Method and device for diagnosing a vehicle transmission

Abstract

The invention relates to a method and an apparatus for diagnosing a vehicle transmission. The method of diagnosing a vehicle transmission based on artificial intelligence may include: collecting vehicle information and shift data of a vehicle; preprocessing and marking the gear shifting data of the transmission; generating, in the feature set stored in the database, a candidate feature set list including a plurality of candidate feature sets satisfying a specific criterion; using artificial intelligence based on supervised learning, for each of the plurality of candidate feature sets, determining an evaluation index of transmission shift comfort associated with the transmission shift data; a feature set of the vehicle is determined by selecting a candidate feature set having the highest evaluation index among candidate feature sets having evaluation indexes above a threshold.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to and the benefit of Korean Patent Application No. 10-2023-0099818 filed in the Korean Intellectual Property Office on July 31, 2023, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present invention relates to a method and device for diagnosing a transmission based on artificial intelligence, and more particularly, to a method and device for diagnosing a transmission based on artificial intelligence that can be optimized by reducing the amount of data required to learn and reflect customer complaint data. Background Art

[0004] The shift feel of an automatic transmission refers to the driver's feeling of the acceleration fluctuations, engine speed fluctuations, and noise generated by the engagement / disengagement of the automatic transmission's shift clutch. The driver mainly feels the shift feel from the acceleration fluctuations, and determines the failure of the transmission based on the shift feel (in other words, the driver's emotions). For example, based on the driver's sensitivity, if an impactful shift occurs or the shift is slow, it is recognized that a transmission failure has occurred. Each manufacturer ensures the proper shift feel of the automatic transmission through shift feel adjustments by relevant researchers and various human evaluation and improvement processes. However, due to deviations in hardware (H / W) manufacturing, some mass-produced vehicles may have a shift feel that is lower than the shift feel designed during development, and the shift feel at the development level may also be felt as slow or involving impact based on the driver's emotional standards.

[0005] When a driver's complaint about the automatic shift feel occurs, there is no way to distinguish whether the problem is due to emotional criteria alone or due to a hardware problem, except for the quantitative evaluation method, and there may be a situation where the hardware is replaced without any actual hardware problem. In order to prevent these problems, there is such a logic: it is able to measure the data of a 6-speed or 8-speed automatic multi-speed transmission through a portable diagnostic device (which is called a global diagnosis system (GDS)), determine whether the shift feel is problematic (fault (not good, NG)) or not (normal (OK)), and inform whether to replace parts or replace the entire transmission.

[0006] The supervised learning-based AI technology used to determine the shift feel requires sufficient quantity and quality of learning data, as well as correct labeling. Generally speaking, in order to create a feature set (which is a collection of feature values), the developers of the shift feel obtain data by testing, preprocessing and labeling them, and then use them to train the AI ​​model.

[0007] This method has limitations, among which, first, since the level of shift feel required for each vehicle and vehicle category is different, it is difficult to share a feature set for a specific vehicle. Second, since the powertrain (PT) combinations (engine displacement and transmission type) of vehicles and vehicle categories are diverse, it takes a lot of time to obtain the data required to create an artificial intelligence database DB that considers each vehicle type and PT combination. Third, in order to maintain a certain standard in the emotional field of data-based judgment, only the person in charge of developing the shift feel can participate in labeling, which is time-consuming. Fourth, if the standard of the person in charge of developing the shift feel is different from that of the customer, it is difficult in actual use.

[0008] The above information disclosed in this Background section is only for enhancement of understanding of the background of the invention and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0009] The present invention attempts to provide a method and apparatus for diagnosing a transmission based on artificial intelligence, which can reduce the total amount of data and marking time by determining whether the feature set of the vehicle can be shared, thereby generating an optimized database.

[0010] The present invention attempts to provide a method and apparatus for diagnosing a transmission based on artificial intelligence, which is capable of referring to customer complaint data collected through repair shops, confirming and modifying misdiagnosis of data determined to be faulty, and updating a feature set.

[0011] According to one or more exemplary embodiments of the present invention, a method may include: collecting vehicle information and transmission shift data of a vehicle; preprocessing and marking the transmission shift data; generating a candidate feature set list including multiple candidate feature sets that meet specific criteria from a feature set stored in a database; utilizing artificial intelligence based on supervised learning to determine, for each of the multiple candidate feature sets, an evaluation index of transmission shift comfort associated with the transmission shift data; and determining a feature set of the vehicle by selecting a candidate feature set with the highest evaluation index from among the candidate feature sets with evaluation indexes above a threshold.

[0012] The method may further include generating a new feature set based on additional transmission shift data associated with the vehicle in accordance with a plurality of candidate feature sets not having an evaluation index above a threshold.

[0013] Generating the candidate feature set list may include determining a plurality of feature sets regarding a plurality of other vehicles as the plurality of candidate feature sets, each of the plurality of feature sets regarding the plurality of other vehicles including vehicle information that is the same as at least a portion of the vehicle information of the vehicle.

[0014] The portion of the vehicle information of the vehicle may include at least one of a transmission model, whether an electric motor is used, an engine fuel, an engine displacement, or an engine type.

[0015] The evaluation index may be based on a confusion matrix-based performance evaluation index, which indicates the performance of supervised learning-based artificial intelligence using accuracy and false positive rate.

[0016] Determining the feature set of the vehicle may include selecting a candidate feature set based on a candidate feature set having a highest accuracy rate among candidate feature sets having an accuracy rate above a second threshold and a false positive rate below a third threshold.

[0017] The method may further include: determining the accuracy of the performance evaluation index based on the confusion matrix based on the formula (TP+TN) / (TP+TN+FN+FP), and calculating the false positive rate of the performance evaluation index based on the confusion matrix based on the formula FP / (TP+TN+FN+FP). For the transmission shift comfort related to artificial intelligence based on supervised learning: TP may represent a true positive, which indicates that the positive prediction is correct, TN may represent a true negative, which indicates that the negative prediction is correct, FP may represent a false positive, which indicates that the positive prediction is wrong, and FN may represent a false negative, which indicates that the negative prediction is wrong.

[0018] Determining the feature set of the vehicle may include: obtaining data about the vehicle; preprocessing and labeling the obtained data; training artificial intelligence based on supervised learning based on the preprocessed and labeled data; and storing the feature set generated by the training in a database. Training artificial intelligence based on supervised learning may include utilizing a semi-supervised learning method.

[0019] Semi-supervised learning methods may include the MixMatch algorithm.

