Prediction model establishment method for air inlet leakage detection, detection method and unit

Abstract

The invention provides a method for establishing a prediction model, the model is used for detecting air inlet leakage in an air inlet pipeline of an engine, and the method comprises the following steps: obtaining historical data of related parameters in the air inlet pipeline and engine operation parameters, and performing offline processing on the historical data based on a machine learning algorithm to establish the model; in the above process, through standardization processing and data cross multiplication, historical data generates a higher-dimension updated feature vector from an initial feature vector so as to establish a model. The invention also provides a method for detecting air inlet leakage by using the model, which comprises the following steps of: inputting real-time data of relevant parameters in an air inlet pipeline and engine operation parameters into the model obtained according to the method, and obtaining an output result representing whether the air inlet leakage occurs or not and / or the leakage level on the basis of a machine learning algorithm; and sending alarm information to an operator and / or actively limiting the rotating speed and the torque of the engine based on the output result indicating that the air inlet leakage occurs. The invention further relates to a detection unit for executing the method and a computer readable storage medium.

Description

Technical Field

[0001] The present invention relates to the technical field of diesel vehicles, and in particular to a method for establishing a prediction model, wherein the prediction model is used to detect intake air leakage in an engine intake line. The present invention also relates to a related method and unit for detecting intake air leakage in an engine intake line. Background Art

[0002] In the vehicle engines commonly used in the current market, especially diesel engines, the intake system often leaks due to long-term use of the pipeline aging damage or insufficient sealing of the pipeline connection parts. Intake leakage can cause the diesel engine to fail due to insufficient intake and insufficient power. In addition, if there is damage on the intake pipeline and unclean air enters the cylinder, it will cause serious wear of the diesel engine cylinder liner, piston, piston ring, etc., resulting in black smoke from the diesel engine exhaust, insufficient power, burning oil and other failures. In practice, the working environment of off-road vehicles using diesel engines is quite complex. For example, when using a combine harvester to harvest grain in a farmland, it is necessary to continue working under high vibration, high temperature, high noise and continuous working conditions. Since the window period of the busy farming season is usually only one month or even shorter to half a month, once a leak occurs in the intake pipeline of the diesel engine, it will seriously affect the operating efficiency.

[0003] Current technology for detecting intake air leaks in the intake line relies primarily on physical sensors. Figure 1 The figure shows the engine intake pipe system of a diesel vehicle and the sensor layout structure for detecting intake leakage in the prior art. As shown in the figure, corresponding pressure sensors and temperature sensors are respectively set at various points of the intake pipe (especially where intake leakage is prone to occur), and the upper limit and lower limit under normal operation are set for each sensor. Excluding the situation where the sensor itself fails, when the detection value of the sensor exceeds the upper limit or lower limit of normal operation, the intake leakage is already very serious, thus causing irreparable damage to the diesel engine.

[0004] It should be noted that the "background technology" section is only used to help understand the content of the present invention, so the content disclosed in the "background technology" section may contain some prior art that is not known to those skilled in the art. The content disclosed in the "background technology" section does not mean that the content or the problems to be solved by one or more embodiments of the present invention have been known or recognized by those skilled in the art before the application of the present invention. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects in the above-mentioned prior art and provide an improved method for detecting intake leakage in the engine intake pipe, which includes the establishment of a prediction model, the use of the prediction model in combination with actual operating parameters for detection, and taking corresponding remedial measures based on the detection results. The method according to the present invention can discover abnormal data sets in the leakage process through learning big data, and provide warnings and reminders. When warnings and reminders are issued repeatedly, the vehicle ECU is used to actively limit the engine speed and torque to avoid additional damage to the engine.

[0006] To achieve the above-mentioned purpose, a method for establishing a prediction model is provided, wherein the prediction model is used to detect intake leakage in an engine intake duct, and the method comprises: obtaining historical data of relevant parameters in the engine intake duct and engine operating parameters, and performing offline processing on the historical data based on a machine learning algorithm to establish the prediction model; wherein during the offline processing, the historical data is generated from an initial feature vector into an updated feature vector with a higher dimension through standardization and data cross multiplication for use in establishing the prediction model.

