A vehicle intelligent diagnosis method and device

Abstract

The application discloses a kind of vehicle intelligent diagnosis method and device, wherein, method includes: obtaining the running data of vehicle object to be diagnosed at different time;The running data at different time is mapped, and the working load, abnormal times, first mapping conventional data group and actual performance evaluation index of the object to be diagnosed are obtained;According to the first mapping conventional data group and actual performance evaluation index, obtain the data group that is strongly related with the actual performance evaluation index;The actual performance evaluation index is compared with the ideal performance evaluation index of object to be diagnosed, and the overall comprehensive performance index function of the object to be diagnosed is constructed according to the comparison result;The working load, abnormal times, strongly related data group and overall comprehensive performance index function of object to be diagnosed are imported into pre-established health degree diagnosis model, and the health degree diagnosis data of the object to be diagnosed is obtained.The application can carry out intelligent prediction diagnosis to the parts or system of vehicle.

Description

Technical Field

[0001] This invention belongs to the field of intelligent connected vehicle technology, specifically relating to a vehicle intelligent diagnostic method and device. Background Technology

[0002] Current vehicle diagnostic functions primarily serve to troubleshoot existing faults. As vehicle intelligence and connectivity deepen, the ability to provide early warnings and notifications regarding fault occurrences and component performance becomes increasingly important. However, the industry currently lacks a method that combines existing vehicle operating data with real-time operating data to efficiently predict and intelligently diagnose impending vehicle faults, performance degradation, and component lifespan, and to issue early warnings to users and vehicle manufacturers. Summary of the Invention

[0003] The technical problem to be solved by the embodiments of the present invention is to provide a vehicle intelligent diagnostic method and device to intelligently predict and diagnose the faults, health status, performance degradation and service life of vehicle components or systems.

[0004] To address the aforementioned technical problems, this invention provides a vehicle intelligent diagnostic method, comprising:

[0005] Acquire operational data of the vehicle to be diagnosed at different times;

[0006] The operational data at different times are mapped to obtain the workload, number of anomalies, first-order mapping conventional data set, and actual performance evaluation indicators of the object to be diagnosed.

[0007] Based on the first-order mapping conventional data set and the actual performance evaluation index, obtain a data set that is strongly correlated with the actual performance evaluation index;

[0008] The actual performance evaluation index is compared with the ideal performance evaluation index of the object to be diagnosed, and the overall comprehensive performance index function of the object to be diagnosed is constructed based on the comparison result.

[0009] The workload, number of abnormalities, strongly correlated data sets, and overall comprehensive performance index function of the object to be diagnosed are imported into a pre-established health diagnosis model to obtain the health diagnosis data of the object to be diagnosed.

[0010] Furthermore, the workload, number of abnormalities, strongly correlated data sets, and overall comprehensive performance index function of the object to be diagnosed are imported into a pre-established health diagnosis model to obtain the health diagnosis data of the object to be diagnosed, specifically including:

[0011] Based on the workload, number of anomalies, strongly correlated data sets, and overall comprehensive performance index function at different times, multiple health diagnostic values ​​are calculated, and a health decay curve is obtained.

[0012] Based on the health decay curve, the overall health decay of the object to be diagnosed is predicted, and the time interval and confidence level of the object to be diagnosed being at risk of failure or malfunction are obtained.

[0013] Furthermore, by mapping the operational data at different times, the workload L of the object to be diagnosed is obtained as follows:

[0014]

[0015] Among them, U i I represents the actual operating voltage of the object to be diagnosed. i The actual operating current of the object to be diagnosed is represented by t, where t represents time and i represents each operating moment.

[0016] The number of abnormalities E of the object to be diagnosed is as follows:

[0017]

[0018] Where e represents the fault that occurs at a certain moment, t represents that moment, and r represents the number of times the fault occurs.

[0019] Furthermore, based on the first-order mapping conventional data set and the actual performance evaluation index, a data set strongly correlated with the actual performance evaluation index is obtained, specifically including:

[0020] The first-order conventional data set D1 and the actual performance evaluation index are imported into the first deep learning model. Parameters that have an impact on the actual performance evaluation index are selected from the first-order conventional data set D1, while parameters that have no impact on the actual performance evaluation index are excluded, resulting in a data set D2 that is strongly correlated with the actual performance evaluation index.

[0021] Furthermore, the actual performance evaluation index is compared with the ideal performance evaluation index of the object to be diagnosed, and a comprehensive performance index function of the object to be diagnosed is constructed based on the comparison result, specifically including:

[0022] The actual performance evaluation index P{P1,P2,P3,…,P under actual operating conditions n} and the ideal performance evaluation index P'{P1',P2',P3',…,P of the object to be diagnosed n By comparing '}, we obtain a P / P'{P1 / P1',P2 / P2',P3 / P3',…,P n / P n The data set of '};

[0023] Based on the data set P / P'{P1 / P1',P2 / P2',P3 / P3',…,P n / P n Construct an overall comprehensive performance index function P. 综 P 综 =Func(P / P')=Func(P1 / P1',P2 / P2',P3 / P3',…,P n / P n '), used to determine the overall performance of the object to be diagnosed in its current lifecycle.

