresponding preset value ranges. For example, it can be stipulated that when all indicator values ​​fall within the range indicating good or normal oil quality, the oil grade is labeled as green or "normal"; when at least one indicator value falls within the range indicating poor oil quality, the oil grade is labeled as red or "replacement required"; all other cases (i.e., at least one indicator value falls within the middle range while the values ​​of the others fall within the range indicating normal oil quality) are classified as yellow or "maintenance required." The oil grade can be used as an output of the model.

[0037] Obviously, the order of steps S101 and S102 can be interchanged, that is, step S102 can be performed first, followed by step S101. Alternatively, steps S101 and S102 can be performed simultaneously.

[0038] Next, in step S103, a correlation analysis is performed on the acquired running data and the acquired SOS test results to filter out the running data used as model input, i.e., the first running data. The correlation analysis algorithm measures the overall error between two variables by calculating the covariance. If the trends of the two variables are consistent, the covariance is positive, indicating a positive correlation. If the trends of the two variables are opposite, the covariance is negative, indicating a negative correlation. If the two variables are independent, the covariance is 0, indicating no correlation. Here, the correlation can be calculated using, for example, the Pearson correlation coefficient. The Pearson correlation coefficient measures the linear correlation between two continuous variables, with a value range of [-1, 1], where 0 indicates no correlation, 1 indicates perfect positive correlation, and -1 indicates perfect negative correlation. The closer the correlation coefficient is to 1 or -1, the stronger the correlation between the two variables. The formula for calculating the Pearson correlation coefficient is:

[0039]

[0040] Where cov(X,Y) represents the Pearson correlation coefficient between variables X and Y, X i and Y i These are the i-th observations of variables X and Y, respectively. and are the sample means of variables X and Y, respectively, and n is the number of observations.

[0041] Therefore, to calculate the Pearson correlation coefficient, firstly, based on the operating data obtained in step S101 and the SOS test results obtained in step S102, interrelated operating data sets and SOS test results are determined. These can be referred to as associated operating data sets and associated SOS test results. Here, "interrelated" between operating data sets and SOS test results means that the time point / time period for detecting a set of operating data for an engine is the same as the time point / time period for extracting the oil sample to be analyzed from that engine. That is, a set of operating data and an SOS test result are measured for the same engine at the same time point / time period. For this purpose, SOS test results and operating data sets with the same combination of vehicle model and / or engine model and / or engine oil type and specification, as well as the same total vehicle operating hours, can be identified from the obtained SOS test results and obtained operating data as associated SOS test results and associated operating data sets. In other words, for a specific combination of vehicle model and / or engine model and / or engine oil type and specification, the time series of all operating data in the associated operating data set as a function of vehicle operating hours, as well as the time series of all indicators in the associated SOS test results as a function of vehicle operating hours, can be obtained. It should be noted that at this point, the associated operating data set still includes all acquired engine-related operating data.

[0042] Then, for each combination of vehicle model and / or engine model and / or engine oil type and specification (i.e., the second operating data), Pearson correlation coefficients between each operating data point and each indicator can be calculated based on the time series of each operating data point in the associated operating data group and the time series of each indicator in the associated SOS test results. Based on the absolute values ​​of these Pearson correlation coefficients, the correlation between each operating data point and each indicator can be categorized into different levels. For example, it can be stipulated that if the absolute value of the Pearson correlation coefficient between the corresponding operating data point and the corresponding indicator is between 0.8 and 1, the correlation between the corresponding operating data point and the corresponding indicator is considered high; if the absolute value is between 0.5 and 0.8, they are considered moderately correlated; if the absolute value is between 0.3 and 0.5, they are considered weakly correlated; and if the absolute value is less than 0.3, they are considered uncorrelated. Therefore, by setting a correlation threshold, operating data with an absolute correlation coefficient greater than the threshold can be selected as the first operating data to be input into the model. Different correlation thresholds can be selected to filter out different numbers of first operating data points. In other words, the first operating data is a subset of the associated operating data group. The preset relevance threshold can be, for example, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 or 0.3.

