CN114821856B - Intelligent auxiliary device connected in parallel to automobile traveling computer for rapid automobile maintenance - Google Patents

Abstract

The invention belongs to the field of research and development of automobile maintenance equipment, and discloses an intelligent auxiliary device for rapid automobile maintenance connected in parallel with a driving computer, which can solve the problems of "unable to accurately find specific faulty parts" and "failure diagnosis failure due to inability to connect to the Internet" in manual maintenance. The steps are as follows: through the installation of quick plug-in sensors, combined with the original sensors and on-board computer, the vehicle operation data information is obtained, and transmitted to the intelligent fault diagnosis embedded system with the help of CAN bus technology, and the fault diagnosis is completed with the help of the API model downloaded from the cloud . The API model training is completed on the cloud server, and the BP neural network with the weight threshold optimized by the genetic algorithm is selected. The initial training data comes from the historical database of vehicle operation information constructed under laboratory conditions. Through user feedback, the database is updated. The device is mainly used for fault status identification and fault classification and location in automobile operation.

Description

An intelligent auxiliary device for automobile rapid maintenance connected in parallel with a driving computer

Technical Field

The present invention belongs to the field of automobile maintenance equipment research and development, and in particular relates to an automobile rapid maintenance intelligent auxiliary device which includes an automobile external quick-plug sensor, a cloud service neural network model training, and an intelligent fault diagnosis embedded system connected in parallel with an on-board computer and has the functions of fault state recognition and fault classification and positioning.

Background Art

With the development of the automobile manufacturing industry and the increasing demand for automobile functions by users, the total number of automobiles continues to increase, and the structures of automobile components are becoming increasingly complex. The complex structure leads to an increasing number of types of automobile failures, and the difficulty of identifying and locating automobile failures is also increasing. If there is a lack of effective fault diagnosis methods, the automobile maintenance process will be complicated, the user's waiting time will increase, and the cost of automobile maintenance will also increase. Therefore, how to quickly identify and locate automobile failures has become an issue worthy of attention.

In the existing commonly used local fault diagnosis methods, an automobile fault diagnostic instrument is used to read the fault code in the on-board computer memory to help maintenance personnel find the cause of the vehicle fault. However, the automobile fault diagnostic instrument only completes the interpretation of the fault code and points out the system where the error is located (for example: power module-battery management system fault), but it cannot accurately identify the specific faulty component. Maintenance personnel still need to make further judgments based on experience to find the specific fault point. In the remote fault diagnosis method under the background of the popularization of the Internet of Vehicles, the vehicle-mounted terminal sends the fault signal to the cloud through wireless transmission, and then the detection technology stored in the cloud performs detection and feedback. However, since the Internet of Vehicles can only process the feedback information of traditional on-board sensors and lacks key targeted sensor information, the Internet of Vehicles diagnosis method has defects such as insufficient innate key diagnostic clue information.

In summary, existing equipment diagnostic technology lacks the ability to accurately locate faulty components. At the same time, Internet of Vehicles diagnostic technology cannot deploy sensors in a targeted manner to obtain key diagnostic clues information, and the diagnostic efficiency and level of intelligent diagnosis need to be improved.

Summary of the invention

The purpose of the present invention is to design an intelligent auxiliary device for automobile rapid maintenance in parallel with the driving computer in view of the industry problem that the existing equipment diagnosis technology lacks the ability to accurately locate faulty parts in the field of vehicle information diagnosis, and the vehicle network diagnosis technology cannot deploy sensors in a targeted manner to obtain key diagnostic clue information, and the diagnostic efficiency and intelligent diagnosis level need to be improved. First, based on a large amount of proprietary experimental data, the cutting-edge neural network technology is used to conduct preliminary training on various common faults of the automobile system; then, the trained model is built into the embedded system, so that the embedded system device can identify various common faults of various automobile systems; then, a device that can be quickly installed and disassembled by simply connecting in parallel with the on-board computer is designed to effectively read the sensor information in the on-board system and the alarm information of the on-board computer; finally, a variety of sensors that can be quickly deployed in key parts of the automobile and quickly plugged in and out of the device embedded computer are designed to realize the direct and rapid collection of key fault diagnosis clue information of the vehicle; an intelligent auxiliary device for automobile maintenance with offline accurate information collection is comprehensively constructed to realize the function of quickly identifying the fault state and classifying and locating the fault of the accident vehicle for the maintenance engineer. This invention has the advantages of low cost, high precision, high efficiency and strong applicability.

