Data processing method and device for automobile ECU repair instrument

Abstract

The present invention provides a data processing method and device for an automobile ECU repair instrument. By acquiring ECU repair instrument data, the ECU repair instrument data is passed into a fault detection algorithm for fault detection to obtain a fault detection result. The fault detection algorithm is obtained by optimizing a preset AI algorithm in combination with a first algorithm cost and a second algorithm cost. The first algorithm cost is at least one of a data implicit representation cost and a defect detection cost. The data implicit representation cost represents the degree of distinction between the fault characterization information corresponding to the same vehicle fault in the sample implicit representation in the ECU repair instrument data sample library, and the defect detection cost represents the degree of distinction between the fault classification prediction information corresponding to the same vehicle fault in the sample classification prediction information in the ECU repair instrument data sample library. The present invention can ensure the consistency of fault diagnosis for data processing of the same vehicle fault type, thereby improving the fault detection accuracy of the fault detection algorithm and saving algorithm costs.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a data processing method and device for an automobile ECU repair instrument. Background Art

[0002] As automotive technology continues to develop, automotive electronic control units (ECUs) have become an integral part of the normal operation of vehicles, and their complexity is increasing. Therefore, accurate detection of ECU failures is critical to the maintenance, safety, and performance of vehicles.

[0003] Artificial intelligence (AI) technology is applied to the field of ECU fault detection to improve the accuracy of detection. However, existing AI-based fault detection algorithms face many challenges when processing ECU repair instrument data. When processing ECU repair instrument data of the same vehicle fault type, these algorithms have difficulty ensuring the consistency of fault diagnosis. For example, for the fault type of engine misfire, different data samples may give different diagnostic results due to differences in implicit data representation or classification estimates. This is because the existing algorithms do not fully consider some key factors in ECU repair instrument data during construction and training.

[0004] From a data perspective, existing algorithms do not handle the distinctiveness of fault characterization information and fault classification prediction information of the same fault in data samples very well. Taking data implicit representation as an example, for different data samples of the same fault, the algorithm may not be able to accurately capture its inherently consistent fault characteristics, resulting in differences in diagnostic results. In terms of algorithm training optimization, existing AI algorithms usually require large-scale modifications to the algorithm composition structure in order to improve accuracy. This not only consumes a lot of manpower and material resources, increases the cost of algorithm development, but also easily causes new problems. For example, large-scale modifications may cause the algorithm to overfit the training data, and its adaptability to new data samples in actual applications will become poor; at the same time, the complex algorithm structure will reduce computing efficiency and prolong the time of fault detection, which is very unfavorable in actual automobile maintenance scenarios and cannot meet the automotive industry's demand for efficient and accurate ECU fault detection. Summary of the invention

[0005] In view of this, the present application provides a data processing method and device for an automobile ECU repair instrument.

[0006] The technical solution of this application is implemented as follows:

[0007] On the one hand, the present application provides a method for processing automobile ECU repair instrument data, the method comprising: obtaining ECU repair instrument data; passing the ECU repair instrument data into a fault detection algorithm for fault detection, and obtaining a fault detection result corresponding to the ECU repair instrument data; wherein the fault detection algorithm is obtained by optimizing a preset AI algorithm in combination with a first algorithm cost and a second algorithm cost, the first algorithm cost is at least one of a data implicit representation cost and a defect detection cost, the data implicit representation cost represents the degree of distinction between fault characterization information corresponding to the same vehicle machine fault in a sample implicit representation in an ECU repair instrument data sample library, the defect detection cost represents the degree of distinction between fault classification prediction information corresponding to the same vehicle machine fault in a sample classification prediction information in the ECU repair instrument data sample library, the sample implicit representation and the sample classification prediction information are obtained by passing the ECU repair instrument data sample library into the preset AI algorithm for fault detection, and the second algorithm cost is calculated in combination with the sample classification prediction information and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library.

[0008] In a second aspect, the present application provides an automobile ECU repair instrument data processing device, the device comprising: a data acquisition module, used to acquire ECU repair instrument data; a fault detection module, used to pass the ECU repair instrument data into a fault detection algorithm for fault detection, and obtain a fault detection result corresponding to the ECU repair instrument data; wherein the fault detection algorithm is obtained by optimizing a preset AI algorithm in combination with a first algorithm cost and a second algorithm cost, the first algorithm cost is at least one of a data implicit representation cost and a defect detection cost, the data implicit representation cost represents the degree of distinction between the fault characterization information corresponding to the same vehicle machine fault in the sample implicit representation in the ECU repair instrument data sample library, the defect detection cost represents the degree of distinction between the fault classification prediction information corresponding to the same vehicle machine fault in the sample classification prediction information in the ECU repair instrument data sample library, the sample implicit representation and the sample classification prediction information are obtained by passing the ECU repair instrument data sample library into the preset AI algorithm for fault detection, and the second algorithm cost is calculated in combination with the sample classification prediction information and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library.

