First-aid repair auxiliary decision-making method and system based on machine learning algorithm

Abstract

The invention provides a first-aid repair auxiliary decision-making method and system based on a machine learning algorithm, and relates to the technical field of automobile maintenance, the auxiliary decision-making system comprises a portable maintenance auxiliary terminal, and the surface of the portable maintenance auxiliary terminal is connected with a special interface adapter through a data acquisition cable; the system is further provided with a fault diagnosis module and a multi-task flow data buffering module, and the fault diagnosis module predicts faults of the vehicle through an RVM. The multi-task flow data buffer module adopts multiple tasks simultaneously, can form an independent system and can also be used in cooperation with an extension test system and an automatic test system, effective means support is provided for first-aid repair work, the problem that common detection and maintenance equipment is difficult to operate in a narrow space and a battlefield environment can be solved, and the detection and maintenance efficiency is improved. And the whole maintenance process is high in stability, the fault information checking process is efficient and convenient, and the efficiency of the subsequent repair process is improved.

Description

Technical Field

[0001] The present application relates to the technical field of automobile repair and maintenance, and specifically to an emergency repair auxiliary decision-making method and system based on a machine learning algorithm. Background Art

[0002] The systems inside automobiles are huge, complex in structure, highly technology-intensive, highly information-intensive, and have diversified information security. They involve multiple disciplines such as mechanics, electronics, optoelectronics, control, and communications. Technical inspection, fault diagnosis, and emergency repairs are difficult, and the security technology requirements are high. This puts higher demands on maintenance and security technical means, methods, and equipment.

[0003] In the existing technology, maintenance and support equipment comes from different R&D and production units, and the detection and maintenance equipment of each subsystem is relatively discrete and numerous. They are all large in size, high in cost, and complex in maintenance. The fault diagnosis efficiency is low and it is unable to assist in emergency repair decision-making. The workload of manual classification and distributed technical detection and fault diagnosis is huge, which also brings certain difficulties to vehicle maintenance and support work. In addition, the internal space of most vehicles is small, which also brings certain difficulties to the use of detection and maintenance and support equipment. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of this application is to provide a method and system for emergency repair decision-making assistance based on machine learning algorithms to solve the problems raised in the above-mentioned background technology. This application can be an independent system or can be used in conjunction with an extension test system and an automated test system to provide effective means of support for emergency repair work. It can solve the problem that general detection and maintenance equipment is difficult to operate in narrow spaces and battlefield environments, and the entire maintenance process has high stability, and the fault information viewing process is efficient and convenient.

[0005] In order to achieve the above-mentioned purpose, the present application is implemented through the following technical solutions: The present application provides an emergency repair decision-making assistance method based on a machine learning algorithm, including the following contents: establishing a dynamic fault tree through qualitative and quantitative analysis methods; establishing a troubleshooting guide and a repair guide for each fault cause in the fault tree; arranging fault case information according to the fault name, and organizing it into a database for storage, management and reuse; establishing a fault diagnosis model; connecting to a portable maintenance auxiliary terminal to perform fault diagnosis; recording the troubleshooting data, and regularly updating and maintaining the case library, fault tree model, and troubleshooting and repair guide library information;

[0006] The decision-making method also uses RVM to predict vehicle failures:

[0007] Given a training sample set is the characteristic value in the sample set, and the target value t nIndependent and identically distributed, with mean 0 and variance σ 2 Gaussian noise ε n ,but

[0008] t n =y(X n ; w)+ε n Formula (1.1)

[0009] The model output of RVM can be defined as:

[0010]

[0011] Where N is the number of samples, and the weight vector w = [ω1,...,ω N ] T , φ is a matrix of order N×(N+1), and φ=[φ(x1),φ(x2),...,φ(x N )] T , where φ(x n )=[K(x n ,x1),K(x n ,x2),...,K(x n ,x N )] T , K(x,x i ) is a nonlinear function;

[0012] Since the target value t n Independent, the likelihood function of the training sample set is expressed as:

[0013]

[0014] Where, the target vector t=[t1,...,t N ] T

[0015] According to the structural risk minimization principle of support vector machine, if the weight w is not constrained and equation (1.2) is directly maximized, overfitting may occur. Therefore, each weight in RVM defines its own Gaussian prior probability distribution:

[0016]

[0017] In the formula, α=[α1,α2...,α N ] T is a hyperparameter, which determines the prior distribution of the weight w. Each hyperparameter α i Corresponding to a weight ω i ;

[0018] Given the prior probability distribution and likelihood distribution, the posterior probability distribution of the weights is calculated based on the Bayesian criterion as follows:

[0019]

[0020] In the formula, the mean μ and covariance ∑ of the posterior weights are expressed as:

[0021] μ=σ -2 ∑φ T t formula (1.6)

[0022] ∑=(σ -2 φ T φ+A) -1 Formula (1.7)

[0023] Among them, A=diag(α1,α2,...,α N )

[0024] The mean μ of the posterior distribution of the weights determines the estimate of the weights, and the covariance ∑ represents the uncertainty of the model prediction. Therefore, in order to obtain the weights of the model, it is first necessary to estimate the optimal value of the hyperparameters. The likelihood distribution of the hyperparameters, that is, the Gaussian distribution, is expressed as:

[0025]

[0026] In the formula, covariance C = σ 2 I+φA -1 φ T

[0027] Maximizing the likelihood distribution of hyperparameters yields α and σ 2 The most likely value is calculated by using the iterative estimation method instead of the analytical calculation method to calculate equation (1.7). 2 Take the derivative of equation (1.7) respectively, then set it to zero and rewrite the formula:

[0028]

[0029] In the formula, μ i is the mean of the ith posterior weights obtained by formula (1.6); i =1-α i ∑ ii ,∑ ii is the current α and σ 2 The i-th diagonal element of the posterior weight covariance matrix obtained by formula (1.7);

[0030] While repeatedly calculating equations (1.9) and (1.10), keep updating equations (1.6) and (1.7) until a certain convergence condition is met and the iteration is stopped. During the iteration, a large number of α tending to infinity appear. i According to formula (1.5), we can get p(w|t,α,σ 2 ) appears at the maximum peak as the zero point, so it is considered that its ω i is zero, thus producing sparse solutions, somewhat similar to the SV of SVM, those with non-zero ω i The corresponding learning sample is the relevant vector.