An intelligent auxiliary device for automobile rapid maintenance connected in parallel to a driving computer, including a quick plug-in sensor, an intelligent fault diagnosis embedded system and a cloud platform;

The quick-swap sensor and intelligent fault diagnosis embedded system are quickly plugged in parallel with the original on-board sensors and on-board computers of the car; the modular quick-swap sensor is used to assist in measuring the vehicle status data; the intelligent fault diagnosis embedded system is used to receive the quick-swap sensor signal, the original on-board sensor signal of the car and the original accident alarm information of the on-board computer, and identify the vehicle fault status and classify and locate the fault according to the received information;

The cloud platform includes cloud storage and cloud computing. The cloud storage part builds a historical database of vehicle operation information, which includes original vehicle sensor information, quick-swap sensor information, vehicle computer accident alarm information, and corresponding fault status and fault classification and location information. The cloud computing part uses the cloud storage vehicle operation information historical database to train the fault status identification and fault location core API model.

The cloud storage part updates the database based on user diagnostic feedback;

When training the core API model for fault state identification and fault location, the cloud computing part uses genetic algorithms to optimize BP neural network parameters;

After the cloud storage part updates the database based on the user's diagnostic feedback, the cloud computing part retrains the fault status identification and fault location core API model, and downloads and updates the trained model to the intelligent fault diagnosis embedded system in an online state; the fault status identification and fault location core API model trained on the cloud is downloaded to the vehicle-mounted intelligent fault diagnosis embedded system for vehicle fault status judgment and fault classification and location;

The original vehicle sensor information, quick-swap sensor information, and vehicle computer accident alarm information are transmitted to the intelligent fault diagnosis embedded system through the CAN bus. The intelligent fault diagnosis embedded system performs data preprocessing operations on the information in the database to obtain the input data of the fault state recognition and fault location core API model;

The core API model for fault status identification and fault location first trains the constructed vehicle operation information history database in the Alibaba Cloud server, and establishes a model that uses vehicle operation information data to identify fault states and classify faults, that is, to obtain a mapping from vehicle operation information to specific faults.

In the training of the fault state identification and fault location core API model, it is first necessary to extract training samples from the vehicle operation information history database, and divide the extracted data into a training set, a test set, and a validation set;

For the extracted data, data preprocessing is required before training. The specific steps are as follows:

Step 1: Delete the samples corresponding to the duplicate and missing data from the database, and re-extract new data from the database as supplementary training samples;

Step 2: Considering the existence of data noise, it is necessary to use means including sliding filtering to remove data noise;

Step 3: Use principal component analysis to select multiple feature data;

Step 4: Perform min-max normalization on the feature data selected by the principal component analysis method, complete the random extraction of the training samples, and pre-process the randomly extracted data.

The preprocessed data is used to train the model for fault state recognition and fault classification and location, using a BP neural network combined with a genetic algorithm. The specific steps are as follows:

Step 1: Use the preprocessed data to build multiple BP neural network models. The activation function of the hidden layer is the tanh function, and the activation function of the output layer is the softmax function.

Step 2: Initialize the neural network connection weights and thresholds, and use genetic algorithms to optimize the weights and thresholds;

Step 3: Perform real number coding on the weights and thresholds of each established BP neural network diagnostic model, and randomly select 100 initial individuals with weights and thresholds corresponding to the real number coding to form an initial population;

Step 4: Calculate the loss function and express it as the sum of squared errors; take the inverse of the loss function as the individual fitness function;

Step 5: Perform selection, crossover, and mutation operations on individuals in the current population to form a new population for the next generation;

Step 6: Determine whether the new population obtained in step 5 meets the convergence condition. If so, complete the weight and threshold optimization; if not, return to step 5 and recalculate;

Step 7: Use the data of the best individual in the population as the initial weight and threshold of the optimized BP neural network model, and start iterative training of the BP neural network model until the loss function value is less than the preset threshold, or the number of iterations is reached, and the BP neural network model training is completed;

Step 8: Input the validation set into multiple trained BP neural network models, and select the neural network model with the best performance as the core API model for fault state identification and fault location;

Step 9: Get the accuracy of the API model through the test set.

