Lubricant deterioration determination device, deterioration determination method

Abstract

To improve accuracy of determining deterioration of lubricant.SOLUTION: The present invention provides a lubricant deterioration determination device 100 for determining deterioration of lubricant in an internal combustion engine, the lubricant deterioration determination device having: an input data acquisition unit 5 that acquires data on an A / F, a water temperature, an exhaust temperature, an intake temperature, and a fuel injection amount of an internal combustion engine; an inference unit 6 that inputs the A / F, the water temperature, the exhaust temperature, the intake temperature, and the fuel injection amount into a model that associates the A / F, the water temperature, the exhaust temperature, the intake temperature, and the fuel injection amount with a total acid number to infer the total acid number; and an output unit 8 that outputs information related to the total acid number inferred by the inference unit 6.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a lubricant deterioration determination device and a lubricant deterioration determination method for an internal combustion engine. [Background technology]

[0002] It is known that deterioration of lubricating oil in an internal combustion engine reduces its lubrication effect between parts. However, the degree of lubricating oil deterioration varies depending on factors such as how the vehicle is used, so there is a need for technology that can accurately determine lubricating oil deterioration.

[0003] As an example of such a technology, Patent Document 1 discloses a technology that includes a total acid number estimation map showing the relationship between the total fuel flow rate in an engine and the total acid number of the engine oil, and a total base number estimation map showing the relationship between the time elapsed since an oil change and the total base number of the engine oil, and that determines the deterioration of engine oil based on at least one of a first determination result based on the total acid number estimation map and the total fuel flow rate, and a second determination result based on the total base number estimation map and the elapsed time. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2021-139347 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in conventional technology, the total fuel injection amount is used to determine lubricant deterioration, but there are thought to be other parameters that affect lubricant deterioration, and there was concern that the accuracy of determining lubricant deterioration was insufficient.

[0006] In view of the above-mentioned problems, an object of the present invention is to provide a technique for improving the accuracy of determining deterioration of lubricating oil. [Means for solving the problem]

[0007] In view of the above problems, the present invention provides a lubricant deterioration determination device for determining deterioration of lubricant in an internal combustion engine, and includes an input data acquisition unit that acquires data on A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount in the internal combustion engine, an inference unit that inputs the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount into a model that associates the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount with the total acid number, and infers the total acid number, and an output unit that outputs information related to the total acid number inferred by the inference unit. [Effects of the Invention]

[0008] The present invention can improve the accuracy of determining deterioration of lubricating oil. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a schematic diagram of a lubricant deterioration determination device that determines deterioration of engine lubricant; [Figure 2] FIG. 2 is a diagram illustrating an example of the arrangement of a lubricant deterioration determination device and a learning device. [Figure 3] FIG. 2 is an example of a functional block diagram of a learning device. [Figure 4] FIG. 2 is an example of a functional block diagram of a lubricant deterioration determination device. [Figure 5] FIG. 1 is a diagram showing an example of a neural network constituting a total acid number calculation model. [Figure 6] 1 is a flowchart illustrating an example of a process performed by a lubricant deterioration determination device to determine lubricant deterioration. DETAILED DESCRIPTION OF THE INVENTION

[0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A lubricant deterioration determination device and a lubricant deterioration determination method performed by the lubricant deterioration determination device will be described below as an example of an embodiment of the present invention with reference to the drawings.

[0011] <Method for determining deterioration of lubricating oil in this embodiment> In this embodiment, a total acid number calculation model is prepared that inputs A / F (air-fuel ratio), water temperature, exhaust temperature, intake temperature, and fuel injection amount, and outputs the total acid number. The lubricant deterioration determination device inputs the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount obtained from the vehicle into this total acid number calculation model, and outputs the total acid number.

