A diesel engine oil change identification method based on Internet of Vehicles big data

Abstract

The present invention belongs to the field of big data processing and analysis technology and precision maintenance of diesel engine commercial vehicles. Specifically, it is a diesel engine oil change identification method based on Internet of Vehicles big data. It includes a feature extraction stage and a model training and prediction stage. The feature extraction stage specifically includes the following steps: step 1-1, data cleaning; step 1-2, extracting the target working condition data set; step 1-3, feature extraction. The model training and prediction stage includes the following steps: step 2-1, constructing a supervised learning model; step 2-2, model result prediction; step 2-3, model error analysis; step 2-4, integrating multiple working condition results; step 2-5, model result display. Through vehicle working condition data, the vehicle working condition and key features reflecting the oil viscosity are constructed to accurately identify the time for the user to change the oil, providing a basic data basis for subsequent precise maintenance recommendations.

Description

Technical Field

[0001] The present invention belongs to the field of big data processing and analysis technology and precise maintenance of diesel engine commercial vehicles, and specifically provides an oil change identification method based on Internet of Vehicles big data. Background Art

[0002] As a crucial lubricant in the engine, engine oil provides cooling, rust and corrosion protection, cushioning, and vibration reduction during operation. Low oil levels or abnormal oil viscosity can significantly impact the vehicle's overall fuel consumption and performance. Therefore, choosing the right timing for oil changes and maintenance is crucial for vehicle owners during daily vehicle use.

[0003] Unlike passenger vehicles, commercial vehicles lack direct oil level measurement points. This makes it difficult for engine manufacturers and OEMs to determine whether owners have performed maintenance at appropriate times. This makes it difficult to accurately locate and attribute failures caused by untimely maintenance. Furthermore, given that the mileage of construction machinery doesn't reflect actual operating conditions, current maintenance recommendations for commercial vehicles are based on engine operating hours. Without access to historical maintenance data, engine manufacturers and OEMs struggle to accurately recommend maintenance intervals if owners don't perform maintenance or repairs at designated service stations. Summary of the Invention

[0004] The purpose of the present invention is to provide a diesel engine oil change identification method based on vehicle network big data to address the above-mentioned shortcomings. Through vehicle operating condition data, the vehicle operating condition and key features reflecting the oil viscosity are constructed to accurately identify the time for users to change the oil, providing a basic data basis for subsequent precise maintenance recommendations.

[0005] The present invention is achieved by taking the following technical solutions:

[0006] A diesel engine oil change identification method based on Internet of Vehicles big data includes a feature extraction stage, a model training and prediction stage.

[0007] The feature extraction stage specifically includes the following steps:

[0008] Step 1-1, data cleaning;

[0009] Step 1-2, extracting the target working condition data set;

[0010] Steps 1-3: Feature extraction.

[0011] The model training and prediction phase includes the following steps:

[0012] Step 2-1: Build a supervised learning model;

[0013] Step 2-2, model result prediction;

[0014] Step 2-3, model error analysis;

[0015] Steps 2-4: Fusion of multiple working condition results;

[0016] Steps 2-5: Display of model results.

[0017] Furthermore, the frequency of the original data collected by the Internet of Vehicles used in step 1-1 is 1 Hz, and there are some unreasonable values ​​or missing values ​​in the data, which need to be processed according to the following rules:

[0018] Rule 1: For some sample retention values, such as 65536, 256, etc., the data in this row is directly discarded; for temperature sampling values ​​≤ -270℃, data is judged as invalid values ​​and is also discarded.

[0019] Rule 2: For missing values, if it is a temperature measurement point with low accuracy, interpolation can be performed based on the previous and next temperature values; if it is not a temperature measurement point, the row of data is directly removed.