[0024] Furthermore, the overall comprehensive performance index function P 综 Specifically, it is expressed as follows:

[0025]

[0026] in, a∈{a1, a2, a3, ..., a n},

[0027] n represents the number of evaluation indicators; k represents the sample size, i.e., the number of sampling points for each P / P' group; l represents the intercept; m j Represents the j-th term p j / p j The weight of '; a j Represents the j-th term p j / p j The order of '; ε represents the random error, used to correct P 综 function.

[0028] Furthermore, the health diagnostic model H is based on workload L, number of abnormalities E, strongly correlated data sets D2, and the overall comprehensive performance index function P. 综 The health diagnosis model is established using historical operating data from known working conditions; the specific functional form of the health diagnosis model is as follows:

[0029]

[0030] Where β(D2) represents the superposition effect factor of the strongly correlated data group D2 on the H function; θ is the intercept of the H function, representing the random error, which is used to correct the H function.

[0031] Furthermore, the process of establishing the health status diagnostic model H specifically involves: assigning values ​​to the H model using historical operating data under known working conditions, and combining a portion of the already assigned dependent variable H with its variable P. 综 E, L, and D2 are imported into a second deep learning model for training to obtain the dependent variable H and its variable P.综 The specific functional expressions of E, L, and D2; the remaining part of the dependent variable H, which has already been assigned values, and its variable P. 综 The health status diagnostic model H was obtained by testing E, L, and D2.

[0032] Furthermore, using a third deep learning model, a subset of existing operational data is used to evaluate the overall comprehensive performance index function P. 综 The training process is then performed, and the results are validated using the remaining data to obtain a reliable overall performance index function P that has been trained. 综 The reliable overall comprehensive performance index function P 综 Used to establish the health status diagnostic model H.

[0033] Furthermore, the vehicle intelligent diagnostic method also includes: uploading the obtained health diagnostic data of the object to be diagnosed to the cloud, or reminding and informing the user through the in-vehicle display screen or voice system in the form of HMI interaction, so as to realize full-stack early warning notification.

[0034] The present invention also provides a vehicle intelligent diagnostic device, comprising:

[0035] The first acquisition module is used to acquire the operating data of the vehicle to be diagnosed at different times;

[0036] The second acquisition module is used to map the running data at different times to obtain the workload, number of anomalies, first-order mapping conventional data set and actual performance evaluation indicators of the object to be diagnosed.

[0037] The third acquisition module is used to obtain a data set that is strongly correlated with the actual performance evaluation index based on the first-order mapping conventional data set and the actual performance evaluation index.

[0038] The fourth acquisition module is used to compare the actual performance evaluation index with the ideal performance evaluation index of the object to be diagnosed, and construct the overall comprehensive performance index function of the object to be diagnosed based on the comparison result.

[0039] The intelligent diagnostic module is used to import the workload, number of abnormalities, strongly correlated data groups, and overall comprehensive performance index function of the object to be diagnosed into a pre-established health diagnosis model to obtain the health diagnosis data of the object to be diagnosed.

[0040] Furthermore, the fourth acquisition module is specifically used for:

[0041] The actual performance evaluation index P{P1,P2,P3,…,P under actual operating conditions n} and the ideal performance evaluation index P'{P1',P2',P3',…,P of the object to be diagnosed n By comparing '}, we obtain a P / P'{P1 / P1',P2 / P2',P3 / P3',…,P n / P n The data set of '};

[0042] Based on the data set P / P'{P1 / P1',P2 / P2',P3 / P3',…,P n / P n Construct an overall comprehensive performance index function P. 综 P 综 =Func(P / P')=Func(P1 / P1',P2 / P2',P3 / P3',…,P n / P n '), used to determine the overall performance of the object to be diagnosed in its current lifecycle.

[0043] Furthermore, the overall comprehensive performance index function P 综 Specifically, it is expressed as follows:

[0044]

[0045] in, a∈{a1, a2, a3, ..., a n},

[0046] n represents the number of evaluation indicators; k represents the sample size, i.e., the number of sampling points for each P / P' group; l represents the intercept; m j Represents the j-th term p j / p j The weight of '; a j Represents the j-th term p j / p j The order of '; ε represents the random error, used to correct P 综 function.

[0047] Furthermore, the health diagnostic model H is based on workload L, number of abnormalities E, strongly correlated data sets D2, and the overall comprehensive performance index function P. 综 The health diagnosis model is established using historical operating data from known working conditions; the specific functional form of the health diagnosis model is as follows:

[0048]

[0049] Where β(D2) represents the superposition effect factor of the strongly correlated data group D2 on the H function; θ is the intercept of the H function, representing the random error, which is used to correct the H function.