[0043] Therefore, in step S104, a supervised machine learning algorithm is used to train a classification model based on the determined first operating data and the oil grade associated with the first operating data, so as to establish a mapping relationship between the operating data as the model input and the oil grade as the model output. After the first operating data is determined in step S103, the training dataset of the model can be determined. Each training data in the training dataset includes a training input data and a training output data. The training input data includes the first operating data, i.e., the operating data filtered by correlation analysis in the associated operating data group, and the second operating data corresponding to the first operating data, i.e., the combination of vehicle model and / or engine model and / or engine oil type and specification in the associated operating data group. The training output data is the oil grade labeled by the SOS test results associated with the first operating data. Here, the SOS test results associated with the first operating data used to label the oil grade were detected at the same time point / time period as the first operating data. Here, the second operating data can act as a model switch, that is, different combinations of model parameters are trained for different second operating data. Of course, it's also possible to train a model with parameter combinations applicable to different second-run datasets. Based on this training dataset, a classification model is trained using supervised learning algorithms. The classification model preprocesses the input data, such as through data cleaning, and then performs feature extraction to extract features and establish a mapping between the extracted features and the classification. In supervised learning, given input data and answer labels, the model learns the algorithm for the relationship between features and classification. The trained model can then be used to predict new data.

[0044] Here, multiple classification models, especially various types of classification models, can be trained simultaneously using the same training dataset, such as decision trees, random forests, support vector machines (SVMs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs). Training a classification model involves adjusting its parameters to maximize classification accuracy or minimize prediction error. Random forests and support vector machines are two classification models briefly introduced below.

[0045] Random forests consist of multiple decision trees, and they perform classification tasks by constructing multiple decision trees and combining their results. Random forests employ a bootstrap sampling method: for the original training dataset, samples are randomly drawn with replacement to construct multiple training subsets of the same size as the original dataset but with different sample content. For each sampled training subset, a decision tree is constructed. During the construction of the decision trees, random forests also introduce random feature selection. At each node split, instead of selecting the best splitting feature from all features, a random subset of features is chosen, thus increasing the diversity of the decision trees. Each decision tree makes a classification prediction for the test sample. Finally, the predicted classes of all decision trees are counted, and the class with the most votes is the final classification result of the random forest.

[0046] When using the random forest algorithm to predict engine oil health, establishing an optimization objective function helps adjust model parameters to achieve better performance. Mean Squared Error (MSE) can be used as the loss function, and its mathematical expression is:

[0047]

[0048] Where n is the number of samples, y i This is the actual value. These are model predictions.

[0049] Alternatively, mean absolute error (MAE) can be used as the loss function:

[0050]

[0051] Where n is the number of samples, y i This is the actual value. These are model predictions.

[0052] For data-driven engine oil health prediction, random forests are a good choice because they offer the following advantages:

[0053] Accuracy: Random forests typically offer high accuracy, especially for complex datasets and feature spaces;

[0054] Robustness: Random forests are robust to outliers, noise, and missing data, and can handle diverse and imperfect data.

[0055] Feature importance: Random forests can be used to assess the importance of features, helping to understand which features are most critical for predicting engine oil health.

[0056] Handling large amounts of data: Random forests can effectively handle large amounts of data and high-dimensional feature spaces, making them suitable for many practical engineering problems; and

[0057] Preventing overfitting: Random forests reduce the risk of overfitting and improve generalization ability by integrating multiple decision trees and introducing randomness.

[0058] Support Vector Machines (SVMs) are a supervised learning classification algorithm whose main goal is to find an optimal hyperplane in the feature space that separates data points of different classes as much as possible. This hyperplane can be linear or non-linear. When the data is linearly separable, a linear hyperplane can be found directly. For non-linear data, a kernel function maps the data to a higher-dimensional space, making the data linearly separable in that space. The optimization objective of SVMs is to maximize the distance (margin) from the hyperplane to the nearest data points in each class and minimize classification error. The advantages of SVMs are:

[0059] Strong generalization ability: Support vector machines perform well when dealing with small samples and high-dimensional data, and can effectively avoid overfitting;

[0060] Capable of handling nonlinear problems: With the help of kernel functions, it can flexibly handle various nonlinear data distributions; and

[0061] Good robustness: It has a certain resistance to noise and outliers in the data. Because it mainly focuses on support vectors, i.e., data points located near the margin boundary, a small amount of noisy data has a relatively small impact on the determination of the hyperplane.

[0062] Steps S103 to S104 can be performed cyclically. By selecting different relevance thresholds, different amounts of first running data are selected as model inputs. That is, different dimensions of input data are selected or determined for the model. Different model parameters can be adjusted for the same type of classification model to achieve the optimization goal.