During the database content update process, data is updated according to the following principles:

(1) When the diagnosis result of the intelligent fault diagnosis embedded system is correct and the data of the same fault state identification and fault classification and location in the database does not reach the capacity value, the vehicle operation information data and the fault state identification and fault classification and location data are directly updated to the database;

(2) When the fault diagnosis result of the intelligent fault diagnosis embedded system is correct and the data corresponding to the fault state identification and fault classification location in the database has reached the capacity value, the database is not updated;

(3) When the fault diagnosis result of the intelligent fault diagnosis embedded system is wrong and the data corresponding to the fault state identification and fault classification and location in the database does not reach the capacity value, the vehicle operation information data and the fault state identification and fault classification and location data are directly updated to the database;

(4) When the fault diagnosis result of the intelligent fault diagnosis embedded system is erroneous and the data corresponding to the fault state identification and fault classification and location in the database has reached the capacity value, the vehicle operation information data during diagnosis and the fault state identification and fault classification and location data are used to randomly replace a group of data corresponding to the same fault state identification and fault classification and location in the database.

Cloud data message queue is used for buffering: In order to ensure the matching of information transmission of intelligent fault diagnosis embedded system and cloud database update process, an information queue is added as data buffer in the middle to reduce the access pressure of cloud server; the cloud platform will receive operation information sent by multiple intelligent fault diagnosis embedded systems at the same time, which will be pushed to the message queue after passing through the data storage module; the message queue is designed as a circular queue data structure, and the queue stores vehicle operation history information in first-in-first-out order; a resident process is started to monitor the data storage of the message queue in real time. Once new data information is found in the queue, the data will be taken out for database update, and then the processed information will be deleted from the queue.

The beneficial effects of the present invention are as follows: by adopting quick-plug sensors, different sensors can be selectively used for different models of cars, and data collection is highly flexible and versatile; by combining the car's own sensors, on-board computers and other components, fault status identification and fault classification and positioning can be made more accurate and comprehensive; by applying cloud training methods to build a core API model, the cost of purchasing servers is saved, and the cloud training method also makes the update of algorithm models more convenient; for the embedded system on the vehicle side, models with fault status identification and fault classification and positioning functions are downloaded from the cloud, which avoids the phenomenon that fault diagnosis cannot be performed when the network cannot be connected, and at the same time, the embedded system does not need to perform model training, saving computing power costs; by completing the update of the database through the user's later feedback, the problem of overfitting of the model to the original fault and the problem of classification and positioning of the model to new faults are solved to a certain extent, and the real-time and accuracy of the model are improved.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a general structural diagram of the present invention;

FIG2 is a core API model training method for fault state identification and fault location according to the present invention;

FIG3 is a database updating method applicable to all vehicles of the present invention;

FIG4 is a flow chart of the present invention showing the fault diagnosis API model training process using the BP cloud neural network platform.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with specific embodiments. It should be understood that the embodiments are only used to illustrate the content of the present invention, rather than to limit the present invention.

Embodiment 1:

This embodiment provides a highly reliable intelligent auxiliary device for automobile rapid repair, which is connected in parallel to a vehicle computer and has the functions of fault state recognition and fault classification and location.

Before conducting fault diagnosis and analysis, it is necessary to obtain the original data of the vehicle operation: the collection of the original data of the vehicle operation is completed through components such as quick plug-in sensors, original on-board sensors, on-board computers, data acquisition cards, and signal conditioners. The original data of the vehicle operation is collected by plug-in sensors, original on-board sensors, and on-board computers, and converted into voltage signals by combining the gain parameters of the programmable amplifier. Then, the data acquisition card is used to convert the output continuous signal into a discrete time series signal and output a discrete voltage value.

The hardware equipment used for data collection is:

(1) Quickly plug and unplug sensors and existing vehicle sensors.