[0009] Beneficial effects of the present application: The present application obtains ECU repair instrument data, passes the ECU repair instrument data into a fault detection algorithm for fault detection, and obtains a fault detection result corresponding to the ECU repair instrument data, wherein the fault detection algorithm is obtained by optimizing a preset AI algorithm in combination with a first algorithm cost and a second algorithm cost, the first algorithm cost being at least one of a data implicit representation cost and a defect detection cost, the data implicit representation cost representing the degree of distinction between the fault characterization information corresponding to the same vehicle machine fault in the sample library of the ECU repair instrument data in the sample implicit representation, and the defect detection cost representing the degree of distinction between the same vehicle machine fault in the sample library of the ECU repair instrument data in the sample classification prediction The distinction between the corresponding fault classification prediction information in the estimation information, the sample implicit representation and the sample classification prediction information are obtained by passing the ECU repair instrument data sample library into the preset AI algorithm for fault detection, and the second algorithm cost is determined by combining the sample classification prediction information and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library. The above fault detection algorithm is obtained based on the optimization obtained by combining the first algorithm cost. Data processing for the same vehicle fault type can ensure the consistency of fault diagnosis, thereby improving the fault detection accuracy of the fault detection algorithm, without large-scale modification of the algorithm composition of the fault detection algorithm, saving algorithm cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic diagram of the implementation flow of a data processing method for an automobile ECU repair instrument provided in an embodiment of the present application.

[0011] Figure 2 A schematic diagram of a training process of a fault detection algorithm provided in an embodiment of the present application.

[0012] Figure 3 A schematic diagram of the composition structure of a data processing device for an automobile ECU repair instrument provided in an embodiment of the present application.

[0013] Figure 4 A hardware entity diagram of a computer system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0014] The embodiment of the present application provides a data processing method for an automobile ECU repair instrument, which can be executed by a processor of a computer system, wherein the computer system can refer to a device with data processing capabilities such as a server, a laptop, a tablet computer, a desktop computer, a mobile device, etc.

[0015] Figure 1 A schematic diagram of the implementation flow of a data processing method for an automobile ECU repair instrument provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0016] Step S110: Acquire ECU repair instrument data.

[0017] Step S110 is the initial step in the automobile ECU repair instrument data processing method, which is executed by a computer system and is intended to obtain ECU repair instrument data.

[0018] ECU (Electronic Control Unit) is an important electronic device in the car. It controls various functions of the car, such as engine fuel injection, ignition, transmission shifting, etc. ECU repair instrument data is data related to the status, performance and possible faults of the ECU. These data contain information from multiple aspects, such as ECU input signal data, such as vehicle speed, oil temperature, intake volume and other information transmitted by sensors; there are also ECU internal operating parameter data, such as some intermediate values ​​of program operation, counter values, etc.; and ECU output control signal data, such as injection time control signal for injector nozzle, ignition advance angle control signal for spark plug, etc. The computer system can obtain ECU repair instrument data through a variety of technical means. One feasible way is to read data directly from the ECU repair instrument through the data interface. Assuming that the ECU repair instrument is connected to the computer system through a USB interface, the computer system can use the corresponding USB communication protocol to obtain data. For example, under the Windows operating system, the Windows API (Application Programming Interface) function can be used to implement data reading of USB devices. The specific formula may involve the data transmission formula in the USB communication protocol, such as: transmission data volume = transmission rate × transmission time. The transmission rate here is determined by the specifications of the USB interface. For example, the theoretical transmission rate of USB 2.0 is 480Mbps (megabits per second). In actual applications, the computer system will obtain the corresponding amount of data based on this rate and the set transmission time.

[0019] Another way to obtain the data is to read the ECU repair instrument data from a storage device. If the ECU repair instrument data has been stored on a local hard disk or other storage media (such as a solid-state drive, an external mobile hard disk, etc.), the computer system can access this data through the file system. For example, under the Linux system, the computer system can use the file I / O (Input / Output) operation function to read the ECU repair instrument data stored in a specific file. This may involve the setting of parameters such as the offset of the file read and the number of bytes read to ensure that the complete ECU repair instrument data is accurately obtained. In short, step S110 ensures that the ECU repair instrument data required for subsequent fault detection is obtained through these technical means.

[0020] Step S120: passing the ECU repair instrument data into the fault detection algorithm for fault detection, and obtaining the fault detection result corresponding to the ECU repair instrument data; wherein, the fault detection algorithm is obtained by optimizing the preset AI algorithm in combination with the first algorithm cost and the second algorithm cost, the first algorithm cost is at least one of the data implicit representation cost and the defect detection cost, the data implicit representation cost represents the distinction between the fault characterization information corresponding to the same vehicle machine fault in the sample implicit representation in the ECU repair instrument data sample library, the defect detection cost represents the distinction between the fault classification prediction information corresponding to the same vehicle machine fault in the ECU repair instrument data sample library in the sample classification prediction information, the sample implicit representation and the sample classification prediction information are obtained by passing the ECU repair instrument data sample library into the preset AI algorithm for fault detection, and the second algorithm cost is calculated in combination with the sample classification prediction information and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library.