[0020] According to one or more exemplary embodiments of the present invention, a method for diagnosing a transmission based on artificial intelligence may include: measuring first transmission shift data of a vehicle; preprocessing the measured first transmission shift data; determining the transmission shift comfort using artificial intelligence based on supervised learning and based on a first feature set related to the vehicle; receiving information about whether a customer is dissatisfied with the transmission shift comfort; based on determining that the information about dissatisfaction indicates that the customer has a complaint related to the determined transmission shift comfort, receiving customer complaint data including the preprocessed first transmission shift data and the determined transmission shift comfort; marking misdiagnosis data confirmed by a user based on a cumulative count of the received customer complaint data being above a first threshold; generating second transmission shift data by adding the misdiagnosis data to the first feature set; based on the second transmission shift data, training artificial intelligence based on supervised learning and generating a second feature set related to the vehicle; and storing the second feature set based on an evaluation index of the trained artificial intelligence based on supervised learning being above a second threshold.

[0021] Generating the second transmission shift data may include: determining the label ratio of the second transmission shift data; training artificial intelligence based on supervised learning based on the label ratio satisfying a third threshold; and terminating the training of artificial intelligence based on supervised learning based on the label ratio not satisfying the third threshold.

[0022] The evaluation index may be based on a confusion matrix-based performance evaluation index, which indicates the performance of supervised learning-based artificial intelligence using accuracy and false positive rate.

[0023] Storing the second feature set may include: based on the accuracy rate being above a fourth threshold and the false positive rate being below a fifth threshold, setting the second feature set trained based on supervised learning artificial intelligence as a representative feature set of the vehicle.

[0024] According to one or more exemplary embodiments of the present invention, a device for diagnosing a transmission based on artificial intelligence may include: one or more processors; a memory storing instructions. When the instructions are executed by one or more processors, the device: collects vehicle information and transmission shift data related to the vehicle and pre-processes them; marks the transmission shift data; generates a candidate feature set list including multiple candidate feature sets that meet specific criteria in the feature set stored in the database; determines the evaluation index of the transmission shift comfort related to the transmission shift data for each of the multiple candidate feature sets using artificial intelligence based on supervised learning; determines the feature set of the vehicle by selecting the candidate feature set with the highest evaluation index among the candidate feature sets with evaluation indexes above a threshold.

[0025] When the instructions are executed by one or more processors, the apparatus may further be caused to: generate a new feature set based on additional transmission shift data associated with the vehicle, based on multiple candidate feature sets that do not have an evaluation index above a threshold.

[0026] When the instructions are executed by one or more processors, the apparatus may be caused to: generate a new feature set by utilizing a semi-supervised learning method including a MixMatch algorithm.

[0027] When the instructions are executed by one or more processors, the device can be caused to: generate a candidate feature set list by determining multiple feature sets about multiple other vehicles as multiple candidate feature sets, each of the multiple feature sets about the multiple other vehicles including vehicle information that is the same as at least a portion of the vehicle information of the vehicle.

[0028] The evaluation index may be based on a performance evaluation index based on a confusion matrix, wherein the performance evaluation index based on the confusion matrix indicates the performance of artificial intelligence based on supervised learning using accuracy and false positive rate. When the instructions are executed by one or more processors, the device may be caused to: determine the feature set of the vehicle by selecting a candidate feature set based on a candidate feature set with the highest accuracy rate from among candidate feature sets with an accuracy rate above a second threshold and a false positive rate below a third threshold.

[0029] When the instructions are executed by one or more processors, the device can further enable the device to generate second transmission shift data based on customer complaint data related to transmission shift comfort satisfying a first threshold: by marking misdiagnosis data confirmed by the user, and adding the misdiagnosis data to existing data to generate an updated feature set; based on evaluation indicators related to the updated feature set satisfying a second threshold, determine the updated feature set as the final feature set of the vehicle, and store the updated feature set in a database.

[0030] When the instructions are executed by one or more processors, the device can further: utilize a performance evaluation indicator based on a confusion matrix, wherein the performance evaluation indicator based on the confusion matrix is ​​configured to evaluate the performance of artificial intelligence based on supervised learning according to accuracy and false positive rate; and determine the updated feature set as the final feature set of the vehicle based on the accuracy of the performance evaluation indicator based on the confusion matrix being above a third threshold and the false positive rate of the performance evaluation indicator based on the confusion matrix being below a fourth threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a block diagram of an apparatus for diagnosing a transmission based on artificial intelligence according to an implementation scheme.

[0032] Figure 2is a flow chart of a method for diagnosing a transmission based on artificial intelligence according to an implementation scheme.

[0033] Figure 3 It is shown Figure 2 Flow chart of a method for generating a list of candidate feature sets.

[0034] Figure 4 is a flow chart of a method for diagnosing a transmission based on artificial intelligence according to an implementation scheme.

[0035] Figure 5 is a table diagram showing a confusion matrix for evaluation indicators according to an embodiment.

[0036] Figure 6 is a schematic diagram illustrating a process of diagnosing a vehicle by using a method of diagnosing a transmission based on artificial intelligence according to an embodiment.

[0037] Figure 7 is a schematic diagram illustrating a process of diagnosing a vehicle by using a method of diagnosing a transmission based on artificial intelligence according to an embodiment. DETAILED DESCRIPTION

[0038] Embodiments of the present invention will be described more fully below with reference to the accompanying drawings so that those skilled in the art can easily implement the embodiments. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present invention. In order to illustrate the present invention, parts not related to the specification will be omitted, and throughout the specification, the same reference numerals will be used to refer to the same elements or equivalents.

[0039] In addition, unless explicitly described to the contrary, the word "include" and variations such as "comprising" or "including" are understood to imply the inclusion of the elements described, but not the exclusion of any other elements. Terms including ordinal numbers such as first and second are used to describe various constituent elements, but these constituent elements are not limited by the terms. These terms are only used to distinguish one component from other components.

[0040] In addition, the terms "unit", "component" or "portion", "device" and "module" in the specification refer to a unit that processes at least one function or operation, which can be implemented by hardware, software, or a combination of hardware and software.

[0041] Hereinafter, one or more exemplary embodiments of the present invention will be described with reference to the accompanying drawings.

[0042] Figure 1 is a block diagram of an apparatus for diagnosing a transmission based on artificial intelligence according to an implementation scheme. Figure 1Each component shown in the drawings may be implemented using hardware (eg, one or more processors and a memory storing instructions that can be executed by the one or more processors), software, or a combination of both.