Corresponding sensors are installed to prevent possible faults in the fuel supply system, cooling system, starting system, ignition system, lubrication system and other components. For example, the sensors that measure signals in the fuel supply system include: Fuel tank level sensor: detects whether there is too little fuel in the fuel tank or the oil level is lower than the lower end of the upper oil pipe hole to determine whether the fuel quantity is sufficient; Upper oil pipe vibration signal sensor: bundles the vibration signal sensor around the upper oil pipe. If the upper oil pipe is desoldered, cracked, broken or the oil pipe joint is loose, an abnormal vibration signal will be detected; Fuel pipeline liquid pressure gauge: placed on the pipeline where the gasoline filter is located. If there is blockage, the fuel pressure will be abnormal.

(2)Signal conditioner.

(3) Data acquisition card.

As shown in FIG2 , the embodiment of the present invention is a highly reliable intelligent auxiliary device for automobile rapid repair which is connected in parallel to the on-board computer and has the functions of fault status identification and fault classification and positioning. It first needs to use the vehicle operation information data, fault status identification and fault classification and positioning data measured under laboratory conditions to form a vehicle operation information history database.

As shown in Figure 2, data is randomly extracted from the vehicle operation information history database according to the ratio of 70%, 15%, and 15% to obtain a training set, a test set, and a validation set. In the process of randomly extracting data, it is necessary to extract the fault state parameter data corresponding to each category of fault.

For the data extracted above, it is necessary to first check and delete the duplicate data and missing data from the sample. At the same time, the information is also reported to the database, and the duplicate data and missing data are deleted from the database. At the same time, in order to ensure that the total amount of the extracted samples remains unchanged, it is necessary to re-extract the corresponding number of samples from the database as a supplement; then perform duplicate data and missing data checks, and repeat the operation in sequence until all the data in the sample are qualified.

For the sample data after the duplicate data and missing data inspection, the data is filtered considering the existence of data noise.

For the filtered data, feature extraction is performed; considering that there are too many extracted data features and there may be high redundancy, principal component analysis is used to select multiple feature data and perform data dimensionality reduction; at this point, the random extraction of training samples and the preprocessing steps of the randomly extracted data are completed.

After completing the data preprocessing, as shown in FIG4 , the BP neural network model is trained in combination with the genetic algorithm.

(1) Construct multiple BP neural network models. Each BP neural network contains n input neurons, d hidden layer neurons (the number of hidden layer neurons in different BP neural network models is different), and m output neurons. The activation function of the hidden layer is the tanh activation function, and the activation function of the output layer is the softmax function.

(2) First, the genetic algorithm is used to optimize the parameters of the BP neural network model. The weights and thresholds of each BP neural network diagnostic model established in step 1 are encoded with real numbers, and the encoding length, i.e., the chromosome length of the individual, is S. Then, a population with a size of 100 individuals is randomly generated as 100 random solutions. That is, 100 initial individuals with weights and thresholds corresponding to the real number encoding are randomly selected to form the initial population. Each initial individual represents an initial solution for finding the optimal initial weight and initial threshold.

(3) Calculate the fitness of each individual in the population (initial population) described in step 2. First, calculate the loss function, expressed as the sum of squared errors J(i), and the formula is:

Where i=1,...., N is the number of chromosomes, m is the number of output layer nodes, k is the number of training samples, _Cm represents the actual value of the mth output node, and _ym represents the predicted value of the mth output node;

(4) Calculate the fitness of the individual and take the inverse of the loss function as the individual’s fitness function F(i):

(5) Perform selection, crossover, and mutation operations on individuals in the current population (the population generated in step 2 or the population returned in step 6) to form a new population for the next generation. The higher the fitness of an individual, the greater the probability of being selected. The probability of each individual being selected P(i) is:

For the crossover probability, take 0.4 during training; for the mutation probability, take 0.1 during training.

(6) Determine whether the new population obtained in step 5 meets the convergence condition. If so, complete the genetic algorithm optimization; otherwise, return to step 5 and recalculate.

(7) The optimal individual data in the population that meets the convergence conditions is used as the initial weights and thresholds of the corresponding BP neural network model. The constructed multiple BP neural networks all complete the optimization of the initial weights and thresholds through the genetic algorithm, and then the BP neural network model is trained.