[0021] In step S120, the fault detection algorithm is an algorithm specially designed to detect ECU faults. In the electronic control system of a car, the ECU is responsible for monitoring and controlling numerous car components and functions, and its internal data structure and operating logic are very complex. The fault detection algorithm needs to be able to parse various information in the ECU repair instrument data to determine whether there is a fault, the type of fault, and possible location.

[0022] For example, for an engine control unit (a type of ECU), it receives data from multiple sensors, such as oxygen sensors, water temperature sensors, throttle position sensors, etc. The fault detection algorithm needs to be able to analyze the representation of these sensor data in the ECU repair instrument data. Assuming that under normal circumstances, the output voltage of the oxygen sensor should fluctuate within a specific range (for example, 0.1-0.9V). If the ECU repair instrument data shows that the voltage of the oxygen sensor exceeds this range for a long time, this may mean that the oxygen sensor itself is faulty or there is a problem with the engine fuel injection system related to it. The fault detection algorithm must be able to identify this data anomaly.

[0023] The computer system can use different technical means to transfer the ECU repair instrument data to the fault detection algorithm. One way is to transfer data through the software interface. If the fault detection algorithm is implemented as an independent software module, the computer system can use the inter-process communication (IPC) mechanism provided by the operating system to transfer data. For example, under the Linux system, pipes or shared memory can be used.

[0024] For the pipeline method, assume that the process of the fault detection algorithm is process A, and the process of obtaining the ECU repair instrument data is process B. Process B can write data to the pipeline, and process A reads data from the pipeline. The specific operation involves system calls. For example, in C language, process B can use the write function to write data to the pipeline, and process A uses the read function to read data from the pipeline. There is no specific formula here, but the correctness of data transfer depends on factors such as the matching of write and read operations and the buffer size setting of the pipeline.

[0025] For the shared memory method, the computer system must first create a shared memory area. Process B writes the ECU repair instrument data to the shared memory area, and process A reads the data from the shared memory area. The use of shared memory involves operations such as memory address mapping and permission management. For example, in the Linux system, system calls such as shmget and shmat can be used to create and map shared memory.

[0026] Another possible technical means is based on network communication. If the fault detection algorithm runs on a remote server, the computer system needs to send the locally acquired ECU repair instrument data to the server through the network. This involves the use of network protocols, such as TCP / IP protocol. The computer system encapsulates the data according to the TCP / IP protocol and then sends it to the specified server IP address and port. In the process of data transmission, some network parameters are involved in the calculation. For example, the delay of data transmission can be roughly estimated by the formula: delay = transmission distance / transmission speed + queuing time + processing time. The transmission speed depends on the bandwidth of the network. For example, the theoretical transmission speed of 100Mbps Ethernet is 100Mbps, but the actual transmission speed will be affected by factors such as network congestion.

[0027] When the fault detection algorithm performs fault detection on the ECU repair instrument data, it analyzes the data according to predefined rules and algorithmic logic. Taking the transmission control ECU of a car as an example, the fault detection algorithm may check the shift signal sequence in the ECU repair instrument data. Under normal circumstances, the shift signal should appear in a certain logical order. For example, the shift signal from first gear to second gear should only appear when the vehicle speed and engine speed reach a certain threshold. If the fault detection algorithm finds that the order of the shift signal does not conform to this logic, or the shift signal appears when it should not appear, it may be determined that there is a fault in the transmission control system.

[0028] When analyzing data, the fault detection algorithm may also consider the time series characteristics of the data. For example, for the vehicle's brake system ECU, the data from the brake pressure sensor should be a smooth rise, then a steady hold, and then a slow decline during normal braking. If the fault detection algorithm finds that the brake pressure data has a sudden jump or irregular fluctuation, this may mean that there is a fault in the brake system, such as a brake line leak or a brake pump failure.

[0029] Finally, the fault detection algorithm will come up with a fault detection result based on the analysis of the ECU repair instrument data. This result may be a clear fault code indicating a specific fault type, such as "P0101-Air flow sensor circuit range / performance problem". It may also be a comprehensive diagnostic result, such as "Engine idle speed is unstable, suspected to be related to the intake system or injection system". After the computer system obtains this fault detection result, it can be used for subsequent maintenance operation guidance, fault recording or further fault analysis.

[0030] like Figure 2 As shown, it is a schematic diagram of the training process of the fault detection algorithm, which is trained by the following steps:

[0031] Step S11: obtaining an ECU repair instrument data sample library and a fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library;

[0032] Step S12: the ECU repair instrument data sample library is passed into a preset AI algorithm for fault detection, and an implicit representation of samples corresponding to the ECU repair instrument data sample library and classification estimation information of samples corresponding to the ECU repair instrument data sample library are obtained;

[0033] Step S13: combining the sample implicit representation, performing implicit representation error analysis on the ECU repair instrument data sample library to obtain the data implicit representation cost;

[0034] Step S14: combining sample classification prediction information, performing prediction error analysis on the ECU repair instrument data sample library to obtain defect detection cost;

[0035] Step S15: Determine the second algorithm cost by combining the sample classification prediction information and the fault classification label;

[0036] Step S16: Combine the first algorithm cost and the second algorithm cost to train the preset AI algorithm to obtain a fault detection algorithm.