[0012] It is known that the air-fuel ratio (A / F), water temperature, exhaust temperature, intake temperature, and fuel injection amount contribute to the production of formic acid, acetic acid, nitric acid, and hydrogen sulfide. Furthermore, formic acid, acetic acid, nitric acid, and hydrogen sulfide cause lubricating oil degradation. Therefore, a total acid number calculation model that inputs the air-fuel ratio (A / F), water temperature, exhaust temperature, intake temperature, and fuel injection amount and outputs the total acid number can accurately calculate the total acid number, which is an indicator of lubricating oil degradation.

[0013] <Outline of the internal combustion engine> An example of the configuration of a lubricant deterioration determination device 100 will be described with reference to Fig. 1. Fig. 1 is a schematic diagram of the lubricant deterioration determination device 100 that determines the deterioration of lubricant in an engine 11. The engine 11 of this embodiment may burn gasoline, diesel, hydrogen, natural gas, methanol, ethanol, LPG, dimethyl ether, biofuel, or the like, and the specific fuel that drives the engine 11 by combustion is not important. The engine 11 may also be used to generate power to run a vehicle.

[0014] The lubricant deterioration determination device 100 includes an ECU 12 (electronic control unit) and several sensors. The ECU 12 controls the engine 11. In this embodiment, the ECU 12 determines lubricant deterioration in addition to controlling the engine 11. Various sensors detect the values of various parameters used to determine lubricant deterioration. In FIG. 1, an A / F sensor 13, a water temperature sensor 14, an exhaust temperature sensor 15, and an intake air temperature sensor 16 are connected to the ECU 12. The A / F sensor 13 detects the A / F and transmits it to the ECU 12. The water temperature sensor 14 detects the temperature of the liquid cooling the engine 11 and transmits it to the ECU 12. The exhaust air temperature sensor 15 detects the exhaust air temperature from the engine 11 and transmits it to the ECU 12. The intake air temperature sensor 16 detects the intake air temperature to the engine 11 and transmits it to the ECU 12. The ECU 12 also executes fuel injection amount control 17, which injects fuel into the engine 11 in a controlled manner according to the accelerator opening, etc., and the fuel injection amount is a control value (i.e., a known value).

[0015] In this embodiment, the ECU 12 controls the engine 11 and also inputs the A / F, water temperature, exhaust temperature, and intake temperature into a prepared total acid number calculation model to output the total acid number. However, a device in the vehicle other than the ECU 12 may output the total acid number using the total acid number calculation model.

[0016] This section explains how A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount correlate with the total acid number, along with the factors that affect the total acid number: formic acid / acetic acid, nitric acid, and sulfuric acid. Formic acid / acetic acid are generated by unburned gases during fuel combustion. Also, unburned gases are more likely to be generated when the engine 11 is not warmed up. Therefore, the amount of unburned gas generated correlates with the A / F ratio. Also, whether the engine 11 is warmed up can be measured by the water temperature. The higher the temperature of the combustion gas, the more NO X It is known that NO X The water generated in the engine 11 is mostly due to condensation, but the lower the outside temperature, the more water is generated on the chain cover or head cover. XThe amount of condensation is correlated with the exhaust temperature, and the amount of condensation is correlated with the intake temperature. Hydrogen sulfide is H2S in fuel O4 Therefore, the amount of hydrogen sulfide correlates with the amount of fuel injected.

[0017] Thus, it is clear that the A / F ratio, water temperature, exhaust temperature, intake temperature, and fuel injection amount are appropriate indicators for estimating the total acid number of lubricating oil, as shown in Figure 1.

[0018] <Example of arrangement of lubricant deterioration determination device and learning device> FIG. 2 is a diagram illustrating an example of the arrangement of a lubricant deterioration determination device 100 and a learning device 200. The learning device 200 is an information processing device that learns the correspondence between inputs (in this embodiment, A / F (air-fuel ratio), water temperature, exhaust temperature, intake air temperature, and fuel injection amount) and outputs (in this embodiment, total acid number) to create a total acid number calculation model. Learning (also called machine learning) is a technique for enabling a computer to acquire human-like learning capabilities. It refers to a technique in which a computer autonomously generates algorithms required for judgments such as data identification from learning data acquired in advance and applies these to new data to make predictions. Learning methods for machine learning include supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning, but in this embodiment, supervised learning may be mainly used.