[0043] The method for diagnosing a transmission (also called a gear transmission, gearbox, etc.) based on artificial intelligence is performed by a device 100 for diagnosing a transmission based on artificial intelligence. The device 100 for diagnosing a transmission based on artificial intelligence can be set as a server. Alternatively, the device 100 for diagnosing a transmission based on artificial intelligence can be set up through multiple servers and / or devices. In this case, multiple servers and / or devices can be connected via a network. The device 100 for diagnosing a transmission based on artificial intelligence can be connected to a database DB and an artificial intelligence model AIM via a network. The device 100 for diagnosing a transmission based on artificial intelligence may include a database DB and an artificial intelligence model AIM.

[0044] In an embodiment, the apparatus 100 for diagnosing a transmission based on artificial intelligence can be set as a server, and can be connected to a vehicle diagnostic device set in the form of a user terminal through a network. The apparatus 100 for diagnosing a transmission based on artificial intelligence can optimize the database DB required for the artificial intelligence model AIM based on supervised learning stored in the vehicle's diagnostic device (e.g., a tablet computer). The apparatus 100 for diagnosing a transmission based on artificial intelligence can generate or update a feature set for each vehicle included in the database DB.

[0045] Here, a feature set may refer to elements of input data used by an artificial intelligence model when learning or making decisions. Features represent variables that explain individual observations within a data set, and a collection of multiple features may be a feature set. In artificial intelligence, a feature set may be a weight value learned by receiving data input, a feature set may be defined as a collection of data features or items, and a feature set may be referred to as a feature. For example, in one embodiment, a feature set may be various pieces of information required to determine a shift feel, including engine speed (RPM), transmission state, vehicle speed, and the like. Shift feel may refer to the comfort felt by an occupant of a vehicle when performing a transmission shift. Shift feel may be a measure of subjective comfort reported by an occupant.

[0046] The feature set may include various features depending on the vehicle. The feature set may be different depending on the type of vehicle. For example, feature set A for vehicle A having a first vehicle type may be different from feature set B for vehicle B having a second vehicle type different from the first vehicle type. However, in an embodiment, if the vehicle information of vehicle A is partially the same as the vehicle information of vehicle B, the already generated feature set A may be used for vehicle B without generating feature set B.

[0047] Figure 1The components of the apparatus 100 for diagnosing a transmission based on artificial intelligence are shown in the form of a block diagram. Figures 2 to 7 The generation and optimization of a feature set by updating an artificial intelligence database DB is performed via the apparatus 100 for diagnosing a transmission based on artificial intelligence in detail.

[0048] refer to Figure 1 The device 100 for diagnosing a transmission based on artificial intelligence includes: a preprocessor 110, a marking unit 120, a candidate feature set list generator 130, an evaluation index calculator 140, a feature set determiner 150 and a customer complaint data processor 160.

[0049] The preprocessor 110 can collect and preprocess vehicle information and the shift data of the vehicle. The vehicle information can be used to generate a decision of a candidate feature set list of the user's vehicle in a database DB storing feature sets about multiple vehicles. The preprocessor 110 can obtain vehicle information through the controller area network (CAN) communication of the vehicle, or receive information directly input by a developer (e.g., a manufacturer, an administrator, a user, etc.). For example, the vehicle information may include: power source type (ICE, mHEV, HEV, PHEV, EV), transmission model (6DCT, 7DCT, 4AT, CVT, etc.), engine fuel type (gasoline, diesel, LPG, FFV), engine displacement, engine operation method (MPI, GDI, turbocharger, supercharger, Atkinson cycle, etc.), vehicle drive method (2WD / 4WD), suspension and shock absorber type, etc.

[0050] The preprocessor 110 may collect the shift data required for diagnosing the user's vehicle. The shift data may include information about the vehicle's transmission. For example, the shift data may include engine speed (RPM), gear state, vehicle speed, transmission temperature, etc.

[0051] The preprocessor 110 may preprocess the data into a basic process using an artificial intelligence model. For example, since in the case of a shift clutch control, the control period varies depending on the shift gear, clutch type, and shift control time point, the preprocessor 110 may perform time normalization to match the data to a predetermined time value to use an artificial intelligence model and identify the impact generation time point.

[0052] The labeling unit 120 may label the preprocessed data. Labeling may be a process of assigning a specific label or category to data. This may be a necessary process for a model to learn data and make predictions (mainly in machine learning). For example, the labeling unit 120 may assign a label or category of fault (NG) or normal (OK) to each preprocessed data.

[0053] The candidate feature set list generator 130 may generate a candidate feature set list including a plurality of candidate feature sets that meet a specific criterion among the feature sets stored in the database DB. The specific criterion may be a criterion predetermined by a developer.

[0054] For example, the candidate feature set list generator 130 may determine a plurality of feature sets about a plurality of other vehicles (each feature set including at least a portion of the same vehicle information as the vehicle information of the vehicle) as candidate feature sets for the corresponding vehicle, and include the determined candidate feature sets in the candidate feature set list. Here, the corresponding vehicle may correspond to the vehicle of the user for whom the feature set is to be generated.

[0055] The candidate feature set list generator 130 may evaluate the similarity of data on various combinations of vehicle type and powertrain to determine the availability of another vehicle's feature set for shift feel diagnosis. The candidate feature set list generator 130 may generate the candidate feature set list with respect to the following specific criteria.

[0056] The candidate feature set list generator 130 may determine feature sets of vehicles showing similar torque generation forms for each accelerator pedal position sensor (APS) as candidate feature sets. For example, vehicles having similar torque forms generated according to motor use (ICE, mHEV, HEV, PHEV, etc.), engine fuel type (gasoline, diesel, LPG, FFV, etc.), engine displacement, engine type (MPI, GDI, turbocharger, supercharger, Atkinson, etc.) have a higher possibility of data sharing.

[0057] The candidate feature set list generator 130 may determine the feature sets of vehicles having the same transmission type as candidate feature sets. For example, the acceleration waveform generated when shifting gears is different due to the transmission ratio and the like according to the transmission type (DCT, AT, CVT, etc.), and when the transmission types of the vehicles are the same, the possibility of data sharing may be high.

[0058] The influence of vehicle type such as suspension type, shock absorber type, etc. is not significant. However, since the expectation level of customers may differ according to vehicle type, the candidate feature set list generator 130 may adjust the determination level of the candidate feature set by reflecting customer complaint data.