[0007] According to another aspect of the present invention, a method for detecting intake leakage in an engine intake duct is provided, comprising the following steps: step S10: establishing the prediction model using the method according to any one of claims 1 to 5; step S20: inputting real-time data of relevant parameters in the engine intake duct and engine operating parameters into the prediction model, and obtaining an output result indicating whether intake leakage occurs and / or the leakage level based on the machine learning algorithm; step S30: issuing an alarm message to an operator and / or actively limiting the engine speed and torque based on the output result obtained by step S20 indicating the occurrence of intake leakage; wherein in the process of executing step S20, the real-time data is generated from the initial feature vector into an updated feature vector with a higher dimension through normalization processing and data cross multiplication to obtain the output result.

[0008] According to another aspect of the present invention, a unit for detecting intake leakage in an engine intake duct is provided, which is configured to perform the aforementioned detection method, and the unit comprises: a data acquisition module, which is configured to be able to acquire historical data and real-time data of relevant parameters in the engine intake duct and engine operating parameters; a data processing module, which is configured to perform offline processing on the historical data based on a machine learning algorithm to establish a prediction model for detecting intake leakage, and output a result indicating whether an intake leakage occurs and / or the leakage level based on the machine learning algorithm when the real-time data is input into the prediction model; and an alarm module, which is configured to issue an alarm message to an operator and / or actively limit the engine speed and torque based on the detection of the intake leakage.

[0009] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores a program, and when the program is executed by a processor, the processor is enabled to perform the aforementioned detection method. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The above and other aspects of the present invention will be more thoroughly understood and appreciated with reference to the accompanying drawings, in which:

[0011] Figure 1 The invention shows the engine air intake pipeline system of a diesel vehicle and the sensor layout structure for detecting the air intake leakage thereof according to the prior art;

[0012] Figure 2 A basic flow chart of a method for detecting an intake air leak in an intake line of an engine according to the present invention is shown;

[0013] Figure 3 A flow chart of establishing a prediction model using a machine learning algorithm according to a preferred embodiment of the present invention is shown, wherein the prediction model is used to process real-time data to detect whether an intake air leak occurs;

[0014] Figure 4 Shows the use of Figure 3 The prediction model obtained by the flow chart obtains an output result indicating whether intake air leakage occurs and / or the leakage level based on real-time input data;

[0015] Figure 5 In a two-dimensional coordinate system, normal data points indicating no intake air leakage and abnormal data points indicating intake air leakage are schematically shown;

[0016] Figure 6 A flow chart is shown for detecting intake leaks in an engine intake line using real-time data after obtaining a prediction model and for subsequently executing countermeasures.

[0017] Figure 7 A schematic block diagram of a unit for detecting intake air leakage in an intake line of an engine according to a preferred embodiment of the present invention is shown. DETAILED DESCRIPTION

[0018] In order to make the technical problems, technical solutions and beneficial technical effects to be solved by the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and exemplary embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the scope of protection of the present invention.

[0019] like Figure 2 As shown, the method for detecting intake air leakage in an intake pipe of an engine according to an embodiment of the present invention mainly comprises the following steps S10 to S30:

[0020] First, at S10, relevant parameters in the engine intake pipe and historical data of engine operating parameters are obtained, and the historical data are processed offline based on a machine learning algorithm, especially a clustering algorithm, to establish a prediction model for detecting intake leakage. In this article, the historical data not only includes Figure 1 The pressure, temperature and flow parameters obtained by the pressure sensors, temperature sensors and air flow meters installed at various points in the existing intake pipe are shown, and also include engine operating parameters such as engine speed, fuel injection amount and constant parameters representing different engine operating modes. Figure 1 It can be seen that the pressure and temperature parameters serving as historical data of the engine intake pipeline include but are not limited to the intake pressure P0 and intake temperature T0 at the inlet of the intake pipeline, that is, between the air source and the air filter (AF); the pressure P1 and temperature T1 of the portion of the intake pipeline between the air filter and the air compressor; the pressure P2 and temperature T2 of the portion of the intake pipeline between the air compressor and the charge air cooler (CAC); the pressure P21 and temperature T21 of the portion of the intake pipeline between the charge air cooler and the boost pressure sensor (BPS); the pressure P22 and temperature T22 at the location of the boost pressure sensor; the inlet temperature T4 of the oxidation catalyst (DOC) downstream of the turbocharger (WGT) of the air compressor; the inlet temperature T5 of the particulate filter (DPF) downstream of the oxidation catalyst and the pressure difference ΔP across the particulate filter; the temperature T6 of the portion of the intake pipeline between the particulate filter and the selective catalytic reduction (SCR), etc. It should be noted here that although Figure 1The portion ① between the air filter and the air compressor, the portion ② between the air compressor and the charge air cooler, and the portion ③ between the charge air cooler and the boost pressure sensor in the intake pipe shown belong to intake leakage points ALP commonly seen in practice. However, in terms of obtaining historical data, the present invention is not limited to obtaining the pressure, temperature and / or flow at the above-mentioned points, but relevant data obtained at any position in the intake pipe can be selected as input data to establish a prediction model.