[0020] Furthermore, the operating condition data range applicable to steps 1-2 of the identification method is the warm-up process of the diesel engine after it is started every day; the warm-up process is the regular warm-up process of the engine after the diesel engine is started. During this process, the engine is in an idling state and the speed level will be adjusted according to the ambient temperature, usually fixed in 1 to 2 relatively narrow speed ranges; the coolant temperature is usually used as a measurement indicator in the vehicle during the warm-up process.

[0021] In steps 1-2, the conditions for extracting the target working condition are as follows:

[0022] 1-2-1. The vehicle is in idle state, that is, the engine speed is greater than 0 rpm and the vehicle speed is 0 km / h;

[0023] 1-2-2. The engine coolant temperature is higher than 45°C and lower than 55°C. Within this temperature range, the coolant temperature is approximately equal to the oil temperature in the oil circuit.

[0024] After the data processed in steps 1-2-1 and 1-2-2, the engine speed should be concentrated in 1 to 2 speed ranges, which are 790 rpm to 810 rpm and 890 rpm to 910 rpm. The data in the subsequent two speed ranges must be included in the calculation range.

[0025] Furthermore, steps 1-3 extract key variables that can reflect the oil status and aggregate the original 1 Hz data to reduce the deviation caused by data fluctuations. The specific extraction steps are as follows:

[0026] Step 1-3-1: Select the oil pressure measurement point as the key feature reflecting the oil viscosity. In addition, collect the engine speed, coolant temperature, fuel level, and intake air volume as measurement points reflecting the working conditions.

[0027] Step 1-3-2: Aggregate the 1Hz data into 5s data according to time, calculate the average value, 10th percentile value, 90th percentile value, and median value of engine speed, coolant temperature, fuel volume, and intake air volume in four aggregation methods, and obtain 16 features to be input into the model; select the average value of the corresponding oil pressure as the label to be input into the model.

[0028] Furthermore, step 2-1 differentiates the data set according to the different speed ranges of the engine. Specifically, data from 7 days before the date to be analyzed (excluding the day to be analyzed) are selected as the training set, and 16 variables calculated by four aggregation methods, namely the average value, 10th percentile, 90th percentile, and median of the engine speed, coolant temperature, fuel volume, and intake volume obtained in steps 1-3, are used as features. The average oil pressure at the corresponding time is used as a label, and input into a supervised learning model (XGBoost) to obtain model a corresponding to the speed range of 790rpm to 810rpm and model b corresponding to the speed range of 890rpm to 910rpm; during the training process, the hyperparameter settings of model a and model b are consistent; the hyperparameters include learning rate, number of weak classifiers, downsampling ratio, loss function, maximum depth of leaf nodes, L1 regularization term, and L2 regularization term; XGBoost is an open source supervised learning algorithm, which is imported in the form of a program package during use and called using an API or a packaged function to implement training of the machine learning model.

[0029] Furthermore, in step 2-2, the data of the day to be analyzed is used, and the corresponding data is input into model a and model b respectively according to the different engine speed ranges to obtain the predicted 5s average oil pressure value, which is recorded as predicted oil pressure a and predicted oil pressure b; the data of the day refers to the data of the 1Hz warm-up process stage collected on the day, including measurement points such as engine speed, coolant temperature, vehicle speed, intake flow, ambient temperature and oil pressure.

[0030] Furthermore, in the model error analysis of steps 2-3, the predicted oil pressures a and b output by models a and b are subtracted from the actual 5-second average oil pressure values ​​within the corresponding engine speed ranges to obtain the prediction errors for all data on that day. The corresponding errors for all data on the day being analyzed are then averaged and recorded as error a and error b. Theoretically, oil viscosity changes after an oil change. Using a model trained on historical data to predict data after an oil change will result in significant errors.

[0031] Furthermore, in steps 2-4, since there are two operating ranges for the engine speed, it is necessary to weight the average value of the error a and the error b according to the amount of data of the diesel vehicle in the two operating ranges, where the weight is the proportion of the amount of data in the corresponding range of the engine speed to the total amount of data after screening; the error after weighted calculation is used as the final error of the day to be analyzed to determine whether the engine oil has been changed.