[0059] The evaluation index calculator 140 may diagnose the shift feeling with respect to the shift data, and calculate the evaluation index by using artificial intelligence AIM based on supervised learning based on a plurality of candidate feature sets.

[0060] The evaluation index calculator 140 determines the shift feel by using the candidate feature set list through artificial intelligence based on supervised learning, and thereby evaluates the evaluation index. In this process, the artificial intelligence based on supervised learning AIM receives the candidate feature set of the candidate feature set list as an input value, performs shift feel determination based on the vehicle's shift data, and can store the result to calculate the evaluation index.

[0061] Performance evaluation indicators based on confusion matrices can be used as evaluation indicators. Performance evaluation indicators based on confusion matrices can evaluate the performance of artificial intelligence based on accuracy and error detection rate (i.e., false positive (FP) rate). Accuracy can indicate how accurately artificial intelligence predicts actual phenomena. FP rate can indicate how wrong artificial intelligence predicts actual phenomena.

[0062] The feature set determiner 150 may select a candidate feature set with the highest level from among the candidate feature sets having the evaluation index above the calculated specific level, determine it as the feature set of the user's vehicle, and store the determined feature set of the vehicle in the database. As the accuracy is higher and the FP rate is lower, the evaluation index may have a higher level. The specific level may be preset using the accuracy and the FP rate.

[0063] For example, the feature set determiner 150 may select a candidate feature set with the highest accuracy from among the candidate feature sets whose accuracy is above a preset level and whose FP rate is below a preset level as the feature set of the vehicle. Here, the reason for checking only the accuracy is that misdiagnosis is highly likely to occur depending on the level of the impact, and the FP rate is prone to fluctuate when reflecting the vehicle data of a customer with a complaint. Therefore, the accuracy against external interference and the ability to intuitively confirm the performance can be checked. The preset level may be a level predetermined by a developer.

[0064] When there is no candidate feature set having an accuracy above a preset level and an FP rate below a preset level, the feature set determiner 150 may generate a new feature set after collecting and marking additional shift data about the vehicle, and determine it as the feature set of the vehicle. For example, the feature set determiner 150 may generate a new feature set by utilizing a semi-supervised learning method including a MixMatch algorithm.

[0065] The customer complaint data processor 160 diagnoses the shift feel of the vehicle using artificial intelligence based on supervised learning based on a predetermined feature set of the vehicle.

[0066] When the result of the customer complaint data is received to a certain count or more, the customer complaint data processor 160 confirms whether there is a misdiagnosis through the developer. For example, the customer complaint data processor 160 may send the customer complaint data to the developer, and may receive misdiagnosis data confirming the misdiagnosis. The customer complaint data processor 160 re-marks the confirmed misdiagnosis data, and adds the marked data to the existing data to generate new second shift data. The certain count may be a predetermined value.

[0067] The customer complaint data processor 160 can relearn the artificial intelligence based on supervised learning based on the generated second shift data, generate a new updated feature set, and when the evaluation index calculated for the updated feature set is above a certain level, determine the updated feature set as the final feature set of the vehicle. The updated feature set is generated by reflecting the customer complaint problem to the traditional feature set.

[0068] The customer complaint data processor 160 uses a performance evaluation index based on a confusion matrix (which evaluates the performance of artificial intelligence according to accuracy and FP rate), and if the accuracy is above a preset level and the FP rate is below a preset level, the customer complaint data processor 160 can determine the updated feature set as the final feature set of the vehicle. The customer complaint data processor 160 can update the final feature set of the vehicle in the database DB.

[0069] Figure 2 is a flow chart of a method for diagnosing a transmission based on artificial intelligence according to an implementation scheme. Figure 3 It is shown Figure 2 Flow chart of a method for generating a list of candidate feature sets.

[0070] exist Figure 2 and Figure 3 In the embodiment, the method for diagnosing a transmission based on artificial intelligence can be performed by the device 100 for diagnosing a transmission based on artificial intelligence, the database DB and the artificial intelligence model AIM. The method for diagnosing a transmission based on artificial intelligence can be a method for generating a feature set required for diagnosing a transmission of a vehicle by using feature sets of other vehicles already existing in the database and by using artificial intelligence. Figure 2 Describe it.

[0071] In step S210, in the method for diagnosing a transmission based on artificial intelligence, the apparatus 100 for diagnosing a transmission based on artificial intelligence according to the embodiment can obtain vehicle information and shift data of the vehicle and pre-process them. For example, the apparatus 100 for diagnosing a transmission based on artificial intelligence can obtain vehicle information and shift data about the nth vehicle (vehicle #n) and pre-process them to generate a feature set.

[0072] In step S220 , the apparatus 100 for diagnosing a transmission based on artificial intelligence may perform labeling on the preprocessed data and calculate a label ratio.

[0073] In step S230, the apparatus 100 for diagnosing a transmission based on artificial intelligence may determine whether the label ratio of the data is uniform. In an embodiment, the apparatus 100 for diagnosing a transmission based on artificial intelligence may check the label ratio of the data to ensure the reliability of the evaluation index. For example, when there are more normal (OK) labels in the data than NG1, NG2, NG3, and NG4 labels, the possibility of overfitting to OK increases. Accordingly, the prediction accuracy of the data for the OK label may be higher, but the prediction accuracy of the NG label may be lower.

[0074] Therefore, in step S240 , when the label ratio becomes beyond a preset level, the apparatus 100 for diagnosing a transmission based on artificial intelligence may perform preprocessing and labeling work to maintain a uniform ratio by adding data to insufficient labels or by removing data from excessive labels.

[0075] In step S100, when the label ratio is determined to be uniform, the device 100 for diagnosing a transmission based on artificial intelligence can extract candidate feature sets from the database DB and generate a candidate feature set list to determine the availability (sharing) of feature sets (or feature values) of the database DB. In an embodiment, the candidate feature set list may include multiple candidate feature sets that meet specific criteria among the feature sets of the database DB. For example, the candidate feature set may include a feature set about a vehicle that includes vehicle information similar to the vehicle information of the nth vehicle. Figure 3 Step S100 is described in detail.

[0076] exist Figure 3 In the embodiment, the apparatus 100 for diagnosing a transmission based on artificial intelligence can determine a plurality of feature sets about a plurality of other vehicles as candidate feature sets, each candidate feature set including vehicle information that is the same as at least a portion of the vehicle information of the vehicle. Here, at least a portion of the vehicle information can include: transmission model, motor usage, engine fuel, engine displacement, and engine type.