[0031] Furthermore, the qualitative analysis is used to find out the combination of all fault causes that lead to the fault. This process obtains all fault causes of the fault by analyzing the working principle, fault mechanism, working process, and performance of the vehicle; the quantitative analysis includes the probability of occurrence of the fault cause, the difficulty of troubleshooting, and the vulnerability of the faulty part; the troubleshooting guide provides steps and guidance methods for detecting and troubleshooting the faulty part of the fault cause, and the repair guide is used to provide steps and guidance methods for repairing the faulty part of the fault cause; by establishing a fault diagnosis model, the signals of each device are quantized on the basis of vehicle signal acquisition and conversion to form a one-dimensional feature vector representation, and all eigenvalues ​​are normalized to between [0,1]. The system forms a diagnostic strategy for real-time fault discovery through a machine learning dynamic fault case library, and finally provides a troubleshooting and emergency repair guide.

[0032] Furthermore, the auxiliary decision method uses BP neural network to diagnose vehicle faults, and the diagnosis process includes:

[0033] First, initialize the network and initialize the values ​​of each variable according to the dimension of the parameters, the number of neurons in the hidden layer, and the dimension of the output layer. i ,ω ji , b ji , η, α; then, the network is forward propagated, the output of the previous layer is used as the input of the next layer, and the sigmoid function is used as the activation function, and finally the total error E of the network is obtained; then, the network is back-propagated, according to E and the gradient descent method, the weights and thresholds of the network are continuously iterated and updated. In addition, the impulse term is added, which can cross some narrow local minima and reach a smaller place; finally, it is determined whether the algorithm iteration is completed;

[0034] The auxiliary decision-making method also includes a naive Bayes algorithm model. For the fault diagnosis problem, according to the Bayesian theorem, the fault feature to be classified is set to d, and the fault belongs to category c. k The probability of is set to:

[0035]

[0036] Fault feature d contains n feature attributes, d = {w1, w2, ..., w n}, Naive Bayes assumes that these n feature attributes are independent of each other, so formula (3-2) can be expressed as:

[0037]

[0038] For any fault type, the denominator of the above formula is the same and does not need to be considered, so the parameters to be estimated are p(wi|ck) and p(ck). Then, these parameters can be estimated using the maximum likelihood method:

[0039]

[0040] Where N represents the number of training faults, y j Indicates the class label of the jth fault, x j Represents the features contained in the jth fault, and the final classification results are as follows:

[0041]

[0042] Furthermore, the auxiliary decision method separates the data by constructing a segmentation surface, and the decision boundary of SVM is the maximum margin hyperplane solved for the training samples;

[0043] This surface is also called the maximum margin hyperplane. The optimal separation line can accurately distinguish two different types of data. The calculation formula of the separation line is as follows:

[0044]

[0045] Where w and b are two constants, and the value of x is the horizontal coordinate value. (x i ,y i ) as a set of linearly separable sample points, i = 1, ..., n, y i is the sample label, with a value between -1 and +1;

[0046] The calculation formula for the dist gap between two different categories of data is as follows:

[0047]

[0048] The SVM model makes the support vectors of all categories as far away from the segmentation hyperplane as possible, which can be expressed by the following mathematical formula:

[0049]

[0050] st:y (i) (w Tx (i) +b)≥1,i=1,2,…,m

[0051] The Lagrangian optimization method can be used to find the optimal interval hyperplane, transforming the optimization problem into a dual problem. The optimization problem is transformed into the following:

[0052]

[0053] Since Lagrangian has duality characteristics, the optimization objective can be converted into an equivalent dual problem to solve, so that the optimization objective becomes:

[0054] max β≥0 min w,b L(w,b,β)

[0055] Therefore, for this optimization function, it can correspond to finding the maximum value of the Lagrange multiplier β. For processing linearly inseparable data, a kernel function can be introduced to map low-dimensional data to a high-dimensional space for segmentation. The kernel function includes the following:

[0056] (1) Linear Kernel

[0057] K(x,z)=x·z

[0058] (2) Polynomial Kernel

[0059] K(x,z)=(γx·z+r) d

[0060] (3) Gaussian Kernel

[0061]

[0062] (4) Sigmoid Kernel

[0063] K(x,z)=tanhγx·z+r

[0064] A decision system as used in the above-mentioned decision method, the auxiliary decision system includes a portable maintenance auxiliary terminal, the surface of the portable maintenance auxiliary terminal is connected to a dedicated interface adapter through a data acquisition cable, the surface of the portable maintenance auxiliary terminal is also connected to a display unit part through a data transmission cable, the surface of the display unit is embedded with a display screen, and a locking mechanism is installed at the bottom of the portable maintenance auxiliary terminal. The portable maintenance auxiliary terminal performs signal transfer and physical docking with a vehicle under test through an interface adapter; the system is also equipped with a fault diagnosis module and a multi-task stream data buffer module, the fault diagnosis module uses RVM to predict vehicle faults; the multi-task stream data buffer module uses multiple tasks simultaneously, and the tasks are divided into conventional data acquisition tasks, video data acquisition tasks, message parsing tasks, and processing output tasks.

[0065] Furthermore, a back panel is screwed onto the back of the display unit, a connecting rod is installed on the surface of the back panel, an end panel is integrally formed with one end of the back panel, a rotating shaft is installed on the inner side of the end panel, a supporting shaft is installed on the other end of the back panel, a supporting rod is installed on the surface of the supporting shaft, and a connecting column is installed on the end of the supporting rod.