[0019] In FIG. 2(a), the lubricant deterioration determination device 100 is disposed in the vehicle similarly to the engine 11, but the learning device 200 is constructed in an information processing device connected to the vehicle (e.g., the lubricant deterioration determination device 100) via a network N. The information processing device may exist in the cloud or may exist in a building (on-premise) where the developer works. The learning device 200 may acquire the learning data from the vehicle via the network N. Alternatively, instead of the learning device 200 acquiring the learning data directly from the vehicle, for example, the developer may acquire the learning data from the vehicle and upload it to the learning device 200. The total acid number calculation model created by the learning device 200 may be set in the lubricant deterioration determination device 100 when the vehicle is shipped, or the lubricant deterioration determination device 100 may download it from the network.

[0020] In Figure 2(b), the lubricant deterioration determination device 100 and the learning device 200 are installed outside the vehicle. The learning device 200 is the same as in Figure 2(a). The lubricant deterioration determination device 100 inputs input data sent from the vehicle via the network N into a total acid number calculation model and outputs the total acid number.

[0021] In either arrangement example, the learning device 200 creates a normal amount prediction model from the learning data in the learning phase. In the inference phase, the lubricant deterioration determination device 100 inputs input data acquired from the engine 11 into the generated total acid number calculation model and outputs the total acid number.

[0022] 2, learning device 200 is located outside the vehicle, but the vehicle may also have learning device 200. In this case, the vehicle is assumed to have means for detecting or estimating the total acid number, which serves as training data.

[0023] <About the functions of the learning device> Next, the functions of the learning device 200 will be described with reference to Fig. 3. Fig. 3 is an example of a functional block diagram of the learning device 200 according to this embodiment. The learning device 200 has a learning data acquisition unit 2, a learning data storage unit 3, and a learning unit 4. These functional units of the learning device 200 are functions or means realized by the CPU of the information processing device executing instructions of a program expanded in RAM.

[0024] The learning data acquisition unit 2 acquires learning data 21. The learning data 21 is a plurality of data sets each consisting of input data and output data. The input data are A / F, water temperature, exhaust temperature, intake air temperature, and fuel injection amount, and the output data is total acid number. Here, the total acid number is not a value determined by the instantaneous value of the input data, but is considered to correlate with time-series data from the past. For this reason, the input data is time-series data for A / F, water temperature, exhaust temperature, intake air temperature, and fuel injection amount since the lubricating oil was changed to new oil. Alternatively, the learning device 200 may acquire an integrated value of the input data.

[0025] As an example, a vehicle periodically transmits the A / F ratio, water temperature, exhaust temperature, intake temperature, and fuel injection amount to the learning device 200 via the network N. The learning device 200 stores time-series data of these input data in association with the vehicle. To prevent the input data from becoming overwhelming, the A / F ratio, water temperature, exhaust temperature, and intake temperature may be average values over a fixed period of time. Furthermore, the fuel injection amount may be an integrated value over a fixed period of time. In a model such as a neural network, where a constant number of input data points is preferred, the amount of time-series data often differs depending on the vehicle or even within the same vehicle, as the elapsed time since the lubricant was changed. In other words, the number of time-series data points is likely to vary depending on the vehicle for which lubricant deterioration is being determined. In such a case, the learning device 200 aggregates the time-series data points since the lubricant was changed to a fixed number. For example, if the number of input data is N1 and the number of time-series data collected by the learning device 200 is N2 (>N1), the learning data acquisition unit 2 calculates or integrates (fuel injection amount) the average (A / F, water temperature, exhaust temperature, intake air temperature) for every k pieces of time-series data (=N2 / N1). Alternatively, the learning data acquisition unit 2 may acquire the maximum or minimum value for every k pieces of time-series data for the A / F, water temperature, exhaust temperature, and intake air temperature. This is because the maximum or minimum values for the A / F, water temperature, exhaust temperature, and intake air temperature are more strongly correlated with the total acid number. If N1 > N2 because the time elapsed since the lubricating oil change is short, the learning data acquisition unit 2 may add dummy data to the time-series data. The dummy data may be a constant value, such as zero.