[0021] Next, we will combine Figure 3 The process of establishing the prediction model is described. Figure 3 As shown, it shows the various sub-steps S101 to S104 belonging to step S10, which are specifically described as follows:

[0022] At S101, the above historical data at each time point within a certain time period is obtained. It should be noted that the time period (e.g., from 0 to t) needs to at least completely cover the entire cycle of the engine in a certain operating mode (including normal mode, regeneration mode, and desulfurization mode) from starting, gradually accelerating, accelerating to the maximum operating limit, gradually decelerating, and finally shutting down; and at each certain interval Δt, the pressure, temperature, flow rate, engine speed, injection amount, and parameters representing the engine operating mode at the corresponding point in the intake pipe as described above are obtained as a set of initial feature vectors at the corresponding time point. In other words, at step S101, a total of t / Δt groups of initial feature vectors will be obtained as initial input data.

[0023] Next, at S102, in order to eliminate the interference caused by the large magnitude difference between different physical quantities on data processing, the above initial input data will be standardized. Taking the intake pressure P0 in the intake pipe as an example, as mentioned above, a total of t / Δt P0 values ​​will be obtained during the entire time period (0 to t), from which the maximum value P0 of P0 in this time period can be obtained. max and minimum value P0 min , then for any time point t i Inlet pressure P0 i For example, the standardized pressure value P0′ at this time point is obtained using the following formula i :

[0024]

[0025] The normalization of other physical quantities such as temperature, flow, engine speed, fuel injection amount, etc. is similar to that of pressure, so this article will not go into details. It can be seen that each normalized physical quantity represents a dimensionless number in the interval [0,1]. All normalized physical quantities obtained at any time point constitute a new set of dimensionless feature vectors at that time point.

[0026] Next, at step S103, the standardized data in the dimensionless feature vector obtained from step S102 is cross-multiplied, and the result of the cross-multiplication together with the dimensionless feature vector obtained from step S102 constitutes a set of updated feature vectors with increased dimensions at the corresponding time point. For example, assuming that time t 1 The normalized dimensionless numbers of the intake pressure, intake temperature, and intake flow rate obtained at the same time are 0.1, 0.2, and 0.3, respectively. Then the result of cross-producting the above normalized data will be [0.02, 0.03, 0.06]. Combining this result with the above initial dimensionless feature vector [0.1, 0.2, 0.3] will obtain 1 The updated feature vector with increased dimension is [0.1, 0.2, 0.3, 0.02, 0.03, 0.06]. In other words, after the above processing, the dimension of the feature vector used for model prediction is increased to six dimensions from the three dimensions (i.e., pressure, temperature, and flow) of real data initially obtained by physical sensors, further combined with the virtual data of three new dimensions (i.e., dimensionless pressure multiplied by dimensionless temperature, dimensionless pressure multiplied by dimensionless flow, and dimensionless temperature multiplied by dimensionless flow) introduced by data cross multiplication. As described above, the present invention can obtain relevant parameters such as pressure, temperature and / or flow from sensors arranged at various points in the intake pipe, and can also obtain parameters related to engine operation such as speed, injection amount, operating mode, etc., so the dimension of the real data obtained can be much larger than the three in the above example (assuming a total of N) as needed. It can be understood by those skilled in the art that after being processed by the above step S103, the updated feature vector with increased dimension will have a total of That is, N+N*(N-1) / 2 dimensions. In other words, the updated feature vector obtained at a certain point in time has N+N*(N-1) / 2 components.

[0027] It can be understood that after the above steps S101 to S103, a total of t / Δt groups of feature vectors with N+N*(N-1) / 2 dimensions representing the historical data will be obtained. Next, at S104, a certain number of groups (for example, m groups, m is less than or equal to t / Δt) of updated feature vectors are randomly selected from the multiple groups of updated feature vectors corresponding to each time point obtained in step S103 as training data, and the machine learning algorithm is used to train them and establish the prediction model. In a preferred embodiment of the present invention, the above m groups of feature vectors are processed and modeled by taking the following machine learning algorithm representing the clustering idea as an example.