[0037] Steps S11-S16 in the training process of the fault detection algorithm are the key processes for building an accurate and effective fault detection algorithm. These steps will be explained in detail below from the perspective of execution by a computer system.

[0038] In step S11, the ECU repair instrument data sample library is a collection of numerous ECU repair instrument data samples, which can represent the operating status of the ECU under different conditions, including normal status and various fault states. For example, for the ECU of a car engine, the data in the sample library may cover the operating data of the engine under different working conditions, such as cold start, high-speed driving, idling, etc. The fault classification label is an identifier of the fault type corresponding to each ECU repair instrument data sample. Taking the engine ECU as an example, if the engine corresponding to a certain sample has a fuel injector clogging fault, then its fault classification label may be marked as "fuel injector clogging". The computer system can obtain these data sample libraries and fault classification labels in a variety of ways.

[0039] One way is to obtain it from the historical maintenance record database of the car repair shop. When the repair shop repairs the car, it will record the relevant data of the ECU and the final fault type. The computer system can extract the required ECU repair instrument data samples and corresponding fault classification labels from this huge historical maintenance record database through database query statements. Assuming that the database uses a relational database management system (such as MySQL), the query statement may be similar to "SELECTecu_data, fault_label FROM repair_records WHERE ecu_type='engine_ecu'", where "ecu_data" represents the ECU repair instrument data, "fault_label" represents the fault classification label, and "ecu_type='engine_ecu'" is a screening condition for obtaining engine ECU related data.

[0040] Another way is to collect data in an experimental environment through special data acquisition equipment. For example, in an automobile R&D laboratory, different fault scenarios can be set to simulate various problems that may occur in the actual operation of the car. The computer system collects ECU repair instrument data in these different fault scenarios through the data acquisition equipment connected to the ECU, and the engineer manually marks the fault classification label for each sample according to the set fault situation.

[0041] In step S12, the preset AI algorithm is an algorithm with certain artificial intelligence capabilities, which can analyze and process the input data to realize the fault detection function. For example, a neural network algorithm is a feasible preset AI algorithm, such as RNN, LSTM, etc. Sample implicit representation is a way of representing the intrinsic features of ECU repair instrument data samples. Taking the neural network algorithm as an example, when processing the ECU repair instrument data sample library, the hidden layer of the neural network will perform a series of linear and nonlinear transformations on the input data, and the results of these transformations can be regarded as implicit representations of the samples. For example, for a neural network with multiple input nodes (corresponding to multiple parameters of ECU repair instrument data) and multiple hidden layer nodes, after the hidden layer is calculated, the vector output by each data sample in the hidden layer is its sample implicit representation. This vector contains the information of the data sample after the feature extraction of the neural network. It is not as intuitive as the original data, but it can better reflect the potential relationship between the data sample and the fault.

[0042] The sample classification prediction information is the preliminary prediction result of the preset AI algorithm for the fault classification of each ECU repair instrument data sample. Taking the neural network as an example, the output layer of the neural network may predict the probability that the sample belongs to different fault types based on the input data sample. Assuming that there may be three types of faults in the automotive ECU: fault A, fault B, and no fault, the output of the neural network may be a vector containing three values, such as [0.1, 0.3, 0.6], which respectively represent the estimated probabilities that the data sample belongs to fault A, fault B, and no fault. The computer system inputs the ECU repair instrument data sample library into the preset AI algorithm, and obtains the sample implicit representation and sample classification prediction information of each sample according to the calculation logic of the algorithm.

[0043] In step S13, the data implicit representation cost reflects the difference between the fault representation information corresponding to the same vehicle fault in the sample implicit representation in the ECU repair instrument data sample library. This step aims to evaluate the consistency between samples of the same fault type under the sample implicit representation.

[0044] For example, for engine misfire fault (a type of vehicle computer fault), if there is a large difference in the fault characterization information corresponding to different engine misfire fault samples in the sample implicit representation, then the data implicit representation cost will be high. This may mean that the preset AI algorithm did not capture the essential characteristics of the engine misfire fault well during the feature extraction process. The computer system determines the data implicit representation cost by analyzing the sample implicit representation and calculating the difference measure between different samples. This difference measure can be based on a variety of distance metrics, such as Euclidean distance. Suppose there are two sample implicit representation vectors and , the Euclidean distance formula is , where n is the dimension of the vector. The computer system calculates this distance between all samples of the same fault type and takes it into account to obtain the implicit representation cost of the data.

[0045] In step S14, the defect detection cost represents the degree of distinction between the fault classification prediction information corresponding to the same vehicle fault in the ECU repair instrument data sample library in the sample classification prediction information.