[0026] Since the amount of total acid number is likely to be affected by time, the time elapsed since the lubricant was changed (time series data) may be one of the input data.

[0027] The total acid number, which is output data, is a value that represents the total amount of acidic components in the oil. The measurement method is defined by JIS or the like. Therefore, preferably, a developer or the like removes the lubricating oil from each vehicle and measures the total acid number for each vehicle using a predetermined measurement method. A pair of this total acid number and input data (time-series data) constitutes one piece of learning data 21. Furthermore, if the total acid number or a physical property correlated with the total acid number can be detected by a sensor installed in the vehicle, the learning data acquisition unit 2 may acquire the total acid number detected by the sensor.

[0028] The learning data storage unit 3 stores the learning data 21 acquired by the learning data acquisition unit 2. The learning unit 4 uses a part of the learning data 21 to learn (create a total acid number calculation model) and verifies the total acid number calculation model using the remaining data.

[0029] The learning unit 4 learns the learning data 21 using various machine learning algorithms to generate a total acid number calculation model. The total acid number calculation model is correspondence information that associates input data with the total acid number. In other words, the total acid number calculation model outputs the current total acid number for the input data acquired from the vehicle. A neural network is shown in FIG. 5 as an example of a learning algorithm.

[0030] <Function of the lubricant deterioration detection device> Next, the functions of the lubricant deterioration determination device 100 will be described with reference to Fig. 4. Fig. 4 is a functional block diagram of the lubricant deterioration determination device 100 of this embodiment. The lubricant deterioration determination device 100 has an input data acquisition unit 5, an inference unit 6, a determination unit 7, and an output unit 8. These functional units of the lubricant deterioration determination device 100 are functions or means realized by the microcomputer in the ECU 12 executing instructions of a program expanded in RAM.

[0031] The input data acquisition unit 5 acquires input data 23 (time series data of A / F, time series data of water temperature, time series data of exhaust temperature, time series data of intake temperature, and time series data of fuel injection amount). If the time series data of these input data is held by the ECU 12, it may be acquired from the ECU 12, or if it is stored on the network N, it may be acquired from the network N. Therefore, the input data is not limited to data acquired when the vehicle ignition is ON.

[0032] The inference unit 6 inputs the input data into a total acid number calculation model to calculate the total acid number of the lubricating oil. That is, the inference unit 6 inputs time series data of A / F, time series data of water temperature, time series data of exhaust temperature, time series data of intake air temperature, and time series data of fuel injection amount into the total acid number calculation model to infer the current total acid number.

[0033] The determination unit 7 compares the total acid number inferred by the inference unit 6 with a preset threshold value to determine whether the lubricant has deteriorated. The threshold value is determined as a value at which the total acid number is determined to not satisfy the appropriate performance of the lubricant. The threshold value may vary depending on the type of lubricant or the type of engine 11.

[0034] When the determination unit 7 determines that the lubricating oil has deteriorated, the output unit 8 outputs that fact. For example, the output unit 8 displays a warning on the meter panel or sends a warning by e-mail or the like addressed to the driver's e-mail address.

[0035] <Neural Network> Fig. 5 shows an example of a neural network constituting a total acid number calculation model. The neural network in Fig. 5 is a regression neural network (prediction of continuous values) that outputs one output value for multiple data input to input layer 41.