[0032] Furthermore, in step 2-5, the error data of multiple consecutive days are calculated, and through the error fluctuations in time series, it is determined whether the engine oil has been changed on the same day, and the result of the final model is output.

[0033] Beneficial effects of the present invention: The method of the present invention utilizes the operating data collected by the Internet of Vehicles to accurately identify the timing of oil replacement. It is the basis for automobile companies to provide users with more value-added services related to precise maintenance, and is also a method for calculating user maintenance records at a low cost.

[0034] 1) Using fewer measurement points can solve the problem of not being able to sense oil changes due to the lack of oil level, effectively making up for the manufacturer's monitoring of unofficial maintenance activities and improving the data foundation for providing users with accurate maintenance cycle recommendations.

[0035] 2) The model is applicable to multi-purpose and multi-model diesel engines. As long as the input measurement points are met, only slight adjustments to the operating conditions corresponding to the engine warm-up data are required to detect whether the engine oil has been changed. This model has strong generalization properties.

[0036] 3) By analyzing a large number of vehicles, we can summarize the common oil pressure conditions before and after oil changes, use oil pressure as the basis for determining whether to change the oil, more accurately calculate the oil change cycle, increase the interval between oil changes for users to reduce costs, or remind users to change the oil as soon as possible to reduce engine wear and tear, thereby reducing users' car use costs from an economic perspective, reducing the risk of engine failure from a safety perspective, and optimizing the user's car experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention will be further described below with reference to the accompanying drawings:

[0038] Figure 1 is a flow chart of the method of the present invention;

[0039] Figure 2 This is a data flow diagram of the model training and prediction stages in an embodiment of the present invention;

[0040] Figure 3 2 is a schematic diagram showing the effect of identifying the oil change time in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Definitions of terms and abbreviations in this invention

[0043] Motor oil: also known as engine lubricant, plays the main roles of lubrication and friction reduction, auxiliary cooling, sealing and leak prevention, rust and corrosion prevention, shock absorption and buffering in the daily operation of the engine. It is known as the "blood" of the car.

[0044] Idle: The state in which the vehicle engine is started but no torque is being transmitted to the wheels. In this condition, the engine speed is greater than 0 and the vehicle speed is equal to 0.

[0045] Warming up: Before entering service, diesel vehicles require the engine temperature to be raised to near normal operating temperature to ensure lubrication. This is a routine process for diesel vehicles. During this process, the vehicle remains in idle mode, the engine speed remains stable within a fixed range of 1-2, and the coolant temperature gradually rises to operating temperature.

[0046] Supervised learning model: refers to an algorithm that builds a corresponding machine learning algorithm based on the characteristics and labels of the data and learns the relationship between features and labels.

[0047] At present, commercial vehicles have basically completed the installation of T-Box and Internet of Vehicles infrastructure, and can collect vehicle operating data in the cloud. Specific embodiment:

[0049] Refer to the attached Figure 1 ,The diesel engine oil change identification method based on Internet of Vehicles big data of the present invention includes a feature extraction stage, a model training and prediction stage;

[0050] Perform data cleaning and feature processing in the feature extraction stage;

[0051] During the model training and prediction phase, the dataset is differentiated based on the engine speed range. Specifically, the data from the seven days before the date to be analyzed (excluding the day to be analyzed) is selected as the training set. This data is then fed into a supervised learning model (XGBoost) for model training, followed by model prediction. The data from the day to be analyzed is then used for model prediction.

[0052] Finally, an error analysis is performed on the prediction results of the above model.