[0028] Specifically, firstly, m groups of feature vectors are randomly selected to form a data set [X 1 ,X2 ,X 3 ,……X m ], where X i Represents the value obtained in step S103 at t i The feature vector after the dimension increase at the time point. Then according to the data set [X 1 ,X 2 ,X 3 ,……X m ], calculate the minimum Euclidean space distance from each data point to the other m-1 data points [D 1 ,D 2 ,D 3 ,……D m For the first data point X 1 For , first calculate its Euclidean space distance from the remaining m-1 data points according to the following formula:

[0029]

[0030] Where D 1j refers to the first data point X 1 The set of Euclidean distances to other data points. 1k It refers to the first data point X 1 The component of the kth dimension in this eigenvector; similarly X jk It means that the jth data point X j The component of the kth dimension in this feature vector, where j can be selected from 2, 3, 4, ..., m. After that, the obtained m-1 distances are compared and the minimum value is taken to obtain D 1 , that is to say, D 1 =min[D 12 ,D 13 ,……,D 1m ]. In a similar manner, the second data point X can be obtained in sequence 2 The minimum Euclidean space distance D from the remaining m-1 data points 2 And so on until the mth data point X m The minimum Euclidean space distance D from the remaining m-1 data points m For the i-th data point (i.e., the i-th group of updated feature vectors) X i For , the minimum Euclidean space distance between it and the remaining m-1 data points is calculated using the following formula:

[0031] Where j = 1, 2, 3 ..., m and j ≠ i. After obtaining the minimum Euclidean space distance [D 1 ,D 2 ,D3 ,……D m ] After that, the average of the above minimum Euclidean space distance can be obtained by the following formula to obtain the mean R of the closest distance.

[0032]

[0033] After that, the distance D from each data point to its nearest point is calculated. i The points with the most data are classified into the first category, followed by the second category, the third category, and so on, and the points isolated at the end are noise points (i.e., abnormal points indicating air leakage in the air intake pipeline). In the embodiment of the present invention, if the comparison result shows that a certain D i Exceeding R, this means that for i For the data point represented by , the multidimensional Euclidean space with it as the center and R as the space radius will not contain any other data points, that is, the data point X i Compared with other data points, this data point X is isolated according to the above classification criteria. i Indicates the abnormal point where the intake pipe has an intake leak, and when the leak has occurred, as time accumulates, D i The value exceeding R gradually increases to indicate the gradual increase of the leakage level. That is to say, in the process of establishing the prediction model, the mean value R of the minimum Euclidean space distance of each data point to other data points is finally obtained as the decision boundary value for judging whether the intake pipeline is normal (no intake leakage) or abnormal (intake leakage occurs).

[0034] After the prediction model is established as described above, according to the method of the present invention, Figure 2At step S20 shown, the relevant parameters in the intake pipe when the engine is in the running mode and the real-time data of the engine running parameters are input into the prediction model, and the machine learning algorithm used in step S10 is used to calculate the minimum Euclidean space distance between the updated feature vector at the corresponding time point in a period and the updated feature vector at other earlier time points, and the result is output based on the comparison of the minimum Euclidean space distance and the previously obtained decision boundary value R, and the output result indicates whether the intake leakage occurs and / or the size of the leakage level at each time point. It should be noted here that the updated feature vector at the other earlier time point mentioned above can include all the updated feature vectors from the start of the engine to the current detection time point, and can also include a part of the updated feature vectors. For example, as the engine continues to run, using the real-time data obtained at all time points to detect whether the intake leakage occurs means a significant increase in the amount of calculation. In this case, this part of the updated feature vectors obtained in a short period of time before the current detection time point can be input into the prediction model to detect whether the intake leakage occurs. Alternatively or additionally, the partially updated feature vector at an earlier time point may also represent the updated feature vector corresponding to the normal data point after excluding the abnormal data point detected before the current detection time point. In this case, it is possible to avoid the current detected data point (which is actually also an abnormal data point) being erroneously detected as a normal data point because the minimum Euclidean space distance between the abnormal data point closer in time is less than the decision boundary value, thereby improving the detection accuracy of the method.