[0077] In the implementation scheme, in step S110 , the apparatus 100 for diagnosing a transmission based on artificial intelligence may call vehicle information for generating data of an i-th feature set, that is, may call vehicle information of an i-th vehicle having an i-th feature set.

[0078] In step S120 , the apparatus 100 for diagnosing a transmission based on artificial intelligence may compare the vehicle information of the i-th vehicle with the vehicle information of the n-th vehicle of the user to generate a feature set.

[0079] In step S130, the device 100 for diagnosing a transmission based on artificial intelligence can check whether the i-th vehicle and the n-th vehicle use the same transmission model, use an electric motor, use the same engine fuel, have the same engine displacement, and have the same engine type.

[0080] In step S140, when the i-th vehicle and the n-th vehicle use the same transmission model, use an electric motor, use the same engine fuel, have the same engine displacement, and have the same engine type, the i-th feature set of the i-th vehicle may be stored in the candidate feature set list of the n-th vehicle. That is, when the listed vehicle information of the i-th vehicle and the n-th vehicle are the same, it can be considered that the specific criteria of the candidate feature set are met.

[0081] In step S150, after storing the i-th feature set in the feature set list of the n-th vehicle, it can be checked whether the i+1-th feature set exists in the database. If it exists, in step S160, the vehicle information of the i+1-th vehicle corresponding to the i+1-th feature set (i+1 feature set) is called and compared with the vehicle information of the n-th vehicle, and when a specific standard is met, the i+1 feature set is stored in the candidate feature set list. The device 100 for diagnosing a transmission based on artificial intelligence can repeat the process to generate a candidate feature set list for the n-th vehicle.

[0082] If any of the listed vehicle information of the i-th vehicle and the n-th vehicle is not the same, then at step S150 , the checking of the vehicle information of the i-th vehicle may be stopped, and then it may be immediately checked whether the (i+1)-th feature set exists.

[0083] Return to reference Figure 2 In step S250, the apparatus 100 for diagnosing a transmission based on artificial intelligence can diagnose the shift feeling of the shift data of the nth vehicle, and calculate an evaluation index based on multiple candidate feature sets through artificial intelligence based on supervised learning. The evaluation index can use a performance evaluation index based on a confusion matrix that evaluates the performance of artificial intelligence according to accuracy and FP rate.

[0084] The device 100 for diagnosing a transmission based on artificial intelligence can extract candidate feature sets whose calculated evaluation indicators are above a specific level and store them in a list. In step S260, the device 100 for diagnosing a transmission based on artificial intelligence can extract candidate feature sets whose accuracy is above a preset level and whose FP rate is below a preset level from the candidate feature sets, and store them in a list. For example, the accuracy can be calculated as (TP+TN) / (TP+TN+FN+FP). The FP rate can be calculated as FP / (TP+TN+FN+FP). Here, regarding the prediction of the shift feel by artificial intelligence, the true positive (true positive, TP) represents the situation where the prediction is correct (for example, the number of situations), the true negative (true negative, TN) represents the situation where the prediction is not correct (for example, the number of situations), the false positive (false positive, FP) represents the situation where the prediction is wrong (for example, the number of situations), and the false negative (false negative, FN) represents the situation where the prediction is wrong (for example, the number of situations). This will refer to Figure 5 Provide a detailed explanation.

[0085] In step S270 , the apparatus 100 for diagnosing a transmission based on artificial intelligence may check through a list whether a candidate feature set having an evaluation index above a specific level has been extracted.

[0086] If there is no candidate feature set stored in the list, then in step S281, it can be determined that there is no feature set that can be shared for the nth vehicle. In step S291, in this case, the device 100 for diagnosing a transmission based on artificial intelligence can additionally obtain and preprocess the shift data about the nth vehicle, generate a new feature set by using a semi-supervised learning method, and store it in a database. That is, a new feature set is generated as a feature set corresponding to the nth vehicle. For example, the device 100 for diagnosing a transmission based on artificial intelligence can independently obtain data about the nth vehicle, preprocess and label the obtained data, and generate a new feature set by using the preprocessed and labeled data, via a semi-supervised learning method, and by artificial intelligence learning. In an embodiment, the semi-supervised learning method may include a MixMatch algorithm. Semi-supervised learning is a field of machine learning, and is a method of using labeled sample data and unlabeled sample data together to enable a model to learn. Semi-supervised learning provides a method for improving the performance of a model by utilizing unlabeled data. The MixMatch algorithm is one of the methods used in semi-supervised learning. The MixMatch algorithm is a method for enabling a model to learn by combining unlabeled data and labeled data.

[0087] In step S282, if there is a candidate feature set stored in the list, it is checked whether there are multiple candidate features. In step S300, when there are multiple candidate features, the corresponding candidate feature set can be finally determined as the feature set for the nth vehicle and stored in the database DB.

[0088] In step S292, if there are multiple candidate feature sets stored in the list, the feature set with the highest accuracy may be selected. In step S300, the selected one feature set may be determined as the feature set shared by the nth vehicle and stored in the database.

[0089] Figure 4 is a flow chart of a method for diagnosing a transmission based on artificial intelligence according to another embodiment. Figure 4 An implementation scheme of the method for diagnosing a transmission based on artificial intelligence can be such a method: when the diagnostic results of the transmission of a specific vehicle (whose feature set is already stored in a database) include customer complaints, the feature set of the corresponding vehicle is updated to modify the diagnostic results by reflecting the customer complaints.

[0090] exist Figure 4 In step S410, the shift data of the vehicle of the complaining customer can be measured through the diagnostic device (tablet computer) communicating with the CAN of the vehicle, and preprocessing can be performed.

[0091] In step S420, the shift feel of the vehicle may be diagnosed by using the supervised learning-based artificial intelligence model AIM stored in the diagnostic device. For example, the shift feel of the vehicle may be diagnosed based on a first feature set of the vehicle of the complaining customer.

[0092] In step S430, it can be determined whether the complaining customer is dissatisfied with the diagnosis result. For example, regarding whether the complaining customer is dissatisfied with the diagnosis result, information or data on whether the customer is dissatisfied can be received from the repair shop of the customer's vehicle, etc. When the customer is dissatisfied with the diagnosis result, in step S440, the device 100 (see FIG. 1 ) for diagnosing a transmission based on artificial intelligence can be used to diagnose the transmission based on artificial intelligence. Figure 1 ) can receive customer complaint data including pre-processed shift data and diagnostic results of the vehicle. Like the customer dissatisfaction information, the customer complaint data can be received from a repair shop of the customer's vehicle, etc.