[0050] Implementing this invention has the following beneficial effects: By performing intelligent diagnosis and prediction of vehicle malfunctions, performance degradation, and service life of automotive components and systems, this invention can assess the vehicle's usage status for users, provide real-time vehicle health monitoring and maintenance references, remind users of reasonable vehicle use and appropriate maintenance, and parts replacement plans, reducing safety issues and concerns caused by vehicle malfunctions, and improving user safety and comfort. It can also provide 4S stores with customized maintenance and repair suggestions, reducing maintenance material costs and the difficulty of troubleshooting. Furthermore, it provides vehicle manufacturers with statistical data analysis of vehicle parts and components, providing a basis for design improvement, storage, transportation, and supplier selection. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating a vehicle intelligent diagnostic method according to an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram of the specific process of a vehicle intelligent diagnostic method according to an embodiment of the present invention. Detailed Implementation

[0054] The following description of the embodiments is taken with reference to the accompanying drawings, which illustrate specific embodiments in which the invention can be implemented.

[0055] Please refer to Figure 1 As shown, Embodiment 1 of the present invention provides a vehicle intelligent diagnostic method, including:

[0056] Acquire operational data of the vehicle to be diagnosed at different times;

[0057] The operational data at different times are mapped to obtain the workload, number of anomalies, first-order mapping conventional data set, and actual performance evaluation indicators of the object to be diagnosed.

[0058] Based on the first-order mapping conventional data set and the actual performance evaluation index, obtain a data set that is strongly correlated with the actual performance evaluation index;

[0059] The actual performance evaluation index is compared with the ideal performance evaluation index of the object to be diagnosed, and the overall comprehensive performance index function of the object to be diagnosed is constructed based on the comparison result.

[0060] The workload, number of abnormalities, strongly correlated data sets, and overall comprehensive performance index function of the object to be diagnosed are imported into a pre-established health diagnosis model to obtain the health diagnosis data of the object to be diagnosed.

[0061] Specifically, please combine Figure 2 As shown, this embodiment requires prior design and association of the associated signal matrices for each function, as well as input of the data mapping algorithm library, based on the characteristics of different components and development experience. In this embodiment, the object to be diagnosed refers to the component or system.

[0062] First, based on the vehicle signal matrix, a correlation signal matrix for the vehicle to be diagnosed is constructed. There are two dimensions to constructing the correlation signal matrix: (1) all vehicle data are considered to be related to the part to be diagnosed, and all signals are used to construct a key signal matrix; (2) only some vehicle data are considered to be related to the part to be diagnosed, so for a single part to be diagnosed, only the correlation signal matrix related to it is constructed. In terms of data scale, the data scale of the second approach is smaller.

[0063] Based on the vehicle's functional lifecycle (including each sub-function or sub-action of the function or component), the system interacts with the vehicle's signal matrix, which includes body signals, vehicle GPS signals, vehicle mode, user mode, vehicle gear position, driver / passenger seat sensor signals, vehicle clock signals, vehicle interior / exterior temperature / humidity parameters, vehicle battery parameters, ignition status, and other parameters (including but not limited to these parameters). Through the vehicle signal matrix list, a related signal matrix for the component / system function is constructed (all signals on the vehicle can become a member of the related signal group, which is defined by the manufacturer).

[0064] For example, constructing a related signal matrix for the vehicle's front windshield wipers: selecting the current status signal of the wiper function, service mode, switch control signal, position adjustment signal, speed adjustment signal, wiper working status feedback output signal, wiper maintenance mode request command, rain sensing level signal from the rain and light sensor, as well as GPS positioning from the navigation system, outside temperature / humidity and inside temperature / humidity from the air conditioning system, current discharge current and voltage from the battery sensor and alternator, ignition status from the engine management module, etc., to construct a related signal matrix for the function of this component / system.

[0065] For example, a correlation signal matrix can be constructed for the low-voltage battery of a vehicle: select signals such as the current battery current, current range, battery voltage, battery temperature, cumulative charging capacity, cumulative discharging capacity, real-time maximum battery capacity status value, real-time maximum battery capacity, battery fault signal, GPS positioning from the navigation system, outside temperature / humidity and inside temperature / humidity from the air conditioning system, current discharge current and voltage from the alternator, ignition status from the engine management module, vehicle gear position, and vehicle mode and user mode from the vehicle management module to construct a correlation signal matrix for predicting the health of the battery.