[0035] Figure 4 Shows the use of Figure 3The prediction model EM obtained by the flow chart obtains output data OD indicating whether intake leakage occurs and / or the leakage level based on the relevant parameters in the engine intake pipeline and the real-time data of the engine operating parameters. It can be seen from the figure that in the preferred embodiment of the present invention, the parameters representing the engine operating mode ERM are used as input data to participate in the training of the previous prediction model and subsequently as real-time data input to participate in the detection of intake leakage. In this case, unlike other physical quantities such as intake pressure P0, intake temperature T0, flow rate, engine speed r, injection amount FIQ, etc., the parameters representing the engine operating mode are assigned constant constant parameters. For example, constants 0, 1, and 2 are used to represent the normal mode, regeneration mode, and desulfurization mode of the engine respectively. As an alternative example, other constant parameters such as -2, -1, and 0 can also be used to represent the normal mode, regeneration mode, and desulfurization mode of the engine. It should be noted here that although 0 is used to represent one of the corresponding operating modes of the engine in the initial case, in order to avoid the influence of the product of this parameter and other components of the standardized eigenvector when cross-multiplying the data, a constant parameter very close to 0 (such as 0.01, 0.001, etc.) is usually selected during the calculation.

[0036] Further reading Figure 5 , which shows normal data points indicating no intake leakage (hollow dots in the figure) and abnormal data points indicating intake leakage (solid dots in the figure) in a two-dimensional coordinate system. It should be noted that the horizontal axis in the figure represents time t, and the vertical axis represents the working transient set WTS (working transient set, i.e., the aforementioned multi-dimensional feature vector set) obtained based on real-time data input. As shown in the figure, different areas are divided by "straight lines" to represent abnormal area I, normal area II running in engine regeneration and desulfurization mode, and normal area III running in engine normal mode. It should be noted that the actual decision boundary DB is not strictly a straight line, but a multi-dimensional surface represented by a constraint function. Figure 5 What is shown is only the decision boundary of a straight line and the projection of the data points representing the multi-dimensional feature vector set in the two-dimensional coordinate system. When the constants 0, 1, and 2 are used to represent the normal mode, regeneration mode, and desulfurization mode of the engine for modeling and prediction, the minimum Euclidean space distance between the updated feature vector at a certain time point and the updated feature vector at other earlier time points is greater than the decision boundary value (i.e. R) determined based on the prediction model to determine that an abnormality has occurred (i.e., an intake leak has occurred in the intake pipe, such as Figure 5 Area I).

[0037] Further, the concept of the present invention is not limited to the above preferred embodiments. For example, instead of using a constant constant to represent the engine's operating mode as input data to participate in the training of the prediction model and the detection of intake leakage, different prediction models (for example, a normal mode prediction model, a regeneration mode prediction model, and a desulfurization mode prediction model) are established for different engine operating modes, and the intake leakage status and level are evaluated based on different prediction models. In this case, the way of obtaining the input data, the way of processing, the training of the prediction model, and the detection of intake leakage are exactly the same as those described above, and will not be repeated here.

[0038] Next, if Figure 2 As shown, after the prediction model obtained in step S10 outputs the result of determining whether a leakage occurs in the intake pipeline based on real-time input data at step S20, the method of the present invention issues an alarm message to the operator and / or actively limits the engine speed and torque at step S30 based on the output result obtained in step S20 indicating that an intake leakage occurs. Figure 6 A flowchart is shown for detecting intake leakage in the engine intake line using real-time data after obtaining a prediction model and for subsequent implementation of countermeasures. Figure 6 As can be seen from the previous description of step S20, this step itself includes the following three sub-steps, namely: obtaining real-time data of relevant parameters in the engine intake duct and engine operating parameters at S201, and inputting the real-time data into the prediction model previously obtained through offline training; processing the original real-time data into an updated feature vector with a higher dimension through standardization and data cross multiplication in the prediction model at S202; and calculating the minimum Euclidean space distance between the updated feature vector at a corresponding time point within a period and the updated feature vector at other earlier time points in the prediction model at S203, and outputting the result based on the comparison of the minimum Euclidean space distance with the decision boundary value obtained through the prediction model. Further, as Figure 6 As shown, step S30 may include the following sub-steps: At S301, determine whether the output result obtained by step S20 is normal (i.e., whether intake leakage occurs). If the result is yes, that is, when it is determined that there is no intake leakage in the intake pipe, continue the current operation of the engine and return to step S20 to continuously obtain the above real-time data input prediction model to determine whether the intake pipe leaks at a subsequent time point. If the result is no, that is, when it is determined that there is an intake leakage in the intake pipe, the method proceeds to step S302, at which time the above signal will be fed back to the processor to control a flashing light, a buzzer and / or other alarm devices to send an alarm signal to the operator, thereby reminding the operator that an intake leakage occurs.