[0066] Furthermore, a second suction cup is installed at the end of the plug-in column, a slot is provided at the top of the second suction cup, a rotating sleeve is provided at the top of the support rod, and the rotating sleeve is used to be sleeved on the surface of the support shaft, and the support rod as a whole rotates and folds around the support shaft; a fixed sleeve is integrally formed on the surface of the rotating shaft, and a first suction cup is installed on the side of the fixed sleeve, and the number of the first suction cup and the second suction cup are both two.

[0067] Furthermore, a shell is provided on the surface of the portable maintenance auxiliary terminal, a base is screwed to the bottom of the shell, a side baffle is integrally formed on the side of the shell, a sandwich layer is provided inside the shell, and a storage groove is provided on the top of the shell.

[0068] Furthermore, a front baffle is welded on the top of the base, and a latch is inserted into the surface of the front baffle. The latch passes through the front baffle and is used to be embedded in the slot. The interface adapter is used to be inserted into the storage slot, and the display unit is placed as a whole inside the interlayer.

[0069] Furthermore, the locking mechanism includes a transmission box, a screw rod and a motor. A notch is provided in the middle of the transmission box, and threaded sleeves are provided at both ends of the transmission box. A screw rod is inserted into the interior of the threaded sleeve, and an extrusion plate is integrally formed at the end of the screw rod. A driving shaft is inserted into the output end of the motor; a driving gear is keyed on the surface of the driving shaft, a toothed belt is sleeved on the surface of the driving gear, and a driven gear is sleeved on the other end of the toothed belt, and a plug-in rod is inserted on the side of the driven gear, and the plug-in rod is embedded in the interior of the screw rod. There are two plug-in rods, and the toothed belt penetrates into the interior of the transmission box from the notch, and the driving shaft is inserted into the interior of the base as a whole.

[0070] Beneficial effects of this application:

[0071] 1. This application can provide detailed maintenance information for on-site operators to perform in-situ maintenance by establishing a dynamic vehicle fault case library, including technical information, status information, fault diagnosis methods, troubleshooting steps, verification methods, as well as historical records, spare parts information, etc., providing important references for vehicle maintenance and emergency repair work. It can also realize the detection of the main systems and signals of a certain vehicle, and determine the vehicle status and performance through machine learning algorithms, etc., to provide effective means to support emergency repair work.

[0072] 2. This application provides a mobile auxiliary testing and maintenance technology that can be used on-site, provides troubleshooting and repair guides for vehicle maintenance support and emergency repairs, provides technical support for emergency repair decisions, and solves the problem that general testing and maintenance equipment is difficult to operate in narrow spaces and battlefield environments.

[0073] 3. This application adopts a multi-task method to avoid wasting system resources and affecting work efficiency, and also solves the problem of possible message or data loss. At the same time, through qualitative and quantitative analysis methods, a dynamic fault tree is established, allowing users to locate the cause of the fault more quickly.

[0074] 4. This application obtains information from the test equipment with the help of a portable maintenance auxiliary terminal in the early stage. The dedicated instrument module resources perform signal acquisition and status monitoring on different subsystems of different large and complex technical vehicles, and directly displays and processes the fault information in real time through the display unit. With the help of the display unit's own structure, it can be directly installed on the vehicle's windshield or side window glass, so it can be directly viewed whether it is being repaired outside or inside the vehicle, and the connection stability of the interface adapter part is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A flowchart of a method for assisting decision-making in emergency repair based on a machine learning algorithm is provided for this application;

[0076] Figure 2 This is a schematic diagram of the structure of the portable maintenance auxiliary terminal part of this application;

[0077] Figure 3 This is a schematic diagram of the back structure of the display unit part of this application;

[0078] Figure 4 This is a structural diagram of the second suction cup part in the display unit of this application;

[0079] Figure 5 This is a structural diagram of the interlayer part of the portable maintenance auxiliary terminal of this application;

[0080] Figure 6 This is the external structure diagram of the locking mechanism part of this application;

[0081] Figure 7 This is a schematic diagram of the internal connections of the locking mechanism of this application;

[0082] Figure 8 This is a diagram of the linkage structure inside the base and the transmission box of this application.

[0083] In the figure: 1. portable maintenance auxiliary terminal; 2. interface adapter; 3. locking mechanism; 4. display unit; 5. back plate; 6. connecting rod; 7. end plate; 8. rotating shaft; 9. fixed sleeve; 10. first suction cup; 11. support rod; 12. support shaft; 13. plug-in column; 14. second suction cup; 15. slot; 16. side baffle; 17. interlayer; 18. front baffle; 19. latch; 20. storage slot; 21. transmission box; 22. notch; 23. screw rod; 24. extrusion plate; 25. threaded sleeve; 26. motor; 27. drive shaft; 28. driving gear; 29. ​​toothed belt; 30. driven gear; 31. plug-in rod; 32. housing; 33. rotating sleeve; 34. data acquisition cable; 35. data transmission cable; 36. base. DETAILED DESCRIPTION

[0084] In order to make the technical means, creative features, objectives and effects achieved by this application easy to understand, this application is further explained below in conjunction with specific implementation methods.

[0085] First, see Figures 1 to 8The present embodiment provides the following technical solutions: a repair auxiliary decision system based on a machine learning algorithm, wherein the auxiliary decision system includes a portable maintenance auxiliary terminal 1, the surface of the portable maintenance auxiliary terminal 1 is connected to a dedicated interface adapter 2 through a data acquisition cable 34, and the surface of the portable maintenance auxiliary terminal 1 is also connected to a display unit 4 through a data transmission cable 35, the surface of the display unit 4 is embedded with a display screen, and a locking mechanism 3 is installed at the bottom of the portable maintenance auxiliary terminal 1, and the portable maintenance auxiliary terminal 1 performs signal transfer and physical docking with the vehicle under test through the interface adapter 2; the system is also equipped with a fault diagnosis module and a multi-task stream data buffer module, the fault diagnosis module uses RVM to predict vehicle faults; the multi-task stream data buffer module uses multiple tasks simultaneously, and the tasks are divided into conventional data acquisition tasks, video data acquisition tasks, message parsing tasks, and processing output tasks.