(8) Iteratively training multiple BP neural network models, exiting the neural network training when the loss function is less than a preset threshold or when the target number of iterations is reached, to obtain multiple neural network models;

(9) After obtaining multiple neural network models, in order to verify the performance of the models, each model is tested using a test set, and the model with the best performance on the verification set is selected. The test set is then input into the best performing model to gain a preliminary understanding of the generalization ability of the model.

Download the best performing neural network model mentioned above to the intelligent fault diagnosis embedded system, connect the quick-swap sensors in parallel with the original vehicle-mounted sensors and vehicle-mounted computers, input the vehicle operation information data and the original accident alarm information data into the embedded system for preprocessing (extract the same features used in the initial model building process), and then use the fault state recognition and fault location core API model in the embedded system to perform fault state recognition and fault classification and location.

After fault status identification and fault classification and location are completed, the user will provide feedback on the correctness of the diagnosis results and the correct fault category after actual maintenance; the feedback results will be reviewed by diagnostic experts to ensure the validity of the data.

The database update principle is shown in FIG3 . With the continuous updating of the database, the original laboratory collected data is replaced by the actual fault state recognition and fault classification data, and the amount of data in the database is also increased to the rated capacity. The fault recognition accuracy of the fault state recognition and fault location core API model obtained by training is improved. The automobile rapid maintenance intelligent auxiliary device of the present invention is used for the fault diagnosis of various automobiles. With the increase of data collection in the process of fault diagnosis of various automobiles, the applicability of the fault state recognition and fault location core API model obtained by training is gradually enhanced.

During the database update process, when the cumulative diagnostic errors of a certain fault type reach the set threshold, the BP neural network model is trained using the updated database to re-obtain the BP neural network model for fault state identification and fault classification and location, and this model is updated to the vehicle's intelligent fault diagnosis embedded system.

Table 1. Fault location and classification function table of the present invention

Claims (2)

1. An intelligent auxiliary device for fast maintenance of an automobile, which is connected in parallel with a traveling computer, is characterized by comprising a fast plugging sensor, an intelligent fault diagnosis embedded system and a cloud platform;

the quick plug-in sensor and the intelligent fault diagnosis embedded system are connected with the original vehicle-mounted sensor and the original vehicle-mounted computer of the automobile in parallel in a quick plug-in mode; the modularized quick plug-pull sensor is used for carrying out auxiliary measurement on vehicle state data; the intelligent fault diagnosis embedded system is used for receiving a quick plug-pull sensor signal, an original vehicle-mounted sensor signal of an automobile and original accident alarm information of a vehicle-mounted computer, and carrying out automobile fault state identification and fault classification positioning according to the received information;

the cloud platform comprises cloud storage and cloud computing; the cloud storage part constructs a vehicle operation information historical database which comprises original vehicle-mounted sensor information, rapid plugging and unplugging sensor information, vehicle-mounted computer accident alarm information and corresponding fault state and fault classification and positioning information; the cloud computing part performs fault state recognition and fault positioning core API model training by using a cloud-stored vehicle operation information historical database;

the cloud storage part updates the database according to the user diagnosis result feedback;

the cloud computing part is combined with a genetic algorithm to carry out BP neural network parameter optimization when fault state identification and fault positioning core API model training are carried out;

the cloud computing part updates the database according to the user diagnosis result feedback in the cloud storage part, then carries out fault state recognition and fault positioning core API model training again, and downloads and updates the trained model to the intelligent fault diagnosis embedded system in a networking state; downloading the fault state recognition and fault positioning core API model after cloud training to an intelligent fault diagnosis embedded system of a vehicle-mounted end for judging the fault state of the vehicle and classifying and positioning the fault;

original vehicle-mounted sensor information, rapid plugging sensor information and vehicle-mounted computer accident alarm information are transmitted to an intelligent fault diagnosis embedded system through a CAN bus, and the intelligent fault diagnosis embedded system performs data preprocessing operation on the information in a database to obtain input data of a fault state identification and fault positioning core API model;

the method comprises the steps that a fault state identification and fault location core API model firstly trains a constructed vehicle operation information historical database in an Alice cloud server, and establishes a model for carrying out fault state identification and fault classification location by using vehicle operation information data, namely mapping from vehicle operation information to definite faults is obtained;