[0066] Next, we construct a correlation signal matrix and a first-order data mapping algorithm library. Besides basic vehicle condition and environmental data, since the workload of a component can characterize its operational depreciation over its lifespan, and the number of anomalies can reflect its service life and health status, and component performance can be judged through performance evaluation indicators of different dimensions, the data mapping algorithm can achieve the following tasks (including but not limited to):

[0067] (1) Workload L mapping;

[0068] (2) Mapping of the number of anomalies E;

[0069] (3) Conventional signal mapping yields a first-order conventional data set (DataSet) D1;

[0070] (4) Create an evaluation index P{P1,P2,P3} to assess the actual performance of the component / system under real-world operating conditions. 3, …,P n} and calculate P{P1,P2,P3,…,P} by mapping each existing set of running data. n The dataset contains [database name]. The number of evaluation metrics for different parts is determined based on the actual situation.

[0071] The above content will design and import an algorithm for first-order data mapping of the raw signals from various functional modules of the vehicle, based on the design guidelines and engineering theories of vehicle manufacturers. The mapping algorithm needs to be built based on experience. For example, if the vehicle's interior temperature signal is acquired, (temperature signal × data precision - initial value) × a certain coefficient needs to be multiplied to be equivalent to the temperature of the predicted component. The reliability of this coefficient needs to be judged based on engineering experience. Understandably, the data mapping algorithm will be activated immediately after the vehicle bus network is woken up.

[0072] After constructing the correlation signal matrix and data mapping algorithm library for the object to be diagnosed, the runtime data at different times are subjected to first-order data mapping to obtain:

[0073] (1) Workload L mapping

[0074]

[0075] Among them, U i I represents the actual operating voltage of the component. i The actual operating current of the component is represented by t, where t represents time and i represents each operating moment.

[0076] (2) Mapping of the number of anomalies E

[0077]

[0078] Where e represents the fault that occurs at a certain moment, t represents that moment, and r represents the number of times the fault occurs; it should be noted that (e,t) is a fault array, representing the specific faults that occur and their corresponding times, and the subscript r is the number of faults counted.

[0079] (3) First-order conventional data group D1: The remaining data excluding component workload L and number of abnormalities E;

[0080] (4) Actual performance evaluation indicators and their dataset P{P1,P2,P3,…,P n For example, motor performance can be represented by the increase in the amplitude of jitter during the time required for the motor to perform the same ascending or descending operation under the same external environmental parameters.

[0081] For example:

[0082] (1) Working load L mapping of wiper motor: Total power output of wiper motor, with kW as the vertical axis and time T as the horizontal axis;

[0083] (2) Mapping of wiper malfunctions: Statistics on the number of malfunctions (including wiper output stall, wiper speed malfunction, wiper jamming, etc. If multiple types of malfunctions occur at the same time, they are treated as the total number of multiple types of malfunctions). The vertical axis is the number of times, and the horizontal axis is the time T.

[0084] (3) The conventional signal mapping yields a first-order conventional data set D1: including, but not limited to, mapping the time required for one complete wiping action by combining the wiper current status signal, wiper service mode, wiper switch control signal, wiper position adjustment signal, wiper speed adjustment signal, and wiper working status feedback signal (the time required for one complete wiping action is a sub-member of the first-order conventional data set D1). Then, combining the wiper position signal and the vehicle's wiper radius, the stroke rate of one wiper motor action is mapped. The vehicle's interior and exterior temperatures are continuously processed and compared with the temperature values ​​from the front / rear battery temperature sensors to calculate... The temperature sensor collects the temperature and the component placement difference coefficient. The current vehicle interior and exterior temperature / humidity is continuously processed and multiplied by the wiper motor difference coefficient to map the current motor operating ambient temperature / humidity. The rain sensor signal and the wiper current status are mapped to distinguish between automatic wiper and manual wiper operation scenarios. The current and voltage of the branch connecting the wiper to the battery and generator are converted into the wiper motor circuit model according to the electrical topology diagram. The vehicle GPS location information is mapped to geographical area distribution information through statistical algorithms and used as the geographical influence factor of historical ambient temperature / humidity.

[0085] (4) Actual performance evaluation index of wiper motor P{P1,P2,P3,…,P n} = {Output torque, power, output speed, power factor}

[0086] For example:

[0087] (1) Battery workload L mapping: total power output of battery, the vertical axis is kW, and the horizontal axis is time T;

[0088] (2) Abnormal number E mapping: Abnormal number (current abnormality, voltage abnormality, temperature abnormality, battery certification abnormality, response abnormality) statistics, the vertical axis is the number of times, and the horizontal axis is the time T. If multiple types of faults occur at the same time, they are processed according to the total number of multiple types of faults.