[0086] With the help of the portable maintenance auxiliary terminal 1, information is obtained from the test equipment, and the dedicated instrument module resources perform signal acquisition and status monitoring for different subsystems of different large and complex technical vehicles, and the fault information is directly displayed and processed in real time through the display unit 4. With the help of the display unit 4's own structure, it can be directly installed on the vehicle's windshield or side window glass, so it can be directly viewed whether it is being repaired outside or inside the vehicle, and the connection stability of the interface adapter 2 is higher.

[0087] In this application, the system needs to continuously monitor the data packets of the CAN communication interface, the action data of multiple RS-422 interfaces, the laser power measurement data, the voltage and current monitoring data, and the six-channel video data on the one hand; on the other hand, it needs to classify and process the received CAN frame data packets, analyze the messages, judge errors, save data or record faults, display videos, etc. If the whole process is executed serially, it will inevitably waste system resources and affect work efficiency, which may lead to the problem of message or data loss.

[0088] The method of carrying out multiple tasks simultaneously in this embodiment can effectively solve such problems, but this must be based on scientific and reasonable business division and rational use of resources. This embodiment divides the tasks into four major parts: conventional data collection tasks; video data collection tasks; message analysis tasks; and processing output tasks.

[0089] (1) Conventional data collection tasks: monitor the data packets of the CAN communication interface, the action data of the RS-422 interface, the laser power measurement data, the voltage and current monitoring data, and insert them into the message queue to be parsed at the application layer;

[0090] (2) Video data acquisition task: obtain the data of six channels of video and insert it into the application layer video data queue;

[0091] (3) Message parsing task: parse the data packets of the CAN communication interface and the action data of the RS-422 interface, and insert them into the application layer parsing message queue;

[0092] (4) Process output tasks: obtain data from each queue and output it to the display terminal.

[0093] In this embodiment, a back plate 5 is screwed to the back of the display unit 4, a connecting rod 6 is inserted on the surface of the back plate 5, an end plate 7 is integrally formed at one end of the back plate 5, a rotating shaft 8 is inserted on the inner side of the end plate 7, a supporting shaft 12 is inserted at the other end of the back plate 5, a supporting rod 11 is installed on the surface of the supporting shaft 12, and a plug-in column 13 is inserted at the end of the supporting rod 11. A second suction cup 14 is installed at the end of the plug-in column 13, a slot 15 is provided at the top of the second suction cup 14, a rotating sleeve 33 is provided at the top of the support rod 11, and the rotating sleeve 33 is used to be sleeved on the surface of the supporting shaft 12, and the support rod 11 is rotated and folded around the supporting shaft 12 as a whole. A fixed sleeve 9 is integrally formed on the surface of the rotating shaft 8, and a first suction cup 10 is installed on the side of the fixed sleeve 9, and the number of the first suction cup 10 and the second suction cup 14 are both two.

[0094] Specifically, when the portable maintenance auxiliary terminal 1 is used, the display unit 4 is directly taken out from the inner part of the interlayer 17, and the portable maintenance auxiliary terminal 1 is fixed to the signal transfer and physical docking area of ​​the vehicle by controlling the locking mechanism 3. After the display unit 4 is pulled out, it is directly fixed on the front windshield of the vehicle. At this time, if the inner area of ​​the vehicle is to be inspected, the display unit 4 can be directly formed into a Figure 2 In the state shown, the display screen is directed toward the interior of the vehicle, and the first suction cup 10 is attached to the inner side of the front windshield of the vehicle, the bottom support rod 11 is folded toward the bottom, and the second suction cup 14 is controlled to rotate until the second suction cup 14 is partially extended downward to provide upward support for the entire display unit 4. If the outer area of ​​the vehicle needs to be inspected, the display unit 4 is controlled to rotate to form the following state. Figure 3 state, at this time, directly adsorb the first suction cup 10 and the second suction cup 14 on the windshield, and you can directly view the vehicle fault information status currently collected by the portable maintenance auxiliary terminal 1 through the display screen from the outside of the vehicle, without the need to always hold the entire display unit 4 for mobile use.

[0095] In this embodiment, the surface of the portable maintenance auxiliary terminal 1 is provided with a shell 32, the bottom of the shell 32 is screwed with a base 36, the side of the shell 32 is integrally formed with a side baffle 16, the inside of the shell 32 is provided with an interlayer 17, and the top of the shell 32 is provided with a storage slot 20. A front baffle 18 is welded to the top of the base 36, and a bayonet 19 is inserted on the surface of the front baffle 18. The bayonet 19 is used to be embedded in the inside of the card slot 15 after passing through the front baffle 18, the interface adapter 2 is used to be inserted and placed in the inside of the storage slot 20, and the display unit 4 is placed in the inside of the interlayer 17 as a whole. Specifically, the entire display unit 4 is stored inside the shell 32 through the interlayer 17. When not in use, the display unit 4 is directly embedded in the interlayer 17, and the first suction cup 10 and the second suction cup 14 are adsorbed onto the inner wall of the interlayer 17. The end of the second suction cup 14 is directly locked by passing through the front baffle 18 with the help of the locking pin 19, and the locking mechanism 3 at the bottom can be used to fix the entire display unit 4.

[0096] In this embodiment, the locking mechanism 3 includes a transmission box 21, a screw rod 23 and a motor 26. A notch 22 is provided in the middle of the transmission box 21. Threaded sleeves 25 are provided at both ends of the transmission box 21. The screw rod 23 is inserted into the threaded sleeve 25. The end of the screw rod 23 is integrally formed with an extrusion plate 24. A drive shaft 27 is inserted into the output end of the motor 26. A driving gear 28 is keyed to the surface of the driving gear 28. A toothed belt 29 is sleeved on the surface of the driving gear 28. A driven gear 30 is sleeved on the other end of the toothed belt 29. A plug-in rod 31 is inserted into the side of the driven gear 30. The plug-in rod 31 is embedded in the screw rod 23. There are two plug-in rods 31. The toothed belt 29 penetrates into the transmission box 21 from the notch 22. The drive shaft 27 is inserted into the base 36 as a whole.