[0063] After training each classification model, in step S105, the performance of each model is verified and evaluated to select the model that can accurately predict the health status of the engine oil. Here, multiple classification models trained in the laboratory are deployed to a test vehicle, and oil samples are periodically extracted from a pre-installed oil sampling port on the test vehicle for SOS analysis. The SOS test results are compared with the model prediction results to verify the model, thereby finding the best-performing model, for example, the one with the smallest loss function, as the final training result. Here, the final training result may be, for example, a random forest or a support vector machine. The determined first operating data includes, for example, the total vehicle operating hours, engine speed, engine coolant temperature, actual engine torque, engine friction torque, and the inlet and outlet temperatures of the SCR catalytic converter or exhaust aftertreatment device.

[0064] The finally trained model can be used to predict the engine oil grade of a target vehicle to infer the engine oil's health condition. To this end, the finally trained model can be integrated into the target vehicle's controller or a real-time monitoring system to continuously monitor the engine's operating data and predict the engine oil's health condition in a timely manner.

[0065] Figure 2 A flowchart of a method for monitoring engine oil health according to the present invention is shown. In step S201, input operating data is acquired, which is a portion of engine-related operating data. The input operating data can be detected by various sensors installed on the vehicle, particularly continuously. The input operating data may include first data and second data. The first data may be selected from the following: total vehicle operating hours, engine speed, engine coolant temperature, actual engine torque, engine friction torque, and inlet and outlet temperatures of the SCR catalytic converter (or exhaust aftertreatment device). The second data may be selected from the following: vehicle model, engine model, and engine oil type and specification. In particular, the second data can act as a model switch, with different second data triggering different models or different combinations of model parameters. Here, the quantity and type of the input operating data (i.e., how many and which of the engine-related operating data are included in the input operating data) are determined from the engine-related operating data according to the above reference. Figure 1 The methods described were selected through correlation analysis.

[0066] In step S202, based on the acquired input operating data, a trained model is used to predict the oil grade characterizing the engine oil's health condition. The trained model has been prepared according to the above-mentioned reference... Figure 1 The described method involved training and selection. Specifically, the model input to the trained model was selected from the engine-related operating data through correlation analysis of the engine-related operating data and the associated SOS test results of the engine oil. The model was trained using the selected operating data as training input and the oil grade, labeled with the associated SOS test results, as training output. The SOS test results included one or more, and especially all, of the following indicators: oil viscosity, contamination level, impurity level, and oxidation degree. Here, the correlation analysis used the Pearson correlation coefficient, which was calculated using the following formula:

[0067]

[0068] Where i is the number of observations, Xi Y i Let be the i-th observation values ​​of two variables X and Y, respectively. These are the average values ​​of two variables, where variable X is the engine-related operating data and variable Y is the SOS test result. The engine-related operating data and the associated SOS test result were detected at the same time. Operating data whose absolute value of the Pearson correlation coefficient with the SOS test result is greater than a preset correlation threshold are selected as the model input. The preset correlation threshold is, for example, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4, or 0.3.

[0069] The trained model is a random forest or a support vector machine. Oil quality grades include green, yellow, and red grades, or oil quality grades include "normal," "requires maintenance," and "requires replacement."

[0070] In step S203, the predicted fuel grade is output. The predicted fuel grade can be output to the vehicle's display or to a remote operator, for example, in a control room.

[0071] Then, in optional step S204, appropriate treatment measures are taken for the engine oil as necessary, based on the predicted oil grade. For example, if the oil grade is yellow or "maintenance required," maintenance measures such as cleaning or filtering the engine oil are performed, or the remaining service life of the engine oil is estimated. If the oil grade is red or "change required," further testing is conducted to find the root cause of the engine oil's unhealthy condition, or the remaining service life of the engine oil is estimated, or the engine oil is replaced.

[0072] The present invention also relates to a computer program product comprising instructions which are implemented to perform the methods described above.

[0073] Industrial applicability

[0074] The method for training an engine oil health status prediction model and the method for predicting engine oil health status using a trained model based on detected engine-related operating data of the present invention can be applied to various vehicles, especially construction machinery in mining areas and construction sites, or motor vehicles on roads, especially trucks and passenger cars. By performing correlation analysis on the detected operating data and detected SOS test results over a period of time, a smaller number of operating data points most favorable for prediction are selected from a large pool of operating data, thus determining the smaller-dimensional model input data. Then, the model parameters are trained based on the corresponding training dataset, and after actual verification on test vehicles, the best-performing model is finally selected. The finally selected model is then integrated into the target vehicle's controller or a real-time monitoring system such as a backend system to continuously monitor the engine operating data of the target vehicle and predict the engine oil health status in a timely manner.