[0089] (3) Conventional signal mapping to obtain first-order conventional data group D1: By combining the battery capacity with vehicle mode, user mode signals and the vehicle power-on / off signal status and timestamps, the battery working time during a complete power-on / off process, the battery capacity change during each vehicle power-on / off, and the battery capacity change during vehicle use / sleep time are mapped. Among them, the battery capacity change during vehicle sleep time is equal to the power consumption of the vehicle ECU to the battery in the sleep state. The vehicle interior and exterior temperatures are continuously processed and compared with the temperature value of the battery temperature sensor. The difference coefficient of the air conditioning temperature and humidity sensor and the battery sensor placement is calculated. The current vehicle interior and exterior humidity is continuously processed and multiplied by the difference coefficient and mapped to the current ambient humidity of the battery working environment. The current current and voltage information obtained from all components, batteries and generators that use battery power are mapped to the current battery operating circuit model after electrical topology calculation. The vehicle GPS positioning information is mapped to the geographical distribution information through statistical algorithms and used as the geographical influence factor of historical ambient temperature / humidity. The driver / passenger seat sensor signals are converted to whether there is a person or not in the vehicle, thereby assisting in the influence of human and unmanned factors on the subject to be diagnosed.

[0090] (4) Actual performance evaluation index of the battery P{P1,P2,P3,…,P n} = {charging rate, discharging rate, real-time maximum battery capacity, discharging current, real-time battery capacity change rate, battery chemical aging offset rate}.

[0091] Considering the limited computing power of vehicle-side computation, and to ensure more accurate data range and reduce the computational scale of vehicle computation, this embodiment will use the obtained first-order conventional data set D1 and the actual performance evaluation index P{P1,P2,P3,…,P n The dataset} is imported into the first deep learning model, and a set of data with the actual performance evaluation metric P{P1,P2,P3,…,P} is selected. n The strongly correlated DataSet D2 is used to identify parameters that influence the actual performance evaluation index P in the first-order conventional DataSet D1, while excluding parameters that do not influence P. This achieves precise screening of the input sample (screening methods can include R-squared, T-test, F-test, etc.). In addition, it can also verify whether the actual performance evaluation index P is reliable.

[0092] On the other hand, the actual performance evaluation index P{P1,P2,P3,…,P3} under actual operating conditions will be used. n} and the ideal performance evaluation index P'{P1',P2',P3',…,P n(Ideal performance evaluation indicators are generally included in the specifications or instructions of the object to be diagnosed) are compared to obtain a P / P'{P1 / P1',P2 / P2',P3 / P3',…,P n / P n The data set of '}.

[0093] For example, by comparing the deviations of the motor's actual performance evaluation indicators with the performance data under ideal working conditions, the percentage of the motor's current overall performance value compared to when it left the factory can be mapped; or by comparing the differences in the above performance parameters of the left and right wiper motors, the percentage by which the overall performance value of the left motor is lower than that of the right motor can be determined.

[0094] Based on this, a comprehensive performance index function P is constructed. 综 P 综 =Func(P / P')=Func(P1 / P1',P2 / P2',P3 / P3',…,P n / P n The P / P' array is used to determine the overall performance of the object under diagnosis during its current lifecycle. It can be understood that P / P' is an array that calculates the percentage by dividing P by P'; P... 综 It consists of each P1 / P1', P2 / P2', P3 / P3', ..., P in the array. n / P n 'The result is calculated by taking into account the power effect, multiplying by a certain weighting coefficient, and adding an error term. It can be understood as a polynomial regression operation.'

[0095] P 综 The function is represented as follows:

[0096]

[0097] in, a∈{a1, a2, a3, ..., a n},

[0098] n represents the number of evaluation indicators constructed for different diagnostic subjects across multiple dimensions, and is an integer greater than or equal to 1; k represents the sample size, i.e., the number of sampling points for each P / P' group; l represents the intercept; m j Represents the j-th term p j / p j The weight of '; a j Represents the j-th term p j / p j The order of '; ε represents the random error, used to correct P 综 function.

[0099] The aforementioned workload L, number of abnormalities E, strongly correlated data set D2, and overall comprehensive performance index function P of the subject to be diagnosed are used to analyze the data. 综 Import the pre-established health status diagnosis model H, H = g(P) 综 The health status diagnostic data of the object to be diagnosed is obtained by (E,L,D2). The specific functional form of the health status diagnostic model H is:

[0100]

[0101] Where β(D²) represents the superposition effect of D² on the H function; θ is the intercept of the H function, representing random error, used to correct the H function. Both β(D²) and θ are obtained through machine learning when establishing the health diagnostic function H.

[0102] It should be noted that the pre-established health diagnostic model H in this embodiment is based on workload L, number of abnormalities E, strongly correlated data set D2, and overall comprehensive performance index function P. 综 The data source for establishing the health status diagnostic model H is historical operating data under known working conditions. The process of establishing the health status diagnostic model H is as follows:

[0103] The H model is assigned values ​​using historical operating data under known operating conditions, and a portion of the already assigned dependent variable H is compared with its variable P. 综 E, L, and D2 are imported into a second deep learning model for training to extract the dependent variable H and its variables (P). 综 The specific functional expression relationship of E, L, D2) (the specific functional expression relationship will be different due to different parts, and the final calculated and verified model expression will vary depending on the parts); the remaining part of the dependent variable H that has been assigned a value and its variable P 综 The health status diagnostic model H is obtained by testing E, L, and D2. Influencing factors include, but are not limited to: overall performance, workload, abnormal states / action records, and strongly correlated signal data. As mentioned above, β(D2) and θ are obtained through machine learning in this process.