[0046] For example, for the same type of vehicle computer failure, such as a fuel injection system failure, if in the sample classification prediction information, different fuel injection system failure samples are predicted to have different probability distributions of different failure types, then the defect detection cost will be higher. This indicates that there are problems with the accuracy and consistency of the preset AI algorithm when classifying and predicting such failures. The computer system calculates the defect detection cost by analyzing the sample classification prediction information and comparing the differences between the prediction results of samples of the same fault type. For example, a cross-entropy loss function can be used to measure this difference. For a classification problem with C categories (such as C different types of vehicle computer failures), assuming that the true label is y (a one-hot encoded vector of length C, representing the true fault type of the sample), and the estimated probability is p (also a vector of length C, representing the estimated probability of the sample belonging to each category), the cross-entropy loss function formula is The computer system calculates the cross entropy loss of each pair of samples with the same fault type and performs comprehensive processing to obtain the defect detection cost.

[0047] In step S15, the sample classification prediction information is the fault classification prediction result of the preset AI algorithm for the data sample, and the fault classification label is the actual fault classification of the data sample. The second algorithm cost is determined based on the relationship between the two.

[0048] For example, suppose there is a car ECU data sample whose fault classification label indicates a throttle fault. However, in the sample classification prediction information given by the preset AI algorithm, the estimated probability of throttle fault is very low, while the estimated probabilities of other fault types are higher. This indicates that there is a large deviation between the preset AI algorithm's prediction on this sample and the actual situation. The computer system determines the cost of the second algorithm by comparing the probability distribution in the sample classification prediction information with the fault classification label using a suitable evaluation indicator. One feasible evaluation indicator is accuracy, which is calculated as follows: By calculating the accuracy of the entire ECU repair instrument data sample library and other related indicators, the cost of the second algorithm can be obtained. If the accuracy is low, it means that there is a large gap between the preset AI algorithm and the actual fault classification, and the cost of the second algorithm will be high.

[0049] In step S16, the first algorithm cost includes at least one of the data implicit representation cost and the defect detection cost, which reflects the performance of the preset AI algorithm in data representation and preliminary fault classification prediction; the second algorithm cost reflects the gap between the prediction result of the preset AI algorithm and the actual fault classification.

[0050] The computer system combines these two algorithm costs to adjust the parameters of the preset AI algorithm, thereby training the preset AI algorithm. For example, in a neural network algorithm, this can be achieved through the back-propagation algorithm. The back-propagation algorithm back-propagates the error from the output layer to the input layer based on the calculated algorithm cost (such as the value of the loss function) to adjust the parameters such as the weights and biases of the neural network. Assuming that the first algorithm cost and the second algorithm cost can be combined into a total loss function L, for each weight w in the neural network, the update formula is ,in is the learning rate, which indicates the step size of each parameter update. By continuously adjusting the parameters of the preset AI algorithm according to the algorithm cost, the computer system gradually optimizes the preset AI algorithm until a certain stop condition is reached (such as the algorithm cost convergence or the preset number of training rounds is reached), and finally the optimized fault detection algorithm is obtained. After such training, this fault detection algorithm can better detect faults in ECU repair instrument data and improve the accuracy and reliability of fault detection.

[0051] In summary, steps S11-S16 in the training process of the fault detection algorithm build a fault detection algorithm that can accurately detect automotive ECU faults through a series of complex and orderly operations of the computer system, from data acquisition, preliminary processing to calculation of algorithm cost, and then to final algorithm training.

[0052] In one implementation, step S13, combining the sample implicit representation, performs implicit representation error analysis on the ECU repair instrument data sample library to obtain the data implicit representation cost, including:

[0053] Step S131: splitting each ECU repair instrument data sample in the ECU repair instrument data sample library into data items to obtain a plurality of sample data items corresponding to the ECU repair instrument data sample library;

[0054] Step S132: sorting the multiple sample data items to obtain no less than one sorted data item;

[0055] Step S133: determining that each sorted data item corresponds to at least one piece of involved ECU repair instrument data, and the at least one piece of involved ECU repair instrument data corresponding to each sorted data item is an ECU repair instrument data sample having each sorted data item in the ECU repair instrument data sample library;

[0056] Step S134: combining at least one involved ECU repair instrument data corresponding to each sorted data item, determining at least one involved implicit representation corresponding to each sorted data item, and a sorted implicit representation set corresponding to at least one sorted data item; any one involved implicit representation corresponding to each sorted data item is fault characterization information corresponding to each sorted data item in any one involved ECU repair instrument data corresponding to each sorted data item; the sorted implicit representation set includes at least one involved implicit representation corresponding to each of at least one sorted data items;

[0057] Step S135: combining at least one involved implicit representation corresponding to each sorted data item, determining an involved implicit representation set corresponding to each involved implicit representation; the involved implicit representation set corresponding to any one involved implicit representation includes involved implicit representations other than any one involved implicit representation among at least one involved implicit representation corresponding to the target sorted data item; the target sorted data item is a sorted data item corresponding to any one involved implicit representation among at least one sorted data item;

[0058] Step S136: combining the involved implicit representation set and the sorted implicit representation set corresponding to each involved implicit representation, performing implicit representation similarity analysis on at least one sorted data item to obtain the data implicit representation cost.