[0039] Furthermore, in order to implement graded alarms, an abnormality counter can be used to count the number of abnormalities detected by the prediction model, and then, according to the current count value of the abnormality counter, it is selected whether to issue a more striking alarm message to the operator or to force the engine speed and torque to be actively limited when the operator is distracted, so as to remind the operator to stop the engine in time and perform maintenance on the engine. Figure 6As shown, in the process of determining that no intake leakage occurs in the intake pipeline so that the method of the present invention returns from step S301 to step S20, a judgment step S305 can be additionally set, in which it is judged whether the current count value ErrCnt of the abnormal condition counter is greater than zero. If the result is yes, it means that a result indicating an abnormality (i.e., an intake leakage occurs) was output at a previous time point, but the output result at the current time point is normal. Generally speaking, unless the abnormal result is continuously reported within a certain time window, the occasional abnormal result does not necessarily indicate that an intake leakage has occurred in the intake pipeline. In this case, the method flow proceeds to step S306 to reduce the current count value ErrCnt of the abnormal condition counter by 1 and continue to return to step S20. On the contrary, if the judgment result at step S305 is no, it directly returns to step S20 to detect whether an intake leakage occurs at a subsequent time point. In this way, system false alarms caused by occasional abnormal results can be avoided. On the other hand, if it is determined at step S301 that an intake air leak is detected in the intake line, the method flow proceeds to step S302, at which a flashing light, a buzzer and / or other alarm devices send an alarm signal to the operator at a given frequency and increase the current count value ErrCnt of the abnormal condition counter by 1. Thereafter, the method proceeds from step S302 to step S303 to determine whether an intake air leak is detected within a time window (i.e., a given calibrated continuous driving cycle) based on whether the current count value ErrCnt of the abnormal condition counter exceeds the warning threshold. If the result is no, the method returns to step S20 to detect whether an intake air leak occurs at a subsequent time point; if the result is yes, the method proceeds from step S303 to step S304 to actively limit the engine speed and torque using the vehicle ECU, and at the same time remind the operator in a more eye-catching manner (e.g., flashing with higher frequency and higher brightness, buzzing with higher frequency and louder volume, etc.) that a relatively serious intake air leak has occurred in the intake line. It should be noted that the selection of the above-mentioned warning threshold is proportional to the sampling frequency of the operating parameters. In other words, the higher the sampling frequency f of each sensor set in the intake duct (which means the time interval between each time point for obtaining real-time data is smaller), the larger the warning threshold A can be set to indicate that the abnormal condition in the intake duct is continuously detected within the time window A / f. Correspondingly, the lower the sampling frequency f of each sensor set in the intake duct (which means the time interval between each time point for obtaining real-time data is larger), the smaller the warning threshold A needs to be set to ensure that the abnormal condition in the intake duct is continuously detected within a relatively short period of time, so as to take countermeasures to protect the engine in the first time.

[0040] The method flow of the present invention is explained and illustrated above in combination with the preferred embodiment. In the traditional method of using physical sensors to detect relevant physical quantities, it is usually necessary to detect that the direct physical quantity exceeds the corresponding threshold value through the relevant sensor to determine that the intake leakage occurs only when the intake leakage is already quite serious. With the help of the method of the present invention, in particular, the feature vector set as the input data not only includes the real data obtained by the sensor as in the traditional method, but also includes the virtual data with greatly increased dimensions obtained by processing the above data. Any small data fluctuation caused by the intake leakage will cause a relatively significant output result. Under this premise, the present invention can discover the abnormal data set during the intake leakage process as early as possible through the learning of big data, and provide alarms and reminders to the operator at the first time. When a specific number of alarms occur within a continuous time, the engine speed and torque can be actively limited to avoid additional damage to the engine. In addition, the implementation of the method of the present invention does not require the addition of additional new sensors in the current intake pipeline or the addition of different types of sensors used by the vehicle. In other words, it is not necessary to rebuild the components in the current detection system, thereby saving the cost of modifying the detection system.