[0036] The neural network in FIG. 5 is a neural network with fully connected L layers (the number of weighted intermediate layers 42 and output layers 43 is assumed to be two) from the input layer 41 to the output layer 43. A neural network with a deep hierarchy is called a DNN (Deep Neural Network). The layer between the input layer 41 and the output layer 43 is called an intermediate layer 42 (or hidden layer). The number of intermediate layers 42 and the number of nodes 35a to 35c in each layer have been simplified for the purpose of explanation and are merely examples. The number of nodes 35a in the input layer 41 is the number of elements of the input data (in this embodiment, five elements: A / F, water temperature, exhaust temperature, intake air temperature, and fuel injection amount) multiplied by the number of time-series data. Therefore, x1, x2, etc. correspond to the time-series data of A / F, water temperature, exhaust temperature, intake air temperature, and fuel injection amount, respectively. When predicting continuous values, the number of nodes 35c in the output layer 43 is often one. The value y output by node 35c corresponds to the total acid number.

[0037] In the neural network of Figure 5, all nodes 35a in the input layer 41 are connected to one node 35b in the middle layer 42, and all nodes 35b in the middle layer 42 are connected to one node 35c in the output layer 43 (fully connected). The product of the output z of node 35a in the input layer 41 and the connection weight w is input to node 35b in the middle layer 42, and the product of the output z of node 35b in the middle layer 42 and the connection weight w is input to node 35c in the output layer 43. Equation (1) in Figure 5 shows how to calculate the signal input to node 35c.

[0038] In equation (1), w ji (l,l-1) is the weight between the jth node in the lth layer and the ith node in the l-1th layer, and b j is the bias component in the network. j (l) is the input to the jth node of the lth layer, and z i (l-1) is the output of the i-th node in the l-1th layer. I is the number of nodes in the l-1th layer.

[0039] Also, as shown in equation (2), the input u j(l) is activated by an activation function f, where f represents the node's activation function. Known activation functions include ReLU, tanh, and sigmoid. Node 35 in the lth layer nonlinearizes the input using the activation function and outputs it to node 35 in the l+1th layer. Note that node 35a in the input layer 41 is not activated; it simply transmits the input data to the second layer. In a neural network, this process is repeated from the input layer 41 to the output layer 43.

[0040] An activation function for the output layer is used for node 35c of output layer 43. The activation function for output layer 43 of a regression model is generally an identity function (y=x).

[0041] The learning phase will now be explained. The neural network processes data input to the input layer 41 and outputs an output value from the output layer 43. For example, for input data such as "0.1", "0.3", etc., node 35c of the output layer 43 outputs "5". Assume that training data includes pre-set training data of "7", which corresponds to "0.1" and "0.3". The training data is the total acid number prepared by actual measurement or the like.

[0042] In the learning phase, a loss function is used to evaluate the error between the training data (7) and the output value (5), and the weights w and b are adjusted so that the output value is closer to the training data. The loss function for a regression model can be a function that calculates the squared error. The value of the loss function is propagated to the nodes in the input layer 41 using a calculation method called backpropagation. The weights w and b between the nodes are learned during the propagation process.

[0043] While Figure 5 shows a neural network, if the input data is time-series data, it may be difficult to aggregate the number of time-series data to match the number of nodes 35a in the input layer 41, or convergence may be slow. Known models that use such time-series data as input include recurrent neural networks (RNNs), long short-term memories (LSTMs), and transformers. In an RNN, the calculation results in the intermediate layer 42 are multiplied by a weight and input again (to the intermediate layer 42) for use in the next calculation. Once all input data has been input, the output values are compared with the training data in the output layer, and the weights from the intermediate layer 42 to the intermediate layer 42 are repeatedly updated. LSTMs are models that solve the gradient vanishing problem that RNNs have. Transformers build networks using only a mechanism called attention, without using RNNs or CNNs.

[0044] Note that the inference of continuous values using input data can be realized not only by neural networks but also by regression models, such as multiple regression, ridge regression, lasso regression, and elastic net regression, and is not limited to the method described in this embodiment.