[0059] In step S131, the ECU repair instrument data sample library contains numerous ECU repair instrument data samples, which are information records of the automotive electronic control unit (ECU) in different states (including normal state and fault state). Each data sample is a complex information set, and the computer system needs to perform data item splitting operations on it to obtain more detailed information elements, which are sample data items.

[0060] Taking the data sample of the ECU repair instrument of the automobile engine as an example, a complete data sample may contain multiple operating parameters and status information of the engine at a certain moment. For example, this data sample may contain multiple information such as engine speed, intake temperature, intake volume, injection pulse width, throttle opening, oxygen sensor feedback voltage, etc. The computer system splits this complex data sample according to these different information categories, making each category an independent sample data item.

[0061] The information about engine speed originally existed as a whole in the data sample, but through data item splitting, the computer system extracted it separately as a sample data item. This sample data item can independently reflect the engine speed at that moment without being confused with other information. Similarly, information such as intake temperature and intake volume are also split into independent sample data items. This splitting operation helps the computer system to more deeply analyze the impact of each data element on the overall status of the ECU and possible faults.

[0062] For example, when detecting whether an engine has an incomplete combustion fault, the sample data item of injection pulse width is of great significance. If the injection pulse width is too large or too small, it may lead to incomplete combustion. By separating it from the complex data sample, the computer system can more directly focus on the value change, value range and relationship between this data item and other data items (such as intake volume), so as to more accurately determine whether the engine has a fault and the possible cause of the fault.

[0063] In step S132, after the data items are split in step S131, a large number of sample data items are obtained, but these sample data items may have some problems that are not conducive to subsequent analysis, such as redundant information, irrelevant information, or inconsistent information representation. The computer system needs to sort these sample data items to obtain more valuable and easier to analyze sorted data items.

[0064] Continuing with the example of the automobile engine ECU repair instrument data, there may be some redundant information in the split sample data items. For example, some data samples may contain intake air temperature information expressed in both Celsius and Fahrenheit. In this case, the computer system will sort out the two sample data items representing intake air temperature and may choose to convert them into one unit (such as Celsius) to eliminate this redundancy.

[0065] In addition, there may be some data items that are not relevant to engine fault detection. For example, the engine ECU data sample may contain some status information of the car audio system (probably due to the comprehensiveness of data collection), which is irrelevant to engine fault detection. The computer system will identify and remove such irrelevant data items.

[0066] There is also a situation where there is a certain logical association between data items, but the representation is not clear enough. For example, there may be two sample data items, one is the theoretical intake volume calculated based on the engine speed and throttle opening, and the other is the actual intake volume directly measured. Although these two data items are both related to the intake volume, the representation and meaning are slightly different. The computer system will sort out such data items and may choose to retain one of the more representative data items according to specific needs, or merge them into a data item that can more accurately reflect the intake volume information through certain calculations.

[0067] After such a sorting operation, the sorted data items obtained by the computer system will be more refined, accurate and directly related to fault detection. These sorted data items can better reflect the status and fault characteristics of the ECU and provide a more effective data basis for subsequent analysis.

[0068] In step S133, each sorted data item is associated with some data samples in the ECU repair instrument data sample library, and the computer system needs to determine these associations, that is, find out the involved ECU repair instrument data corresponding to each sorted data item.

[0069] Taking the injection pulse width as an example, the computer system will search for data samples containing injection pulse width information in the entire ECU repair instrument data sample library. These data samples may be collected from different cars under different driving conditions (such as idling, acceleration, high-speed driving, etc.). These data samples are the ECU repair instrument data corresponding to the injection pulse width data item.

[0070] For example, for the throttle opening data item, the computer system will traverse the ECU repair instrument data sample library to find all data samples containing throttle opening information. These data samples may differ due to factors such as different car models, different years of use, and different driving habits, but their common feature is that they all contain the throttle opening data item.

[0071] Determining the involved ECU repair instrument data corresponding to each collated data item is crucial for analyzing the relationship between each collated data item and the ECU fault. By analyzing these involved ECU repair instrument data, the computer system can gain an in-depth understanding of the value range, change rules, and mutual relationship between the collated data item and other data items under different circumstances (different cars, different working conditions, etc.), so as to better judge the impact of the collated data item on fault detection. For example, for the collated data item of injection pulse width, by analyzing its corresponding involved ECU repair instrument data, the computer system can determine the reasonable value range of the injection pulse width during normal engine operation, as well as the abnormal changes in the injection pulse width under different fault conditions (such as injector nozzle blockage, oil pump failure, etc.), thereby providing a more accurate basis for the fault detection algorithm.

[0072] In step S134, during the analysis of the automobile ECU repair instrument data, the sorted data items obtained in the previous steps and their corresponding involved ECU repair instrument data are the basis for further mining data features. On this basis, the computer system determines the involved implicit representation and the sorted implicit representation set.