[0104] For example, by importing wiper health value, overall wiper performance, wiper load, abnormal number records, wiper operation execution time, and whether the wiper operation was manual, along with wiper motor temperature / humidity, electrical environment, and geographical influencing factors into a deep learning model, a diagnostic function for wiper motor health can be derived.

[0105] Workload L, number of anomalies E, strongly correlated data set D2, and overall comprehensive performance index function P 综 The process of obtaining P is the same as described above and will not be repeated here. Specifically, using a third deep learning model, a portion of the existing running data is used to analyze P. 综The function is trained, and then tested using the remaining data to obtain a reliable trained P. 综 The function. It should also be noted that the strongly correlated data group D2 obtained from the aforementioned screening is also beneficial for constructing the H model: it reduces the sample size for calculation, makes the data range more accurate, and indirectly ensures the reliability of P as the input source of the H model.

[0106] Subsequently, this reliable H model is deployed to the vehicle-side edge computing platform to form a vehicle-side predictive diagnostic model. Then, the workload, number of anomalies, strongly correlated data sets, and overall comprehensive performance index function of the object to be diagnosed at different times T are fed into the vehicle-side predictive diagnostic model. A high-performance computer runs a full-stack edge computing and artificial intelligence learning algorithm to calculate multiple health diagnostic values. Through multiple iterations, a function H = f(T) of H with respect to T is obtained, specifically represented as a health decay curve. This health decay curve can be used to predict the overall health decay of the object to be diagnosed, determining the time interval and confidence level at which the object may experience fault or failure risks. As the input source of the H model, P... 综 E, L, and D2 are all calculated through real-time running data mapping, and a set of P is updated at every moment. 综 E, L, D2, and H and P 综 The model relationships of E, L, and D2 have been verified. A health diagnostic value can be calculated using the above parameters at each time step, resulting in a health decay curve. It should be noted that this embodiment does not define specific full-stack edge computing and artificial intelligence learning algorithms; as long as the results pass mathematical and engineering tests, they are acceptable. As an example, the RUL model, various RNN-based models, and grey prediction models can be used for prediction.

[0107] For example, by importing the wiper motor health value diagnostic model and its current operating data into the predictive diagnostic model for prediction, it can be concluded that the overall health value of the motor will approach 0 after running for t to t+n hours, and the confidence level of the overall health value being 0 between t and t+n can be obtained.

[0108] This embodiment can also formulate corresponding early warning strategies or cloud reporting strategies, uploading predicted diagnostic, service life, performance and other information to the cloud operation center for monitoring and status archiving, so as to notify users via SMS or vehicle manufacturer; at the same time, it can also remind and inform users through the in-vehicle display screen or voice system in the form of HMI interaction, realizing full-stack early warning notification; ensuring that the component or system can be repaired / maintained / replaced before failure, or guiding research and development design.

[0109] Corresponding to the vehicle intelligent diagnostic method described in Embodiment 1 of the present invention, Embodiment 2 of the present invention also provides a vehicle intelligent diagnostic device, comprising:

[0110] The first acquisition module is used to acquire the operating data of the vehicle to be diagnosed at different times;

[0111] The second acquisition module is used to map the running data at different times to obtain the workload, number of anomalies, first-order mapping conventional data set and actual performance evaluation indicators of the object to be diagnosed.

[0112] The third acquisition module is used to obtain a data set that is strongly correlated with the actual performance evaluation index based on the first-order mapping conventional data set and the actual performance evaluation index.

[0113] The fourth acquisition module is used to compare the actual performance evaluation index with the ideal performance evaluation index of the object to be diagnosed, and construct the overall comprehensive performance index function of the object to be diagnosed based on the comparison result.

[0114] The intelligent diagnostic module is used to import the workload, number of abnormalities, strongly correlated data groups, and overall comprehensive performance index function of the object to be diagnosed into a pre-established health diagnosis model to obtain the health diagnosis data of the object to be diagnosed.

[0115] Furthermore, the fourth acquisition module is specifically used for:

[0116] The actual performance evaluation index P{P1,P2,P3,…,P under actual operating conditions n} and the ideal performance evaluation index P'{P1',P2',P3',…,P of the object to be diagnosed n By comparing '}, we obtain a P / P'{P1 / P1',P2 / P2',P3 / P3',…,P n / P n The data set of '};

[0117] Based on the data set P / P'{P1 / P1',P2 / P2',P3 / P3',…,P n / P n Construct an overall comprehensive performance index function P. 综 P 综 =Func(P / P')=Func(P1 / P1',P2 / P2',P3 / P3',…,P n / P n '), used to determine the overall performance of the object to be diagnosed in its current lifecycle.