[0093] In step S450, the apparatus 100 for diagnosing a transmission based on artificial intelligence may determine whether the cumulative count of received customer complaint data is above a specific count. In step S460, when the cumulative count of received customer complaint data is above a specific count, the misdiagnosis may be confirmed by the developer, and the marking may be re-executed for the confirmed misdiagnosis data.

[0094] In step S470, the device 100 for diagnosing a transmission based on artificial intelligence can check whether the cumulative count of misdiagnosis data is above a specific standard. If the level of customer dissatisfaction is very small, the customer's acceptance level of shift shock is likely to be higher than the customer level assumed during development. In this case, the decision criteria of the artificial intelligence may be problematic, so the existing feature set can be used as is until the data is accumulated on the server. When the misdiagnosis data is above a specific count, this means that there is a difference between the actual customer level and the developed level in the fault diagnosis problem, or the actual vehicle has a tendency to have problems different from those assumed during development, which requires re-learning of artificial intelligence. Therefore, the specific count can be a preset value within a range that can exclude situations where there are few complaining customers.

[0095] In step S480, when the cumulative count of misdiagnosis data is above a specific count, artificial intelligence re-learning is required, and accordingly, the device 100 for diagnosing a transmission based on artificial intelligence can generate a data set including new shifting data by adding the re-marked misdiagnosis data to the existing original data used to generate a feature set for the vehicle.

[0096] The device 100 for diagnosing a transmission based on artificial intelligence calculates the label ratio of the new data set. That is, in step S490, it is determined whether the label ratio of the new data set exceeds a predetermined value. To this end, the ratio can be obtained by counting the number of data for each label and dividing it by the total number of data. If the label ratio exceeds the specified value range, the re-learning of the artificial intelligence can be stopped, and the developer can be notified.

[0097] In step S500, when the label ratio is within a specified value range, the evaluation data can be separated, and the remaining data can be used to train artificial intelligence based on supervised learning.

[0098] For example, when the label ratio of the existing dataset is (1:1:1:1:1), and the specified criterion is set to 2, the following case can be considered.

[0099] Existing (1:1:1:1:1) -> (1.5:1.4:1.5+1:1:1.3): Failed. Because the label ratio exceeds the specified value range, AI learning is stopped and the developer is notified.

[0100] Existing (1:1:1:1:1) -> (1.5:1.4:1.5:1:1.3+0.4): Success. Since the label ratio is within the specified value range, the evaluation data is separated from the new dataset, and the remaining data is used to train the AI. Therefore, when the ratio of data labels is biased, the AI ​​decision criterion can be prevented from being distorted. In other words, the label ratio of the dataset used for supervised learning is configured to be uniform to achieve correct judgment.

[0101] The apparatus 100 for diagnosing a transmission based on artificial intelligence may determine whether an evaluation index regarding a new feature set of a vehicle generated by relearning a data set including second shift data by artificial intelligence based on supervised learning is above a preset level. That is, in step S510, the apparatus 100 for diagnosing a transmission based on artificial intelligence determines whether the accuracy is above a preset level and whether the FP rate is below a preset level as a result of evaluating the evaluation data.

[0102] If the accuracy or FP rate does not meet the preset level, the relearning of the AI ​​can be stopped and the developer can be notified. For example, if the evaluation metrics on a new data set reflecting misdiagnosed data are not good, this may mean that the preset learning variables (i.e., hyperparameters) of the AI ​​algorithm do not match the new data set, or that outliers are mixed in the new data set. In this case, it is impossible to solve the problem without the intervention of the developer, so the developer is informed of the problem so that appropriate measures can be taken.

[0103] In step S520, if the accuracy is above the preset level and the FP rate is below the preset level, the new feature set learned by the artificial intelligence is set as the representative feature set of the corresponding vehicle type and can be registered in the database DB. Thereafter, in step S530, the new feature set can be updated in the tablet computer of the diagnostic device.

[0104] Figure 5 is a table diagram showing a confusion matrix for evaluation indicators according to an embodiment.

[0105] In the confusion matrix-based performance evaluation index for transmission fault diagnosis, accuracy can indicate how accurately artificial intelligence predicts actual phenomena. For example, accuracy can be calculated as (TP+TN) / (TP+TN+FN+FP). The FP rate can indicate the degree of error in artificial intelligence's prediction of actual phenomena. The FP rate is an important index for fault determination, and if a fault (NG) is determined to be normal (OK) due to misdiagnosis, customer reliability decreases, which may lead to fatal failures. The FP rate can be calculated as FP / (TP+TN+FN+FP).

[0106] Here, regarding the prediction of artificial intelligence for gear shifting feel, true positive TP (true positive, TP) can represent situations where the affirmative prediction (normal) is correct (for example, the number of situations), true negative (true negative, TN) can represent situations where the negative prediction (fault) is correct (for example, the number of situations), false positive (false positive, FP) can represent situations where the affirmative prediction (normal) is wrong (for example, the number of situations), and false negative (false negative, FN) can represent situations where the negative prediction (fault) is wrong (for example, the number of situations).

[0107] When determining the accuracy and FP rate, the label ratio of the labeled data should also be uniform. For example, when the ratio of the data (OK:NG1:NG2:NG3:NG4) is 4:1:1:1:1 instead of 1:1:1:1:1, the ratio of OK data is higher than that of other NG data, so when TP is high, a distortion effect of increased accuracy may occur. That is, when TP increases compared to other cases in the equation of "Accuracy = (TP+TN) / (TP+TN+FN+FP)", the accuracy converges to 1, and a distortion effect of ignoring the influence of TN may occur.

[0108] Figure 6 is a schematic diagram illustrating a process of diagnosing a vehicle by using a method of diagnosing a transmission based on artificial intelligence according to an embodiment. Figure 7 is a schematic diagram illustrating a process of diagnosing a vehicle by using a method of diagnosing a transmission based on artificial intelligence according to an embodiment.

[0109] The method and device for diagnosing a transmission based on artificial intelligence according to the embodiment will refer to Figure 6 and Figure 7 Give a description.

[0110] refer to Figure 6 and Figure 7 The apparatus for diagnosing a transmission based on artificial intelligence according to the embodiment may be provided as a server 100. The server 100 may be connected to a diagnostic device DE provided as a tablet computer.