[0073] Take the ECU repair instrument data of the automobile engine as an example, assuming that the injection time is a collated data item. The computer system has determined the involved ECU repair instrument data corresponding to the injection time, and these data samples come from different automobile operating conditions, different engine states (normal or faulty states), etc. For each involved ECU repair instrument data sample containing the collated data item of injection time, the computer system processes it through a preset AI algorithm (such as a neural network algorithm) to obtain a sample implicit representation of each sample.

[0074] In these sample implicit representations, the fault characterization information related to the injection time data item is the involved implicit representation corresponding to the injection time data item. For example, in the hidden layer output of the neural network, the values ​​of some nodes may be closely related to the impact of the injection time on the engine fault. The values ​​of these nodes are combined to form the involved implicit representation of the injection time data item in the specific involved ECU repair instrument data sample.

[0075] For other sorted data items, such as air intake, throttle opening, etc., the computer system also determines their corresponding implicit representations in the same way. Then, the implicit representations corresponding to all these sorted data items together constitute a sorted implicit representation set. This set comprehensively contains the implicit representations of the fault characterization information of each sorted data item in different data samples.

[0076] The construction of this set of collated implicit representations helps the computer system to grasp the potential relationship between different collated data items and faults as a whole. For example, when analyzing the engine's inadequate combustion fault, the relationship between the involved implicit representations corresponding to collated data items such as injection time and intake volume in the collated implicit representation set can reflect their coordinated change rules when the fault occurs. If, in the case of an inadequate combustion fault, the involved implicit representation of the injection time shows that the injection time is too long, and the involved implicit representation of the intake volume shows that the intake volume is insufficient, then through the collated implicit representation set, the computer system can more comprehensively understand the internal relationship between these factors and provide richer information for subsequent fault detection and analysis.

[0077] In step S135, in order to analyze the relationship between each involved implicit representation in more depth, the computer system needs to determine the involved implicit representation set corresponding to each involved implicit representation, that is, the associated implicit representation set. Still taking the injection time as an example, assuming that the involved implicit representation of a certain injection time has been determined. The computer system will, among all the involved implicit representations corresponding to the injection time, except for the current involved implicit representation, the other involved implicit representations constitute the involved implicit representation set corresponding to this involved implicit representation. This way of determining the involved implicit representation set helps the computer system analyze the difference between a single involved implicit representation and other involved implicit representations belonging to the same sorted data item. For example, under different engine operating conditions, the involved implicit representation of the injection time may be different. For the involved implicit representation of the injection time under a certain specific operating condition, by comparing it with the corresponding involved implicit representation set, the computer system can determine whether the involved implicit representation of the injection time under this specific operating condition is significantly different from that under other operating conditions. If the implicit representation of the injection time when the engine is running at high speed is very different from the implicit representation of the injection time when the engine is idling (in their corresponding implicit representation sets), this may suggest that the injection time characterizes the engine fault differently under different operating conditions, thereby providing more detailed information for the fault detection algorithm. For other sorting data items, such as intake volume, the computer system will also determine the corresponding implicit representation set for each implicit representation. This can comprehensively analyze the relationship between the fault representation information of each sorting data item under different circumstances, which helps to more accurately evaluate the consistency of the same fault type in different data samples.

[0078] In step S136, the computer system performs implicit representation similarity analysis by combining each involved implicit representation with its corresponding involved implicit representation set and the sorted implicit representation set, thereby obtaining the data implicit representation cost.

[0079] Taking the engine ECU repair instrument data as an example, assuming that the injection time is the sorted data item of interest, for an implicated implicit representation of the injection time, the computer system will perform similarity analysis on it with the corresponding implicated implicit representation set (the implicated implicit representations of other injection times) and the elements in the entire sorted implicit representation set (including the implicated implicit representations of other sorted data items).

[0080] There are many ways to perform similarity analysis, such as calculating the distance between vectors (such as the Euclidean distance). Assume that under normal engine operation, different implicit representations of injection time should have high similarity. If the computer system finds through calculation that, in some cases, the Euclidean distance between a certain implicit representation of injection time and other implicit representations in its corresponding implicit representation set is large, this means that in these cases, there is a large difference in the implicit representation of the normal engine operation state by the injection time. This difference may be caused by data acquisition errors, individual differences in engines, or other factors, but it reflects the inconsistency in the implicit representation of the data.

[0081] At the same time, the computer system will also consider the relationship between the involved implicit representation of the injection time and the involved implicit representation of other sorted data items (such as intake volume, throttle opening, etc.) in the sorted implicit representation set. If there is an abnormal relationship between the involved implicit representation of the injection time and the involved implicit representation of the intake volume (for example, under normal circumstances, they should have a certain coordinated change relationship, but the actual calculation finds that this is not the case), this will also affect the overall similarity evaluation.