[0118] Furthermore, the overall comprehensive performance index function P 综 Specifically, it is expressed as follows:

[0119]

[0120] in, a∈{a1, a2, a3, ..., a n},

[0121] n represents the number of evaluation indicators; k represents the sample size, i.e., the number of sampling points for each P / P' group; l represents the intercept; m j Represents the j-th term p j / p j The weight of '; a j Represents the j-th term p j / p j The order of '; ε represents the random error, used to correct P 综 function.

[0122] Furthermore, the health diagnostic model H is based on workload L, number of abnormalities E, strongly correlated data sets D2, and the overall comprehensive performance index function P. 综 The health diagnosis model is established using historical operating data from known working conditions; the specific functional form of the health diagnosis model is as follows:

[0123]

[0124] Where β(D2) represents the superposition effect factor of the strongly correlated data group D2 on the H function; θ is the intercept of the H function, representing the random error, which is used to correct the H function.

[0125] For the working principle and process of this embodiment, please refer to the description of Embodiment 1 of the present invention, which will not be repeated here.

[0126] As can be seen from the above description, compared with the prior art, the beneficial effects of the present invention are as follows: The present invention can assess the vehicle's usage status for users by performing intelligent diagnosis and prediction of vehicle faults, performance degradation, and service life of automotive components and systems. It can provide real-time vehicle health monitoring and maintenance references, remind users to use the vehicle reasonably, perform appropriate maintenance, and plan for parts replacement, thereby reducing safety issues and concerns caused by vehicle faults and improving the user's safety and comfort. It can also provide 4S stores with customized maintenance and repair suggestions, reducing maintenance material costs and the difficulty of troubleshooting. Furthermore, it can provide vehicle manufacturers with statistical data analysis of vehicle parts and components, providing a basis for design improvement, storage, transportation, and supplier selection.

[0127] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A vehicle intelligent diagnosis method, characterized in that, The method comprises the following steps: acquiring running data of a vehicle to be diagnosed at different time points; mapping the running data at different time points to obtain a working load, an abnormality number, a first-mapping regular data set and an actual performance evaluation index of the vehicle to be diagnosed; obtaining a data set strongly correlated with the actual performance evaluation index according to the first-mapping regular data set and the actual performance evaluation index; comparing the actual performance evaluation index with an ideal performance evaluation index of the vehicle to be diagnosed, and constructing an overall comprehensive performance index function of the vehicle to be diagnosed according to a comparison result; inputting the working load, the abnormality number, the strongly correlated data set and the overall comprehensive performance index function of the vehicle to be diagnosed into a pre-established health degree diagnosis model to obtain health degree diagnosis data of the vehicle to be diagnosed.

2. The vehicle intelligent diagnosis method of claim 1, wherein, The method for obtaining the health degree diagnosis data of the vehicle to be diagnosed comprises the following steps: calculating a plurality of health degree diagnosis values according to the working load, the abnormality number, the strongly correlated data set and the overall comprehensive performance index function at different time points to obtain a health degree attenuation curve; predicting an overall health attenuation condition of the vehicle to be diagnosed according to the health degree attenuation curve to obtain a time interval and a confidence degree of a failure risk or an invalidation risk of the vehicle to be diagnosed.

3. The vehicle intelligent diagnostic method of claim 1, wherein, The working load L of the vehicle to be diagnosed obtained by mapping the running data at different time points is as follows: wherein U i represents the actual working voltage of the object to be diagnosed, I i represents the actual working current of the object to be diagnosed, t represents time, and i represents each working time; The abnormality number E of the vehicle to be diagnosed obtained by mapping the running data at different time points is as follows: wherein e represents a failure occurring at a certain time point, t represents the time point, and r represents a number of times of the failure.

4. The vehicle intelligent diagnosis method of claim 3, wherein, The method for obtaining the data set strongly correlated with the actual performance evaluation index comprises the following steps: inputting the first-mapping regular data set D1 and the actual performance evaluation index into a first deep learning model, screening parameters having an influence on the actual performance evaluation index from the first-mapping regular data set D1, and excluding parameters having no influence on the actual performance evaluation index to obtain a data set D2 strongly correlated with the actual performance evaluation index.

5. The vehicle intelligent diagnostic method of claim 4, wherein, The method for constructing the overall comprehensive performance index function of the vehicle to be diagnosed comprises the following steps: The actual performance evaluation index P{P1, P2, P3, …, P n} in actual operation is compared with the ideal performance evaluation index P'{P1', P2', P3', …, P n '} of the object to be diagnosed, to obtain a data set of P / P'{P1 / P1', P2 / P2', P3 / P3', …, P n / P n '}; According to the data set P / P'{P1 / P1', P2 / P2', P3 / P3', …, P n / P n '} an overall comprehensive performance index function P 综 , P 综 =Func(P / P')=Func(P1 / P1', P2 / P2', P3 / P3', …, P n / P n ') is constructed, which is used to judge the overall comprehensive performance of the object to be diagnosed in the current life cycle.