[0097] Specifically, the decision system on the portable terminal adopts the "general platform + dedicated module (including dedicated interface adapter 2)" architecture, which can be used for on-site real-time maintenance of vehicles. It provides hardware system support through a general hardware platform, obtains information from the test equipment through an interface, and uses dedicated instrument module resources to collect signals and monitor the status of different subsystems of different large and complex technical vehicles, and performs signal testing and fault diagnosis through auxiliary decision-making system software, and finally realizes auxiliary decision-making for vehicle repair. Therefore, by connecting the interface adapter 2 to the signal transfer and physical docking area of ​​the vehicle, the fault information of the vehicle can be collected. Then, according to the working process of a certain vehicle, the general test is divided into five stages: power-on, use, monitoring, and fault handling, and the signal characteristics of each stage are tested and processed separately.

[0098] After completing the physical interface connection of the vehicle, directly start the motor 26 at the bottom, and drive the drive shaft 27 and the active gear 28 to rotate through the motor 26, and then drive the plug-in rod 31 to rotate with the help of the toothed belt 29 and the driven gear 30, so that the two plug-in rods 31 control the rotation of the screw rod 23 from the inside of the screw rod 23, and finally extend from the inside of the threaded sleeve 25 to the outside, and get stuck in the internal area of ​​the vehicle. After this point is fixed, even if the display unit 4 or the housing 32 of the portable maintenance auxiliary terminal 1 is hit and pulled during subsequent maintenance, the end of the data acquisition cable 34 can be fixed by the locking unit part at the bottom, thereby avoiding external collisions from interfering with the stability of the connection of the interface adapter 2.

[0099] In a second aspect, this embodiment further provides a decision-making assistance method using the above system, comprising the following steps:

[0100] Step 1. Establish a dynamic fault tree through qualitative and quantitative analysis methods: Qualitative analysis is to find out the combination of all fault causes that lead to a certain fault. For each fault in the case library, all fault causes of the fault are obtained by analyzing the working principle, fault mechanism, working process, and performance of the vehicle. For example, by analyzing a certain terminal, it is found that the causes of its screen edge key failure are the following four: line fault, screen edge key control board fault, screen edge key fault, and main control module fault. On the basis of qualitative analysis, quantitative analysis is performed on each fault cause, the importance is calculated, and the fault causes are sorted according to the importance to obtain a fault tree.

[0101] Quantitative analysis includes three aspects:

[0102] ① Probability of occurrence of fault causes: The probability of occurrence of each fault cause is calculated by the number of times each fault cause occurs in the fault case library, and changes in real time as the case library is updated.

[0103] ② Difficulty of troubleshooting: Determined by the difficulty of detecting and repairing the faulty part, and derived from case data analysis of the user unit.

[0104] ③ Vulnerability of the fault part: Based on the test data, the vulnerability data of the fault part is obtained.

[0105] Through experimental simulation and case verification, the weights of the above three aspects are calculated. Taking a terminal as an example, the data of quantitative analysis of the screen edge key failure is shown in Table 1.

[0106] Table 1 Fault quantitative analysis data table

[0107]

[0108] Step 2: For each fault cause in the fault tree, establish a troubleshooting guide and repair guide for the fault cause. After the fault tree is established, troubleshoot and repair each fault cause in turn according to the order in the fault tree. In order to increase the troubleshooting and repair speed and improve the maintenance capabilities of maintenance support personnel, establish a wizard-style troubleshooting and repair guide, and use the guide to guide troubleshooters to troubleshoot and repair the fault cause. For each fault cause in the fault tree, establish a troubleshooting guide and repair guide for the fault cause.

[0109] The troubleshooting guide provides steps and guidance for detecting and troubleshooting the fault location of the cause of the fault.

[0110] The repair guide provides the steps and guidance for repairing the fault cause and fault location.

[0111] According to the order of the fault tree, each fault cause is detected and checked according to the troubleshooting guide. If the part is indeed faulty, repair the faulty part according to the repair guide; otherwise, continue to check the next fault cause.

[0112] The troubleshooting guide and repair guide mainly include the following contents:

[0113] ① Personnel: staffing and division of labor for testing or repair;

[0114] ② Tools: tools involved in the operation;

[0115] ③Operation steps: detailed steps for troubleshooting, testing or repair;

[0116] ④ Video or picture guidance: Video or picture guidance on how to operate;

[0117] ⑤Precautions: List the precautions during operation;

[0118] Step 3: Sort the fault case information by fault name and organize it into a database for storage, management and reuse. A fault case is a collection of relevant information about a specific fault, including fault type, fault name, fault phenomenon, fault cause and repair method. Building a fault case library is to sort out a large amount of case knowledge, sort the fault case information by fault name, and organize it into a database for storage, management and reuse;

[0119] Step 4: Establish a fault diagnosis model. The process of maintenance personnel when troubleshooting: first determine the fault based on the fault phenomenon; then think about the several causes of the fault, and determine the troubleshooting order based on experience; then troubleshoot and repair each cause of the fault in order; finally, summarize the learning from this repair and transform it into experience. The fault diagnosis model provided in this embodiment simulates the way of thinking of maintenance personnel when analyzing and troubleshooting, and establishes corresponding modules to complete the diagnosis of faults - based on the signal acquisition and conversion of a certain vehicle, the signals of each device are quantized to form a one-dimensional feature vector representation, and all eigenvalues ​​are normalized to [0,1]. The decision-making support expert system forms a diagnostic strategy for real-time fault discovery through a machine learning dynamic fault case library, and provides troubleshooting and emergency repair guidelines;

[0120] Step 5: Access the portable maintenance auxiliary terminal 1 to perform fault diagnosis and obtain fault information;

[0121] Step 6: When the maintenance personnel locate a certain fault part and repair the fault, the troubleshooting record is automatically stored, the corresponding fault case is updated, and the probability importance of the fault cause in the fault tree model is also updated accordingly, so that the case library and fault tree model can be upgraded with the learning of troubleshooting experience. If the actual fault is not stored in the case library, or a new fault cause appears, the troubleshooting record is recorded, and the system maintenance personnel regularly update and maintain the case library, fault tree model, and troubleshooting guide library information.