in the fault state identification and fault location core API model training, firstly, training samples need to be extracted from a vehicle operation information historical database, and extracted data are divided into a training set, a test set and a verification set;

for extracted data, before training, data preprocessing is firstly required, and the method comprises the following specific steps:

step 1: deleting samples corresponding to the detected data repetition and data deletion from the database, and re-extracting new data from the database to supplement the training samples;

step 2: in consideration of the existence of data noise, the data noise needs to be removed by adopting means including sliding filtering;

and step 3: selecting various characteristic data by adopting principal component analysis;

and 4, step 4: performing min-max normalization on the characteristic data selected by adopting a principal component analysis method, finishing random extraction of the training sample, and preprocessing the data after random extraction;

using the preprocessed data to train the model for fault state recognition and fault classification positioning, and adopting a BP neural network and a genetic algorithm, wherein the specific steps are as follows:

step 1: constructing a plurality of BP neural network models by using the preprocessed data, wherein the activation function of the hidden layer is a tanh function, and the activation function of the output layer adopts a softmax function;

step 2: initializing a neural network connection weight and a threshold, and optimizing the weight and the threshold by adopting a genetic algorithm;

and step 3: carrying out real number coding on the weight and the threshold of each BP neural network diagnosis model obtained by building, and randomly selecting 100 initial individuals of the weight and the threshold corresponding to the real number coding to form an initial population;

and 4, step 4: calculating a loss function expressed by the sum of squared errors; taking the reciprocal of the loss function as an individual fitness function;

and 5: selecting, crossing and mutating individuals in the current population to form a new population of the next generation;

and 6: judging whether the new population obtained in the step 5 reaches a convergence condition, and finishing weight and threshold optimization if the new population reaches the convergence condition; if the convergence condition is not reached, returning to the step 5 for recalculation;

and 7: taking the data of the optimal individuals in the population as the initial weight and the threshold of the optimized BP neural network model, starting iterative training on the BP neural network model until the loss function value is smaller than the preset threshold or the number of iterations is reached, and finishing the training of the BP neural network model;

and 8: inputting the verification set into a plurality of trained BP neural network models, and selecting the neural network model with the best performance as a fault state identification and fault positioning core API model;

and step 9: obtaining the accuracy of the API model through a test set;

in the updating process of the database content, data updating is carried out according to the following principles:

(1) When the diagnosis result of the intelligent fault diagnosis embedded system is correct and the data of the same fault state identification and fault classification positioning in the database does not reach the capacity value, directly updating the vehicle operation information data, the fault state identification and fault classification positioning data into the database;

(2) When the intelligent fault diagnosis embedded system has correct fault diagnosis results and the data corresponding to fault state identification and fault classification positioning in the database reaches a capacity value, the database is not updated;

(3) When the intelligent fault diagnosis embedded system has wrong fault diagnosis results and data corresponding to fault state identification and fault classification positioning in the database does not reach a capacity value, directly updating vehicle operation information data, fault state identification and fault classification positioning data to the database;

(4) When the intelligent fault diagnosis embedded system has wrong fault diagnosis results and the data corresponding to fault state identification and fault classification positioning in the database reaches a capacity value, the vehicle operation information data and the fault state identification and fault classification positioning data during diagnosis are used for randomly replacing a group of data corresponding to the same fault state identification and fault classification positioning in the database.

2. The intelligent auxiliary device for automobile rapid maintenance according to claim 1, wherein the cloud data message queue is adopted for buffering: in order to ensure that the information transmission of the intelligent fault diagnosis embedded system is matched with the updating process of the cloud database, an information queue is additionally arranged in the middle to serve as a data buffer area, so that the access pressure of a cloud server is reduced; the cloud platform receives operation information simultaneously sent by the intelligent fault diagnosis embedded systems, and the operation information is pushed to a message queue through the data storage module; the message queue is designed into a circular queue data structure, and the queue stores vehicle running history information according to a first-in first-out sequence; starting a resident process, monitoring the data storage condition of the message queue in real time, taking out the data for updating the database once finding that new data information arrives in the queue, and then deleting the processed information in the queue.

Images