[0053] Refer to the attached Figure 2 The data flow diagram of the model training and prediction phase is as follows:

[0054] 1) Read vehicle operation data to the computing platform;

[0055] 2) Eliminate outliers and invalid data;

[0056] 3) Since model training uses data from seven days (T-7, T-1) and model prediction uses data from one day (T), we need to extract operating conditions from eight days of data. We select two engine speed ranges: coolant temperature between [45°C, 55°C] and engine speed between [790 rpm, 810 rpm] and [890 rpm, 910 rpm], as the engine warm-up conditions for subsequent model training and prediction.

[0057] 4) Using the hot engine data corresponding to [T-7, T-1], the original 1Hz data was aggregated into 5s frequency data. The features to be aggregated included engine speed, coolant temperature, fuel level, and intake air volume. Aggregation methods included mean, 10th percentile, 90th percentile, and median, resulting in a total of 16 features. The label to be aggregated was oil pressure, and the label used for model training was obtained by taking the 5s average oil pressure. Two speed ranges were distinguished and the model was trained using the XGBoost algorithm to obtain oil pressure prediction models a and b.

[0058] 5) Using the hot engine data from day T, the same method is used to aggregate the 5-second operating condition characteristics and average oil pressure. These are input into Model a and Model b for the corresponding speed ranges, respectively, to obtain the predicted oil pressure a and predicted oil pressure b. These are then subtracted from the actual oil pressure a and actual oil pressure b for the corresponding operating condition, and averaged across each speed range to obtain the average prediction error a and average prediction error b of the model.

[0059] 6) Based on the proportion of the two speed ranges in the total warm-up conditions, the weights of the corresponding conditions are obtained, and the weighted average prediction error a and average prediction error b are calculated to obtain the final average error;

[0060] 7) Vehicles with an error value ≥20 are considered to have undergone a sudden change, that is, the oil has been changed; vehicles with an error value ≤20 are considered to have no significant change in the error, that is, the oil has not been changed.

[0061] Figure 3 This is the final output of the example. The horizontal axis represents the date, and the vertical axis represents the oil viscosity composite index (solid line) and the weighted average error (dashed line). By observing data from multiple consecutive days, we can see that the weighted average error experienced a significant fluctuation on September 25, 2021. Comparing this with the actual oil change records, the vehicle underwent an oil change on September 24, 2021, demonstrating that this method can effectively detect vehicle oil changes.

Claims

1. A diesel engine oil change identification method based on Internet of Vehicles big data, characterized in that: Includes feature extraction stage, model training and prediction stage; The feature extraction stage specifically includes the following steps: Step 1-1, data cleaning; Step 1-2, extracting the target working condition data set; Steps 1-3, feature extraction; The model training and prediction phase includes the following steps: Step 2-1: Build a supervised learning model; Step 2-2, model result prediction; Step 2-3, model error analysis; Steps 2-4: Fusion of multiple working condition results; Steps 2-5: Display of model results; Steps 1-3 extract key variables that reflect the oil status and aggregate the original 1 Hz data to reduce deviations caused by data fluctuations. The specific extraction steps are as follows: Step 1-3-1: Select the oil pressure measurement point as the key feature reflecting the oil viscosity. In addition, collect the engine speed, coolant temperature, fuel level, and intake air volume as measurement points reflecting the working conditions. Step 1-3-2: Aggregate the 1Hz data into 5s data according to time, calculate the average value, 10th percentile value, 90th percentile value, and median value of engine speed, coolant temperature, fuel quantity, and intake air quantity in four aggregation methods, and obtain 16 features to be input into the model; The average value corresponding to the oil pressure is selected as the label to be input into the model; Step 2-1 differentiates the data set according to the different speed ranges of the engine. Specifically, the data from the 7 days before the date to be analyzed is selected as the training set. The 16 variables calculated by four aggregation methods, namely the average value, 10th percentile, 90th percentile, and median of the engine speed, coolant temperature, fuel volume, and intake volume obtained in steps 1-3, are used as features. The average oil pressure at the corresponding time is used as the label, and the supervised learning model is input to obtain model a corresponding to the speed range of 790rpm to 810rpm and model b corresponding to the speed range of 890rpm to 910rpm. During the training process, the hyperparameter settings of model a and model b are consistent. The hyperparameters include learning rate, number of weak classifiers, downsampling ratio, loss function, maximum depth of leaf nodes, L1 regularization term, and L2 regularization term.