[0122] In a third aspect, this embodiment also provides a method for diagnosing vehicle faults using a BP neural network, and the specific diagnosis process and algorithm content include:

[0123] First, initialize the network and initialize the values ​​of each variable according to the dimension of the parameters, the number of neurons in the hidden layer, and the dimension of the output layer. i ,ω ji , b ji , η, α; then, the network is forward propagated, the output of the previous layer is used as the input of the next layer, and the sigmoid function is used as the activation function, and finally the total error E of the network is obtained; then, the network is back-propagated, according to E and the gradient descent method, the weights and thresholds of the network are continuously iterated and updated. In addition, the impulse term is added, which can cross some narrow local minima and reach a smaller place; finally, it is determined whether the algorithm iteration is completed;

[0124] The auxiliary decision-making method also includes a naive Bayes algorithm model. For the fault diagnosis problem, according to the Bayesian theorem, the fault feature to be classified is set to d, and the fault belongs to category c. k The probability of is set to:

[0125]

[0126] Fault feature d contains n feature attributes, d = {w1, w2, ..., w n}, Naive Bayes assumes that these n feature attributes are independent of each other, so formula (3-2) can be expressed as:

[0127]

[0128] For any fault type, the denominator of the above formula is the same and does not need to be considered, so the parameters to be estimated are p(wi|ck) and p(ck). Then, these parameters can be estimated using the maximum likelihood method:

[0129]

[0130] Where N represents the number of training faults, y j Indicates the class label of the jth fault, x j Represents the features contained in the jth fault, and the final classification results are as follows:

[0131]

[0132] This auxiliary decision method separates the data by constructing a segmentation surface, and the decision boundary of SVM is the maximum margin hyperplane solved for the training samples;

[0133] This surface is also called the maximum margin hyperplane. The optimal separation line can accurately distinguish two different types of data. The calculation formula of the separation line is as follows:

[0134]

[0135] Where w and b are two constants, and the value of x is the horizontal coordinate value. (x i ,y i ) as a set of linearly separable sample points, i = 1, ..., n, y i is the sample label, with a value between -1 and +1;

[0136] The calculation formula for the dist gap between two different categories of data is as follows:

[0137]

[0138] The SVM model makes the support vectors of all categories as far away from the segmentation hyperplane as possible, which can be expressed by the following mathematical formula:

[0139]

[0140] st:y (i) (w T x (i)+b)≥1,i=1,2,…,m

[0141] The Lagrangian optimization method can be used to find the optimal interval hyperplane, transforming the optimization problem into a dual problem. The optimization problem is transformed into the following:

[0142]

[0143] Since Lagrangian has duality characteristics, the optimization objective can be converted into an equivalent dual problem to solve, so that the optimization objective becomes:

[0144] max β≥0 min w,b L(w,b,β)

[0145] Therefore, for this optimization function, it can correspond to finding the maximum value of the Lagrange multiplier β. For processing linearly inseparable data, a kernel function can be introduced to map low-dimensional data to a high-dimensional space for segmentation. The kernel function includes the following:

[0146] (1) Linear Kernel

[0147] K(x,z)=x·z

[0148] (2) Polynomial Kernel

[0149] K(x,z)=(γx·z+r) d

[0150] (3) Gaussian Kernel

[0151]

[0152] (4) Sigmoid Kernel

[0153] K(x,z)=tanhγx·z+r

[0154] Fourth, this embodiment also provides a method for predicting vehicle failures using RVM, and the specific algorithm flow is as follows:

[0155] Given a training sample set is the characteristic value in the sample set, and the target value t n Independent and identically distributed, with mean 0 and variance σ 2 Gaussian noise ε n ,but

[0156] t n =y(X n; w)+ε n Formula (1.1)

[0157] The model output of RVM can be defined as:

[0158]

[0159] Where N is the number of samples, and the weight vector w = [ω1,...,ω N ] T , φ is a matrix of order N×(N+1), and φ=[φ(x1),φ(x2),...,φ(x N )] T , where φ(x n )=[K(x n ,x1),K(x n ,x2),...,K(x n ,x N )] T , K(x,x i ) is a nonlinear function;

[0160] Since the target value t n Independent, the likelihood function of the training sample set is expressed as:

[0161]

[0162] Where, the target vector t=[t1,...,t N ] T

[0163] According to the structural risk minimization principle of support vector machine, if the weight w is not constrained and equation (1.2) is directly maximized, overfitting may occur. Therefore, each weight in RVM defines its own Gaussian prior probability distribution:

[0164]

[0165] In the formula, α=[α1,α2...,α N ] T is a hyperparameter, which determines the prior distribution of the weight w. Each hyperparameter α i Corresponding to a weight ω i ;

[0166] Given the prior probability distribution and likelihood distribution, the posterior probability distribution of the weights is calculated based on the Bayesian criterion as follows:

[0167]

[0168] In the formula, the mean μ and covariance ∑ of the posterior weights are expressed as:

[0169] μ=σ -2 ∑φ T t formula (1.6)

[0170] ∑=(σ -2 φ T φ+A) -1 Formula (1.7)

[0171] Among them, A=diag(α1,α2,...,α N )

[0172] The mean μ of the posterior distribution of the weights determines the estimate of the weights, and the covariance ∑ represents the uncertainty of the model prediction. Therefore, in order to obtain the weights of the model, it is first necessary to estimate the optimal value of the hyperparameters. The likelihood distribution of the hyperparameters, that is, the Gaussian distribution, is expressed as:

[0173]

[0174] In the formula, covariance C = σ 2 I+φA -1 φ T

[0175] Maximizing the likelihood distribution of hyperparameters yields α and σ 2 The most likely value is calculated by using the iterative estimation method instead of the analytical calculation method to calculate equation (1.7). 2 Take the derivative of equation (1.7) respectively, then set it to zero and rewrite the formula:

[0176]

[0177] In the formula, μ i is the mean of the ith posterior weights obtained by formula (1.6); i =1-α i ∑ ii ,∑ ii is the current α and σ 2 The i-th diagonal element of the posterior weight covariance matrix obtained by formula (1.7);

[0178] While repeatedly calculating equations (1.9) and (1.10), keep updating equations (1.6) and (1.7) until a certain convergence condition is met and the iteration is stopped. During the iteration, a large number of α tending to infinity appear. i According to formula (1.5), we can get p(w|t,α,σ 2 ) appears at the maximum peak as the zero point, so it is considered that its ω i is zero, thus producing sparse solutions, somewhat similar to the SV of SVM, those with non-zero ω iThe corresponding learning sample is the relevance vector (RV).

[0179] Based on the analysis of the above embodiments, the modeling steps of RVM are expressed as:

[0180] Step 1. Initialize parameter α i and σ 2 ;

[0181] Step 2. Calculate the weighted posterior statistics μ, ∑;

[0182] Step 3. Calculate all γ i , and re-estimate α i and σ 2 ;

[0183] Step 4. If converged, execute Step 5, otherwise re-execute Step 2.

[0184] Step 5. Find RVs and build the RVM structure.

[0185] The above shows and describes the basic principles, main features and advantages of the present application. For those skilled in the art, it is obvious that the present application is not limited to the details of the above-mentioned exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or basic features of the present application.

[0186] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A repair auxiliary decision-making method based on machine learning algorithm, characterized in that: The following steps are involved: Establish a dynamic fault tree through qualitative and quantitative analysis methods; For each fault cause in the fault tree, establish a troubleshooting guide and repair guide for the fault cause; Organize fault case information by fault name and form a database for storage, management and reuse; Establish a fault diagnosis model; access a portable maintenance auxiliary terminal to perform fault diagnosis; record the troubleshooting data, and regularly update and maintain the case library, fault tree model, and troubleshooting and repair guide library information; The decision method also uses RVM to predict vehicle failure: Given a training sample set is the characteristic value in the sample set, and the target value t n Independent and identically distributed, with mean 0 and variance σ 2 Gaussian noise ε n ,but t n = y(X n ; w) + ε n Equation (1.1) The model output of RVM can be defined as: Where N is the number of samples, and the weight vector w = [ω1,...,ω N ] T , φ is a matrix of order N×(N+1), and φ=[φ(x1),φ(x2),...,φ(x N )] T , where φ(x n )=[K(x n ,x1),K(x n ,x2),...,K(x n ,x N )] T , K(x,x i ) is a nonlinear function; Since the target value t n Independent, the likelihood function of the training sample set is expressed as: Where, the target vector t=[t1,...,t N ] T According to the structural risk minimization principle of support vector machine, if the weight w is not constrained and equation (1.2) is directly maximized, overfitting may occur. Therefore, each weight in RVM defines its own Gaussian prior probability distribution: In the formula, α=[α1,α2...,α N ] T is a hyperparameter, which determines the prior distribution of the weight w. Each hyperparameter α i Corresponding to a weight ω i ; Given the prior probability distribution and likelihood distribution, the posterior probability distribution of the weights is calculated based on the Bayesian criterion as follows: In the formula, the mean μ and covariance ∑ of the posterior weights are expressed as: μ = σ -2 ∑φ T t Equation (1.6) ∑=(σ -2 f T φ+A) -1 formula(1.7) Where, A=diag(α1,α2,...,α N ) The mean μ of the posterior distribution of the weights determines the estimate of the weights, and the covariance ∑ represents the uncertainty of the model prediction. In order to obtain the weights of the model, the optimal value of the hyperparameter is first estimated. The likelihood distribution of the hyperparameter, that is, the Gaussian distribution, is expressed as: In the formula, covariance C = σ 2 I+φA -1 φ T Maximizing the likelihood distribution of hyperparameters yields α and σ 2 The most likely value is calculated by using the iterative estimation method instead of the analytical calculation method to calculate equation (1.7). 2 Derivative (1.7) is taken separately, then set it equal to zero and rewrite the formula as: In the formula, μ i is the mean of the ith posterior weights obtained by formula (1.6); i =1-α i ∑ ii ,∑ ii is the current α and σ 2 The i-th diagonal element of the posterior weight covariance matrix obtained by formula (1.7); While repeatedly calculating equations (1.9) and (1.10), keep updating equations (1.6) and (1.7) until convergence and stop iteration. During iteration, a large number of α tending to infinity appear. i According to formula (1.5), we can get p(w|t,α,σ 2 ) appears at the maximum peak as the zero point, and the non-zero ω i The corresponding learning sample is the relevant vector.

2. The method for assisting decision-making in emergency repair based on a machine learning algorithm according to claim 1, characterized in that: The qualitative analysis is used to find out the combination of all fault causes that lead to the fault, and all fault causes of the fault are obtained by analyzing the working principle, fault mechanism, working process, and performance of the vehicle; the quantitative analysis includes the probability of occurrence of the fault cause, the difficulty of troubleshooting, and the vulnerability of the faulty part; the troubleshooting guide provides steps and guidance methods for detecting and troubleshooting the faulty part of the fault cause, and the repair guide is used to provide steps and guidance methods for repairing the faulty part of the fault cause; by establishing a fault diagnosis model, the signals of each device are quantized on the basis of vehicle signal acquisition and conversion to form a one-dimensional feature vector representation, and all eigenvalues ​​are normalized to between [0,1]. The system forms a diagnostic strategy for real-time fault discovery through a machine learning dynamic fault case library, and finally provides a troubleshooting and emergency repair guide.