[0075] The foregoing description is merely an exemplary embodiment relating to the spirit and principles of the present invention. Those skilled in the art will understand that various changes can be made to the described examples without departing from the spirit and principles, and such changes and their various equivalents are contemplated by the inventors and fall within the scope defined by the claims of the present invention.

Claims

1. A method for monitoring the health status of engine oil in a vehicle, the method comprising the following steps: Acquire input operating data, which is a portion of engine-related operating data. Based on the acquired input operating data, a trained model is used to predict the oil grade that characterizes the health status of the engine oil. Output the predicted oil grade. Preferably, appropriate treatment measures are taken for the engine oil based on the predicted oil grade. Its features are, The input to the trained model is selected from the engine-related operating data by performing correlation analysis on the engine-related operating data and the results of periodic oil sample analysis tests of the associated engine oil.

2. The method according to claim 1, characterized in that, The input operating data includes first data and second data. The first data is selected from the following data: total vehicle operating hours, engine speed, engine coolant temperature, actual engine torque, engine friction torque, and inlet and outlet temperatures of the exhaust aftertreatment device. The second data is selected from the following data: vehicle model, engine model, and engine oil type and specification.

3. The method according to claim 1 or 2, characterized in that, The input operating data is detected by corresponding sensors installed on the vehicle, and in particular, the input operating data is detected continuously.

4. The method according to any one of claims 1 to 3, characterized in that, The correlation analysis was performed using the Pearson correlation coefficient, which is calculated using the following formula: Where i is the number of observations, X i Y i Let be the i-th observation values ​​of two variables X and Y, respectively. These are the averages of the two variables. Wherein, variable X is the engine-related operating data, and variable Y is the periodic oil sample analysis test result.

5. The method according to claim 4, characterized in that, The engine-related operating data and the associated periodic oil sample analysis test results were detected at the same time.

6. The method according to claim 4 or 5, characterized in that, The model input is selected from the operational data whose absolute value of the Pearson correlation coefficient with the periodic oil sample analysis test results is greater than a preset correlation threshold. The preset correlation threshold is 0.9, 0.8, 0.7, 0.6, 0.5, 0.4 or 0.

3.

7. The method according to any one of claims 4 to 6, characterized in that, The results of the periodic oil sample analysis test include one or more, and especially all, of the following indicators: oil viscosity, contamination level, impurity level, and oxidation level.

8. The method according to any one of claims 1 to 7, characterized in that, The trained model is trained based on selected operational data as training input data and oil grade obtained from the results of associated periodic oil sample analysis and testing as training output data.

9. The method according to any one of claims 1 to 8, characterized in that, The training of the trained model includes the following steps: Obtain engine-related operating data; Obtain periodic oil sample analysis and testing results, and determine the oil grade; A correlation analysis was performed on the acquired operational data and the results of periodic oil sample analysis to select the first operational data as the model input. A classification model is trained using a supervised machine learning algorithm based on the determined first operating data and the oil grade associated with the first operating data to obtain the trained model.

10. The method according to claim 9, characterized in that, The periodic oil sample analysis test results used to label the oil grade associated with the first operating data were detected at the same time as the first operating data.

11. The method according to claim 9 or 10, characterized in that, Multiple classification models are trained simultaneously based on the determined first operating data and the oil grade associated with the first operating data, and the performance of the multiple classification models is evaluated on a test vehicle. The best performing classification model is selected as the trained model.

12. The method according to any one of claims 9 to 11, characterized in that, The classification model is one or more of decision trees, random forests, support vector machines, convolutional neural networks, and recurrent neural networks, and / or the classification model with the minimum loss function is selected as the trained model.

13. The method according to claim 12, characterized in that, Mean squared error or mean absolute error is used as the loss function.

14. The method according to any one of claims 1 to 13, characterized in that, The vehicle in question is construction machinery, such as a mining truck or a heavy-duty truck. and / or The oil grade includes green, yellow and red grades; or the oil grade includes "normal", "requires maintenance" and "requires replacement".

15. A computer program product comprising instructions that are implemented to perform the method according to any one of claims 1 to 14.

Images