[0111] The device 100 for diagnosing a transmission based on artificial intelligence can be connected to a database DB and an artificial intelligence model AIM ( Figure 6 The device 100 for diagnosing a transmission based on artificial intelligence may include a database DB and an artificial intelligence model AIM ( Figure 7 ).

[0112] The device 100 for diagnosing a transmission based on artificial intelligence can optimize the artificial intelligence model AIM and the database DB. In an embodiment, the device 100 for diagnosing a transmission based on artificial intelligence can enhance the diagnostic performance of the transmission fault of the vehicle through the diagnostic device DE by optimizing the artificial intelligence model AIM and the database DB.

[0113] The implementation scheme is a method of determining by utilizing an artificial intelligence model that performs supervised learning through sensor values ​​of vehicle acceleration sensors installed identically for each vehicle type and vehicle status signals, rather than a method of determining in a threshold manner using an acceleration sensor in a diagnostic device.

[0114] That is, after training the artificial intelligence model based on supervised learning with the data marked by the developer at the time of development, the maintenance or evaluation personnel simultaneously measure the vehicle acceleration sensor value and the vehicle state value signal through the vehicle CAN communication measurement function of the diagnostic device, and determine and finely classify whether the situation is problematic by using the artificial intelligence model of supervised learning and the computing power of the diagnostic device or server. Afterwards, when it is determined to be a fault, instead of replacing the transmission or transmission components, the shift control learning value of the transmission control unit (TCU) is initialized, and the solenoid current control amount is adjusted to ensure that the shift control learning is optimally performed to improve the impact.

[0115] To this end, the diagnostic device of the embodiment may include a device capable of measuring the driving data of the vehicle through CAN communication and performing calculation, learning and displaying the results. In addition, the diagnostic device of the embodiment may be implemented as a diagnostic system including a diagnostic server for calculation and learning. The operating sequence of the diagnostic device or the diagnostic system is as follows, and an additional control compensation method may be performed according to the fault diagnosis result.

[0116] 1) Using the diagnostic device DE, the drive data during the shift control of the transmission determined to be problematic is measured at least 3 times. The drive data may include engine speed, torque converter speed, longitudinal acceleration, accelerator pedal opening, engine torque, transmission output speed, transmission temperature, engine temperature, etc. The drive data is measured from various sensors installed in the vehicle and received by the vehicle control device, and the diagnostic device receives the drive data through the vehicle control device and CAN communication.

[0117] 2) The diagnostic device DE can receive driving data, perform calculations (data preprocessing, determine whether the gear shift is OK / NG, determine whether control compensation can improve the problem in the NG case, and determine the amount of control compensation), and after sending the driving data to the artificial intelligence-based transmission diagnosis device 100 set as a separate server (or diagnostic server), receive the results of the performed calculations based on (for example) limited memory size, etc.

[0118] 3) Preprocessing the vehicle driving data using the artificial intelligence model (normalizing the maximum and minimum values ​​to predetermined specific values, and downsampling the signal value and measurement period).

[0119] 4) Pre-calculate the characteristic value of the acceleration signal of the pre-processed data and the period of each control range for determining whether to compensate for the control.

[0120] 5) The pre-processed data is input into an artificial intelligence model based on supervised learning (e.g., AIM, random forest, DNN, etc.), and finely classified as normal or faulty. For example, for data measured more than three times, the final decision is made to determine more than two situations.

[0121] 6) When it is determined to be an abnormal situation, it is determined whether the impact can be resolved by hardware (H / W) replacement or control compensation through a threshold-based determination method based on the calculated value of item 4). When the diagnostic server calculates, the decision result and the result of hardware (H / W) replacement or control compensation are sent to the diagnostic device through communication.

[0122] 7) Detailed decision results and whether the problem can be solved by H / W replacement or control compensation are displayed through the diagnostic equipment.

[0123] 8) When determining the control offset, all problematic shift learning values ​​of the transmission stored in the non-volatile memory in the transmission control unit (TCU) are initialized.

[0124] 9) After processing 8), a target of current control compensation is determined, and a compensation amount of solenoid current control is determined by utilizing the time of each control range of 4).

[0125] 10) For the contents corresponding to 8) and 9), the non-volatile memory in the transmission control unit is refreshed (reprogrammed) (TCU refresh). This provides the effect of allocating optimized control data of the transmission control unit for each vehicle's transmission hardware (H / W).

[0126] In the above-mentioned processing 1) to processing 10) of fault diagnosis and compensation of a transmission by utilizing the diagnostic device DE, the device 100 for diagnosing a transmission based on artificial intelligence can be set as a server, and can generate and update feature sets for each vehicle type and power system required for the artificial intelligence model AIM utilizing the diagnostic device DE. To this end, when it is necessary to generate a feature set, the device 100 for diagnosing a transmission based on artificial intelligence can extract and evaluate a list of similar candidate feature sets from the database DB to check whether they can be shared. In addition, the device 100 for diagnosing a transmission based on artificial intelligence can check and modify misdiagnosis regarding customer complaint data collected from repair shops, and reflect it to the artificial intelligence model. That is, the feature set reflecting the customer complaint data can be newly generated, and the database DB can be updated.

[0127] While the invention has been described in connection with what are presently considered to be practical embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but on the contrary is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A method for diagnosing a transmission based on artificial intelligence, comprising: Collect vehicle information and transmission shift data of the vehicle; Preprocess and mark the transmission shift data; generating a candidate feature set list including a plurality of candidate feature sets satisfying a specific criterion from among the feature sets stored in the database; Determining, for each of a plurality of candidate feature sets, an evaluation index of transmission shift comfort associated with the transmission shift data using artificial intelligence based on supervised learning; The feature set of the vehicle is determined by selecting a candidate feature set having a highest evaluation index among the candidate feature sets having evaluation indexes above a threshold value.

2. The method according to claim 1, further comprising: A new feature set is generated based on additional transmission shift data associated with the vehicle in accordance with a plurality of candidate feature sets that do not have an evaluation index above a threshold.

3. The method according to claim 1, wherein: Generate a list of candidate feature sets including: A plurality of feature sets regarding a plurality of other vehicles are determined as a plurality of candidate feature sets, each of the plurality of feature sets regarding the plurality of other vehicles including vehicle information that is the same as at least a portion of the vehicle information of the vehicle.

4. The method according to claim 3, wherein: The portion of the vehicle information of the vehicle includes at least one of a transmission model, whether an electric motor is used, an engine fuel, an engine displacement, or an engine type.