2. The diesel engine oil change identification method based on Internet of Vehicles big data according to claim 1 is characterized in that: The frequency of the original IoV data collected in step 1-1 is 1 Hz, and some unreasonable or missing values ​​exist in the data. This data needs to be processed according to the following rules: Rule 1: For some sample retention values, the row of data is directly removed; for temperature sampling values ​​≤ -270°C, the data is judged as invalid and also removed; Rule 2: For missing values, if it is a temperature measurement point with low accuracy, interpolate based on the previous and next temperature values; if it is not a temperature measurement point, directly remove the row of data.

3. The diesel engine oil change identification method based on Internet of Vehicles big data according to claim 1 is characterized in that: Steps 1-2 are applicable to the operating condition data range of the identification method, which is the warm-up process of a diesel engine after starting up every day. The warm-up process is the normal engine temperature rise process after the diesel engine is started. During this process, the engine is in an idling state and the speed level is adjusted according to the ambient temperature, usually fixed in one to two relatively narrow speed ranges. The temperature rise process is usually measured using the coolant temperature in the vehicle as an indicator.

4. The diesel engine oil change identification method based on Internet of Vehicles big data according to claim 3 is characterized in that: In steps 1-2, the conditions for extracting the target working condition are as follows: 1-2-1. The vehicle is in idle state, that is, the engine speed is greater than 0 rpm and the vehicle speed is 0 km / h; 1-2-2. The engine coolant temperature is higher than 45°C and lower than 55°C. Within this temperature range, the coolant temperature is approximately equal to the oil temperature in the oil circuit. After the data processed in steps 1-2-1 and 1-2-2, the engine speed should be concentrated in 1 to 2 speed ranges, which are 790 rpm to 810 rpm and 890 rpm to 910 rpm. The data in the subsequent two speed ranges must be included in the calculation range.

5. The diesel engine oil change identification method based on Internet of Vehicles big data according to claim 1 is characterized in that: In step 2-2, the data of the day to be analyzed is used, and the corresponding data is input into model a and model b respectively according to the different engine speed ranges to obtain the predicted 5-s average oil pressure value, which is recorded as predicted oil pressure a and predicted oil pressure b; the data of the day refers to the data of the 1Hz warm-up process stage collected on the day, including engine speed, coolant temperature, vehicle speed, intake flow, ambient temperature and oil pressure measurement points.

6. The diesel engine oil change identification method based on Internet of Vehicles big data according to claim 5 is characterized in that: In the model error analysis of step 2-3, the predicted oil pressure a and predicted oil pressure b output by model a and model b are respectively subtracted from the actual 5-second average oil pressure value in the corresponding engine speed range to obtain the prediction error of all data of the model on that day, and the average value of the corresponding errors of all data on the day to be analyzed is calculated and recorded as error a and error b.

7. The diesel engine oil change identification method based on Internet of Vehicles big data according to claim 6 is characterized in that: In steps 2-4, since there are two operating ranges for the engine speed, it is necessary to weight the average value of the error a and the error b according to the amount of data of the diesel vehicle in the two operating ranges, where the weight is the proportion of the amount of data in the corresponding range of engine speed to the total amount of data after screening; the weighted error is used as the final error for the day to be analyzed to determine whether the engine oil has been changed.

8. The diesel engine oil change identification method based on Internet of Vehicles big data according to claim 6 is characterized in that: In the steps 2-5, the error data for multiple consecutive days are calculated, and through the time series error fluctuations, it is determined whether the engine oil has been changed on the same day, and the result of the final model is output.

Images