3. The method for emergency repair decision support based on machine learning algorithm according to claim 1, characterized in that: The auxiliary decision-making method uses BP neural network to diagnose vehicle faults. The diagnosis process includes: First, initialize the network and initialize the values ​​of each variable according to the dimension of the parameters, the number of neurons in the hidden layer, and the dimension of the output layer. i ,ω ji 、b ji , η, α; then, the network is forward propagated, the output of the previous layer is used as the input of the next layer, and the sigmoid function is used as the activation function, and finally the total error E of the network is obtained; then, the network is back-propagated, according to E and the gradient descent method, the weights and thresholds of the network are continuously iterated and updated. In addition, the impulse term is added, which can cross some narrow local minima and reach a smaller place; finally, it is determined whether the algorithm iteration is completed; The auxiliary decision-making method also includes a naive Bayes algorithm model. For the fault diagnosis problem, according to the Bayesian theorem, the fault feature to be classified is set to d, and the fault belongs to category c. k The probability of is set to: Fault feature d contains n feature attributes, d = {w1, w2, ..., w n }, Naive Bayes assumes that these n feature attributes are independent of each other, so formula (3-2) can be expressed as: The parameters to be estimated are p(wi|ck) and p(ck), and these parameters are estimated using the maximum likelihood method: Where N represents the number of training faults, y j Indicates the class label of the jth fault, x j Represents the features contained in the jth fault, and the final classification results are as follows:

4. The emergency repair decision-making auxiliary method based on machine learning algorithm according to claim 1, characterized in that: This auxiliary decision method separates the data by constructing a segmentation surface, and the decision boundary of SVM is the maximum margin hyperplane solved for the training samples; The optimal separation line of the maximum margin hyperplane can accurately distinguish two different types of data. The calculation formula of the separation line is as follows: Where w and b are two constants, and the value of x is the horizontal coordinate value. (x i ,y i ) as a set of linearly separable sample points, i = 1, ..., n, y i is the sample label, with a value between -1 and +1; The calculation formula for the dist gap between two different categories of data is as follows: The SVM model is expressed mathematically as follows: The Lagrangian optimization method is used to find the optimal interval hyperplane, transforming the optimization problem into a dual problem. The optimization problem is transformed into the following: The optimization goal of the Lagrangian transformation optimization problem is to solve the equivalent dual problem, and the optimization goal becomes as follows: max β≥0 minutes w,b L(w,b,β) The optimization function corresponds to finding the maximum value of the Lagrange multiplier β. For processing linearly inseparable data, a kernel function can be introduced to map low-dimensional data to a high-dimensional space for segmentation. The kernel function is a linear kernel function, a polynomial kernel function, a Gaussian kernel function or a Sigmoid kernel function.

5. A decision-making system using the decision-making method according to claim 1, characterized in that: The auxiliary decision system includes a portable maintenance auxiliary terminal, the surface of the portable maintenance auxiliary terminal is connected to a dedicated interface adapter through a data acquisition cable, the surface of the portable maintenance auxiliary terminal is also connected to a display unit through a data transmission cable, the surface of the display unit is embedded with a display screen, and a locking mechanism is installed at the bottom of the portable maintenance auxiliary terminal. The portable maintenance auxiliary terminal performs signal transfer and physical docking with the vehicle under test through an interface adapter; the system is also equipped with a fault diagnosis module and a multi-task stream data buffer module, the fault diagnosis module uses RVM to predict vehicle faults; the multi-task stream data buffer module uses multiple tasks simultaneously, and the tasks are divided into conventional data acquisition tasks, video data acquisition tasks, message parsing tasks and processing output tasks.

6. The system according to claim 5, characterized in that: A back panel is screwed onto the back of the display unit, a connecting rod is inserted into the surface of the back panel, an end panel is integrally formed at one end of the back panel, a rotating shaft is inserted into the inner side of the end panel, a supporting shaft is inserted into the other end of the back panel, a supporting rod is installed on the surface of the supporting shaft, and a plug-in column is inserted into the end of the supporting rod.

7. The system according to claim 6, characterized in that: A second suction cup is installed at the end of the plug-in column, and a slot is provided at the top of the second suction cup. A rotating sleeve is provided on the top of the support rod, and the rotating sleeve is used to be sleeved on the surface of the support shaft, and the support rod as a whole rotates and folds around the support shaft; a fixed sleeve is integrally formed on the surface of the rotating shaft, and a first suction cup is installed on the side of the fixed sleeve, and the number of the first suction cup and the second suction cup are both two.

8. The system according to claim 5, characterized in that: The surface of the portable maintenance auxiliary terminal is provided with a shell, the bottom of the shell is screwed with a base, the side of the shell is integrally formed with a side baffle, the inside of the shell is provided with an interlayer, and the top of the shell is provided with a storage groove.

9. The system according to claim 8, characterized in that: A front baffle is welded on the top of the base, and a latch is inserted on the surface of the front baffle. The latch passes through the front baffle and is used to be embedded in the slot. The interface adapter is inserted into the storage slot, and the display unit is placed as a whole in the interlayer.

10. The system according to claim 7, characterized in that: The locking mechanism includes a transmission box, a screw rod and a motor. A notch is provided in the middle of the transmission box. Threaded sleeves are provided at both ends of the transmission box. A screw rod is inserted into the interior of the threaded sleeve. An extrusion plate is integrally formed at the end of the screw rod. A driving shaft is inserted into the output end of the motor. A driving gear is keyed on the surface of the driving shaft. A toothed belt is sleeved on the surface of the driving gear. A driven gear is sleeved on the other end of the toothed belt. A plug-in rod is inserted on the side of the driven gear. The plug-in rod is embedded in the interior of the screw rod. There are two plug-in rods. The toothed belt penetrates into the interior of the transmission box from the notch. The driving shaft is inserted into the interior of the base as a whole.

Images