5. The method according to claim 3, wherein: The evaluation index is based on a confusion matrix-based performance evaluation index, which uses accuracy and false positive rate to indicate the performance of artificial intelligence based on supervised learning.

6. The method according to claim 5, wherein: The set of features that determine a vehicle includes: Among the candidate feature sets having accuracy rates above a second threshold and false positive rates below a third threshold, a candidate feature set based on the candidate feature set having the highest accuracy rate is selected.

7. The method according to claim 6, further comprising: The accuracy of the performance evaluation indicator based on the confusion matrix is ​​determined based on the formula (TP+TN) / (TP+TN+FN+FP), and the false positive rate of the performance evaluation indicator based on the confusion matrix is ​​calculated based on the formula FP / (TP+TN+FN+FP). Among them, for the transmission shifting comfort related to artificial intelligence based on supervised learning, TP means true positive, which indicates that the affirmative prediction is correct, TN means true negative, which indicates that the negative prediction is correct, FP means false positive, which indicates that the affirmative prediction is wrong, and FN means false negative, which indicates that the negative prediction is wrong.

8. The method according to claim 2, wherein: The set of features that determine a vehicle includes: obtain data about the vehicle; Preprocess and label the acquired data; Train artificial intelligence based on supervised learning based on preprocessed and labeled data; The feature set generated by training is stored in the database. Among them, training artificial intelligence based on supervised learning includes utilizing semi-supervised learning methods.

9. The method according to claim 8, wherein: The semi-supervised learning method includes the MixMatch algorithm.

10. A method for diagnosing a transmission based on artificial intelligence, the method comprising: measuring first transmission shift data of the vehicle; preprocessing the measured first transmission shift data; Determining a transmission shift comfort using artificial intelligence based on supervised learning and based on a first feature set associated with the vehicle; receiving information on whether the customer is dissatisfied with the transmission shifting comfort; receiving customer complaint data including preprocessed first transmission shift data and the determined transmission shift comfort based on determining that the information regarding dissatisfaction indicates that the customer has a complaint related to the determined transmission shift comfort; marking misdiagnosed data confirmed by a user based on a cumulative count of received customer complaint data being above a first threshold; generating second transmission shift data by adding the misdiagnosis data to the first feature set; Based on the second transmission shift data, training artificial intelligence based on supervised learning and generating a second feature set related to the vehicle; Based on the evaluation index of the trained supervised learning-based artificial intelligence being above a second threshold, a second feature set is stored.

11. The method according to claim 10, wherein: Generating the second transmission shift data includes: determining a label ratio of second transmission shift data; Based on the label ratio meeting the third threshold, artificial intelligence based on supervised learning is trained; Based on the fact that the label ratio does not meet the third threshold, the training of artificial intelligence based on supervised learning is terminated.

12. The method according to claim 10, wherein: The evaluation index is based on a confusion matrix-based performance evaluation index, which uses accuracy and false positive rate to indicate the performance of artificial intelligence based on supervised learning.

13. The method according to claim 12, wherein: Storing the second feature set includes: Based on the accuracy being above the fourth threshold and the false positive rate being below the fifth threshold, the second feature set trained by the supervised learning-based artificial intelligence is set as the representative feature set of the vehicle.

14. A device for diagnosing a transmission based on artificial intelligence, the device comprising: one or more processors; as well as a memory storing instructions that, when executed by one or more processors, cause the apparatus to: collecting vehicle information and transmission shift data related to the vehicle, and preprocessing the vehicle information and transmission shift data related to the vehicle; Marking the transmission shift data; generating a candidate feature set list including a plurality of candidate feature sets satisfying a specific criterion from among the feature sets stored in the database; Determining, for each of a plurality of candidate feature sets, an evaluation index of transmission shift comfort associated with the transmission shift data using artificial intelligence based on supervised learning; The feature set of the vehicle is determined by selecting a candidate feature set having a highest evaluation index among the candidate feature sets having evaluation indexes above a threshold value.

15. The device for diagnosing a transmission based on artificial intelligence according to claim 14, wherein: When the instructions are executed by one or more processors, the apparatus is further caused to: generate a new feature set based on additional transmission shift data associated with the vehicle based on the plurality of candidate feature sets having no evaluation index above the threshold.

16. The device for diagnosing a transmission based on artificial intelligence according to claim 15, wherein: When the instructions are executed by one or more processors, the apparatus is caused to: generate a new feature set by utilizing a semi-supervised learning method including a MixMatch algorithm.

17. The device for diagnosing a transmission based on artificial intelligence according to claim 14, wherein: When the instructions are executed by one or more processors, the device generates a candidate feature set list by determining multiple feature sets about multiple other vehicles as multiple candidate feature sets, each of the multiple feature sets about the multiple other vehicles including vehicle information that is the same as at least a portion of the vehicle information of the vehicle.

18. The device for diagnosing a transmission based on artificial intelligence according to claim 14, wherein: The evaluation indicator is based on a confusion matrix-based performance evaluation indicator, wherein the confusion matrix-based performance evaluation indicator indicates the performance of artificial intelligence based on supervised learning using accuracy and false positive rate; When the instructions are executed by one or more processors, the device determines a feature set of a vehicle by selecting a candidate feature set based on a candidate feature set with a highest accuracy rate from among candidate feature sets with an accuracy rate above a second threshold and a false positive rate below a third threshold.

19. The device for diagnosing a transmission based on artificial intelligence according to claim 14, wherein: When the instructions are executed by one or more processors, the device further causes the following to occur: The second transmission shift data is generated by: Marking misdiagnosed data confirmed by the user; Adding misdiagnosis data to existing data to generate an updated feature set; Based on the evaluation index associated with the updated feature set satisfying the second threshold, the updated feature set is determined as a final feature set of the vehicle, and the updated feature set is stored in the database.

20. The device for diagnosing a transmission based on artificial intelligence according to claim 19, wherein: When the instructions are executed by one or more processors, the apparatus is further caused to: Using a confusion matrix-based performance evaluation metric, wherein the confusion matrix-based performance evaluation metric is configured to evaluate the performance of the supervised learning-based artificial intelligence according to accuracy and false positive rate; According to the accuracy of the performance evaluation indicator based on the confusion matrix being above the third threshold and the false positive rate of the performance evaluation indicator based on the confusion matrix being below the fourth threshold, the updated feature set is determined as the final feature set of the vehicle.

Images