6. The vehicle intelligent diagnostic method of claim 5, wherein, Overall performance index function P 综 As follows: wherein n represents the number of evaluation indexes; k represents the number of sample points of each group of P / P'; l represents the intercept; m j represents the weight of the jth p j / p j ; a j represents the order of the jth p j / p j ; and ε represents a random error for correcting the P 综 function.

7. The vehicle intelligent diagnostic method of claim 6, wherein, The health degree diagnosis model H is based on the working load L, the number of exceptions E, the strongly related data set D2 and the overall comprehensive performance index function P 综 The health degree diagnosis model H is based on the working load L, the number of exceptions E, the strongly related data set D2 and the overall comprehensive performance index function P wherein β(D2) represents an additive influence factor of the strongly correlated data set D2 on the H function, θ is an intercept of the H function, and φ represents a random error for correcting the H function.

8. The vehicle intelligent diagnostic method of claim 7, wherein, The process of establishing the health degree diagnosis model H is: using the historical operation data of the known working conditions to assign values to the H model, and using a part of the dependent variable H and its variable P 综 , E, L, D2 to import a second deep learning model for training to obtain the dependent variable H and its variable P 综 , E, L, D2 of the specific function expression relationship; using the remaining another part of the dependent variable H and its variable P 综 , E, L, D2 for testing, and finally obtaining a reliable health degree diagnosis model H.

9. The vehicle intelligent diagnosis method of claim 8, wherein, Using the third deep learning model, using a part of the existing operation data to the overall comprehensive performance index function P 综 Training, testing with the remaining part of the data, to get a trained reliable overall comprehensive performance index function P 综 ; The reliable overall comprehensive performance index function P 综 for establishing the health degree diagnostic model H.

10. The vehicle intelligent diagnostic method of any one of claims 1-9, wherein, The method further comprises the following steps: uploading the health degree diagnosis data of the vehicle to be diagnosed to a cloud, or reminding and informing a user in an HMI interactive manner through a display screen or a voice system in the vehicle to realize full-stack early warning.

11. A vehicle intelligent diagnosis device, characterized in that, The method comprises the following steps: a first acquisition module configured to acquire running data of a vehicle to be diagnosed at different time points; a second acquisition module configured to map the running data at different time points to obtain a working load, an abnormality number, a first-mapping regular data set and an actual performance evaluation index of the vehicle to be diagnosed; The third obtaining module is configured to obtain a data group strongly correlated with the actual performance evaluation index according to the first-order mapping regular data group and the actual performance evaluation index; The fourth obtaining module is configured to compare the actual performance evaluation index with an ideal performance evaluation index of the object to be diagnosed, and construct an overall comprehensive performance index function of the object to be diagnosed according to a comparison result; The intelligent diagnosis module is configured to input the workload, the number of abnormalities, the strongly correlated data group and the overall comprehensive performance index function of the object to be diagnosed into a pre-established health degree diagnosis model, and obtain health degree diagnosis data of the object to be diagnosed.

12. The vehicle intelligent diagnostic apparatus of claim 11, wherein, The fourth obtaining module is configured to: The actual performance evaluation index P{P1, P2, P3, …, P n} in the actual running condition is compared with the ideal performance evaluation index P'{P1', P2', P3', …, P n '} of the object to be diagnosed, to obtain a data set of P / P'{P1 / P1', P2 / P2', P3 / P3', …, P n / P n '}; According to the data set P / P'{P1 / P1', P2 / P2', P3 / P3', …, P n / n ' constructs an overall comprehensive performance index function P 综 , P 综 =Func(P / P')=Func(P1 / P1', P2 / P2', P3 / P3', …, P n / n '), which is used to judge the overall comprehensive performance of the object to be diagnosed in the current life cycle.

13. The vehicle intelligent diagnostic apparatus of claim 12, wherein, Overall performance index function P 综 As follows: Wherein, n represents the number of evaluation indexes; k represents the number of sample points of each group of P / P'; l represents the intercept; m j represents the weight of the jth p j / p j ; a j represents the order of the jth p j / p j ; and ε represents a random error for correcting the P 综 function.

14. The vehicle intelligent diagnostic apparatus of claim 13, wherein, The health degree diagnosis model H is based on the working load L, the number of exceptions E, the strongly related data set D2 and the overall comprehensive performance index function P 综 The health degree diagnosis model H is based on the working load L, the number of exceptions E, the strongly related data set D2 and the overall comprehensive performance index function P Wherein, β (D2) represents a superimposed influence factor of the strongly correlated data group D2 on the H function; θ is an intercept of the H function, representing a random error, and used for correcting the H function.

Images