[0041] It should be noted that, although the establishment of the prediction model and the detection of intake leakage are described in this article by taking the minimum distance mean algorithm based on the clustering idea as an example. However, those skilled in the art may also think of using other clustering algorithms such as K-means, DBSCAN, etc. In addition, machine learning algorithms such as support vector machines, linear regression, etc. other than clustering algorithms may also be used to achieve the establishment of the prediction model and the detection of intake leakage. It should be noted that no matter what machine learning algorithm is used, the core idea is to process the currently obtained real data to obtain the "increased dimension" feature vector as the input data for first training and establishing the prediction model and subsequently for intake leakage evaluation without reconstructing the traditional intake leakage detection system (that is, without adding new physical sensors or adding new sensor types to obtain additional physical parameters). Taking the support vector machine algorithm as an example, after obtaining the "increased dimension" feature vector as input data, the decision function of the prediction model corresponds to a decision hyperplane with a total dimension minus one, and the new real-time data point can be judged based on the decision hyperplane whether it is abnormal.

[0042] Corresponding to the method for detecting intake air leakage in the intake pipe in the above embodiment, the present invention also proposes a corresponding intake air leakage detection unit 10. Figure 7As shown, the intake leakage detection unit 10 includes a data acquisition module 101, which is configured to acquire historical data and real-time data of relevant parameters in the engine intake pipeline and engine operating parameters; a data processing module 102, which is configured to perform offline processing on the historical data based on a machine learning algorithm to establish a prediction model for detecting intake leakage, and output a result indicating whether an intake leakage occurs and / or the leakage level based on the machine learning algorithm when the real-time data is input into the prediction model; and an alarm module 103, which is configured to issue an alarm message to the operator and / or actively limit the engine speed and torque based on the detection of the intake leakage.

[0043] The unit for detecting intake air leakage in the engine intake pipe provided in the above embodiments can execute the intake air leakage detection method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method. The various steps and technical details of the detection method described above can be stored in the detection unit in the form of software, or implemented in the form of a combination of software and hardware. For technical details not described in detail in the above embodiments, please refer to the intake air leakage detection method provided in the preferred embodiment of the present invention.

[0044] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0045] The above describes in detail with the aid of the accompanying drawings a feasible but non-limiting implementation of the method and unit for detecting intake air leakage in an intake line according to the present invention. For those skilled in the art, without departing from the scope and essence of the present disclosure as set forth in the following claims, modifications and supplements to the technology and structure and the re-combination of the features in each embodiment should be deemed to be included in the scope of the present invention. Therefore, these modifications and supplements that can be conceived under the teachings of the present invention should be regarded as part of the present disclosure. The scope of the present disclosure is defined by the claims attached below, and includes equivalent technologies known at the filing date of the present disclosure and equivalent technologies that have not yet been foreseen.

Claims

1. A method for establishing a prediction model, wherein the prediction model is used to detect intake air leakage in an intake line of an engine, the method comprising: Acquire relevant parameters in the engine intake pipeline and historical data of engine operating parameters, and perform offline processing on the historical data based on a machine learning algorithm to establish the prediction model; During the offline processing, the historical data is generated from an initial feature vector into an updated feature vector with a higher dimension through standardization and data cross multiplication for establishing the prediction model.

2. The method according to claim 1, wherein the relevant parameters in the engine intake line include one or more of the following: intake pressure, intake temperature, intake flow rate at the inlet of the intake line; pressure, temperature, flow rate at the point where leakage is likely to occur in the intake line; temperature at each component of the selective catalytic reduction system located downstream of the intake line; and The engine operating parameters include one or more of an engine speed, a fuel injection amount, and an engine operating mode.

3. The method according to claim 1 or 2, comprising the following sub-steps: Step S101: acquiring the historical data at each time point within a certain time period, wherein the time period at least completely covers the periodic cycle of the engine in an operating mode; Step S102: normalizing the historical data to obtain a dimensionless feature vector representing the historical data; Step S103: performing data cross multiplication on the dimensionless feature vector obtained in step S102 and forming a set of updated feature vectors at the corresponding time point together with the dimensionless feature vector; Step S104: randomly selecting a certain number of updated feature vectors from the multiple groups of updated feature vectors corresponding to each time point obtained in step S103 as training data, and training the training data based on the machine learning algorithm to establish the prediction model.