[0045] <Action or Processing> The flow of the process by which the lubricant deterioration determination device 100 determines lubricant deterioration will be described with reference to Fig. 6. Fig. 6 is a flowchart illustrating the process by which the lubricant deterioration determination device 100 determines lubricant deterioration. The process in Fig. 6 is executed, for example, when the ignition is turned off, at regular intervals while driving, or in response to a driver's instruction.

[0046] First, the input data acquisition unit 5 acquires time series data of A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount (S210). The time series data may be acquired from a storage means on a network or from the ECU 12.

[0047] Next, the input data acquisition unit 5 generates input data from these time series data (S220). For example, the input data acquisition unit 5 aggregates the time series data so that the number of input data becomes a number determined by the total acid number calculation model.

[0048] Next, the inference unit 6 inputs the input data into the total acid number calculation model (230). As a result, the inference unit 6 outputs the total acid number (S240). The judgment unit 7 compares the output total acid number with a threshold value to judge whether the lubricating oil has deteriorated.

[0049] If the determination unit 7 determines that the lubricant has deteriorated, the output unit 8 displays that the lubricant has deteriorated on the meter panel or notifies the driver by email. The output unit 8 may output the estimated total acid number. The output unit 8 may also output that the lubricant has deteriorated, based on a model created by machine learning. This allows the driver to know whether the lubricant has actually deteriorated as measured by a sensor or simply estimated, and to take appropriate action. The output unit 8 may also determine the deterioration of the lubricant in about three stages. The output unit 8 may also store previously estimated total acid numbers and dates and times, and output, based on this past history, when the total acid number will exceed a threshold (when it will be determined that the lubricant has deteriorated).

[0050] <Major Effects> The lubricant deterioration determination device 100 of this embodiment utilizes the fact that the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount are correlated with the amounts of formic acid, acetic acid, nitric acid, and hydrogen sulfide produced, and can accurately calculate the total acid number, which is an indicator of lubricant deterioration, using a total acid number calculation model that inputs the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount and outputs the total acid number.

[0051] In this embodiment, the method for determining deterioration of lubricating oil in a vehicle's internal combustion engine has been described, but the method can be applied to any moving body equipped with an internal combustion engine, such as a motorcycle, a ship, an airplane, a submarine, etc. Furthermore, the method can be applied to any moving body equipped with an internal combustion engine, such as a generator, etc. [Explanation of symbols]

[0052] 11 Engine 12 ECU 100 Lubricating oil deterioration determination device 200 Learning Device

Claims

1. A lubricant deterioration determination device for determining deterioration of lubricant in an internal combustion engine, an input data acquisition unit that acquires data on an A / F ratio, a water temperature, an exhaust temperature, an intake temperature, and a fuel injection amount in the internal combustion engine; an inference unit that infers the total acid number by inputting the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount into a model that associates the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount with the total acid number; an output unit that outputs information about the total acid number inferred by the inference unit; A lubricant deterioration determination device having the same.

2. 2. The lubricating oil deterioration determination device according to claim 1, wherein the model receives time series data of the A / F, time series data of water temperature, time series data of exhaust temperature, time series data of intake air temperature, and time series data of fuel injection amount as inputs, and outputs the total acid number.

3. The lubricant deterioration determination device according to claim 1 or 2, wherein the output unit outputs, as information about the total acid number, information indicating that the lubricant has deteriorated according to the model.

4. A deterioration determination method performed by a lubricant deterioration determination device that determines deterioration of lubricant in an internal combustion engine, comprising: A process of acquiring data on A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount in the internal combustion engine; a process of inputting the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount into a model in which the A / F, water temperature, exhaust temperature, intake temperature, and fuel injection amount correspond to the total acid number, and inferring the total acid number; outputting information about the inferred total acid number; A deterioration determination method.

Images