4. The method according to claim 3, wherein the machine learning algorithm is a clustering-based minimum feature distance mean algorithm, and the decision boundary value R of the prediction model is calculated by the formula Calculate, where m represents the number of groups of randomly selected updated feature vectors, D i It represents the minimum Euclidean space distance between the i-th group of updated feature vectors in the selected updated feature vectors and the remaining m-1 groups of updated feature vectors, which is calculated by the following formula: Where j = 1, 2, 3, ..., m and j ≠ i, X ik represents the feature vector component in the kth dimension in the i-th group of updated feature vectors, and N represents the total number of relevant parameters in the engine intake pipeline and the engine operating parameters.

5. The method according to claim 2, wherein when the engine operating mode is used as input data for establishing a prediction model, different constants are assigned to the normal mode, regeneration mode and desulfurization mode of the engine respectively; when the engine operating mode is not used as input data, the method establishes prediction models for different engine operating modes based on other input data to detect intake leakage.

6. A method for detecting an intake air leak in an engine intake line, comprising the following steps: step S10: Establishing the prediction model using the method according to any one of claims 1 to 5; Step S20: inputting relevant parameters in the engine intake pipeline and real-time data of engine operating parameters into the prediction model, and obtaining an output result indicating whether intake leakage occurs and / or the leakage level based on the machine learning algorithm; Step S30: issuing an alarm message to the operator and / or actively limiting the engine speed and torque based on the output result obtained in step S20 indicating that an intake air leakage occurs; In the process of executing step S20, the real-time data is generated from the initial feature vector into an updated feature vector with a higher dimension through standardization processing and data cross multiplication to obtain the output result.

7. The method of claim 6, wherein generating the real-time data from an initial feature vector into an updated feature vector having a higher dimension comprises: Obtaining real-time data corresponding to each time point within a certain period of time during engine operation; Performing standardization processing on the real-time data to obtain a dimensionless feature vector representing the real-time data at a corresponding time point; as well as The dimensionless feature vector representing the real-time data at the corresponding time point is cross-multiplied and together with the dimensionless feature vector representing the real-time data at the corresponding time point, a set of updated feature vectors representing the real-time data at the corresponding time point is formed.

8. The method according to claim 7, wherein in step S20, the real-time data of the relevant parameters in the engine intake pipe and the engine operating parameters are used as input data, and the minimum Euclidean space distance between a set of updated feature vectors representing the real-time data at the corresponding time point and the updated feature vectors at other earlier time points is calculated based on the machine learning algorithm. When the minimum Euclidean space distance is greater than the decision boundary value determined by the prediction model, an abnormal result indicating that an intake leakage has occurred is output, otherwise a normal result indicating that no intake leakage has occurred is output.

9. The method according to any one of claims 6 to 8, wherein step S30 further comprises the following sub-steps: Step S301: Determine whether the result output in step S20 is normal, if the result is normal, return to step S20, if the result is abnormal, proceed to step S302; Step S302: Send an alarm signal to the operator at a given frequency, increase the current count value of the abnormal condition counter by 1, and proceed to step S303; Step S303: determining whether the current count value reaches a warning threshold, the value of which is related to the time window range for detecting continuous intake air leakage and the sampling frequency of the input data. If the result is no, the process returns to step S20 for subsequent determination. If the result is yes, the process proceeds to step S304. Step S304: Actively limit the engine speed and torque using the vehicle ECU, and send an alarm signal to the operator at a frequency higher than the given frequency.

10. The method according to claim 9, wherein in the process of returning from step S301 to step S20, it further comprises: Step S305, determine whether the current count value of the abnormal status counter is greater than zero. If the result is yes, proceed to step S306, reduce the current count value by 1 and return to step S20; if the result is no, directly return to step S20.

11. A unit for detecting intake air leakage in an intake line of an engine, which is configured to perform the method according to any one of claims 6 to 10, the unit comprising: A data acquisition module configured to acquire relevant parameters in the engine intake pipeline and historical data and real-time data of engine operating parameters; a data processing module configured to perform offline processing on the historical data based on a machine learning algorithm to establish a prediction model for detecting intake air leakage, and output a result indicating whether intake air leakage occurs and / or a leakage level based on the machine learning algorithm when the real-time data is input into the prediction model; as well as A warning module is configured to issue a warning message to an operator and / or actively limit the speed and torque of the engine based on the detection of the intake air leak.

12. A computer-readable storage medium storing a program, which, when executed by a processor, enables the processor to perform the method according to any one of claims 6 to 10.

Images