[0082] By performing such implicit representation similarity analysis on all sorted data items, the computer system comprehensively considers these results to determine the data implicit representation cost. This data implicit representation cost reflects the difference between the fault representation information corresponding to the same vehicle fault in the sample implicit representation in the ECU repair instrument data sample library. If the data implicit representation cost is high, it means that there are many inconsistencies in the implicit representation of the data, which may affect the accuracy of the fault detection algorithm. For example, when the fault detection algorithm converts the ECU repair instrument data into implicit representation for fault judgment, if the data implicit representation cost is high, it may cause the algorithm to deviate from the judgment of the same fault, because the implicit representation of the same fault in different data samples is quite different, making it difficult to accurately identify the fault characteristics.

[0083] In one implementation, step S136, combining the involved implicit representation set and the sorted implicit representation set corresponding to each involved implicit representation, performs implicit representation similarity analysis on at least one sorted data item to obtain the data implicit representation cost, including:

[0084] Step S1361: processing and sorting out multiple involved implicit representations in the implicit representation set one by one;

[0085] Step S1362: Determine, in combination with the sorted implicit representation set, a plurality of first involved implicit representations corresponding to the currently processed involved implicit representation; the plurality of first involved implicit representations are involved implicit representations in the sorted implicit representation set other than the currently processed involved implicit representation;

[0086] Step S1363: selecting a target implicit representation set corresponding to the currently processed implicit representation from the implicit representation set corresponding to each implicit representation; the target implicit representation set includes a plurality of second implicit representations;

[0087] Step S1364: performing commonality measurement on the currently processed implicit representation and each first implicit representation to obtain a first commonality measurement value corresponding to each first implicit representation; the first commonality measurement value corresponding to each first implicit representation indicates the matching degree between each first implicit representation and the currently processed implicit representation;

[0088] Step S1365: performing commonality measurement on the currently processed implicit representation and each second implicit representation to obtain a second commonality measurement value corresponding to each second implicit representation; the second commonality measurement value corresponding to each second implicit representation indicates the matching degree between each second implicit representation and the currently processed implicit representation;

[0089] Step S1366: Determine a third algorithm cost corresponding to the currently processed implicit representation by combining the second commonality metric value corresponding to each second implicit representation and the first commonality metric values ​​corresponding to each of the plurality of first implicit representations;

[0090] Step S1367: At the end of the processing, the multiple third algorithm costs obtained during the processing are averaged to obtain the data implicit representation cost.

[0091] In step S1361, the sorted implicit representation set includes at least one involved implicit representation corresponding to each sorted data item. Taking the automobile engine ECU repair instrument data as an example, it is assumed that the sorted data items include injection time, intake volume, throttle opening, etc. For the sorted data item of injection time, its involved implicit representation is the fault characterization information corresponding to the injection time obtained from the involved ECU repair instrument data sample containing the injection time through a preset AI algorithm (for example, a feature vector related to the injection time output by the hidden layer of the neural network). Similarly, the intake volume and throttle opening also have their own corresponding involved implicit representations, and these involved implicit representations together constitute the sorted implicit representation set.

[0092] The computer system processes each implicit representation in this set. For example, when processing the implicit representation of injection time, the computer system will perform a series of analysis operations on this specific implicit representation, which will be carried out through subsequent sub-steps. This one-by-one processing method ensures that each factor that may affect the cost of the implicit representation of data can be carefully examined.

[0093] Next is step S1362, combining the sorted implicit representation set, determining multiple first involved implicit representations corresponding to the currently processed involved implicit representation; multiple first involved implicit representations are involved implicit representations in the sorted implicit representation set, other than the currently processed involved implicit representation. Continuing with the above engine ECU repair instrument data as an example, when the computer system is processing a certain involved implicit representation of injection time, it will check other involved implicit representations in the sorted implicit representation set. These other involved implicit representations (i.e., involved implicit representations other than the currently processed involved implicit representation of injection time) are the so-called first involved implicit representations. For example, if the currently processed implicit representation of injection time is in vector form v_{ft} (where ft represents injection time), then the implicit representation of intake air volume vector v_{ia} (where ia represents intake air volume) and the implicit representation of throttle opening degree vector v_{td} (where td represents throttle opening degree) are all first implicit representations.

[0094] This operation of determining the first implicate implicit representation is to establish a comparative relationship between the implicate implicit representation currently being processed and other implicate implicit representations in the set. In the context of fault detection, by comparing the implicate implicit representation of the injection time with the implicate implicit representation of other collated data items such as the intake volume and throttle opening, the association and difference in implicit representation between different data items can be found. For example, when the engine is operating normally, there may be a certain inherent logical relationship between the implicit representation of the injection time and the implicit representation of the intake volume (e.g., as the intake volume increases, the injection time may be adjusted accordingly), while in a fault state, this relationship may be destroyed. By determining the first implicate implicit representation and making a comparison, the computer system can capture the change in this relationship, thereby providing a basis for determining the cost of the implicit representation of the data.

[0095] Then comes step S1363, in which a target implicit representation set corresponding to the currently processed implicit representation is selected from the implicit representation set corresponding to each implicit representation; the target implicit representation set includes multiple second implicit representations. Still taking the engine ECU repair instrument data as an example, for the sorting data item of injection time, each implicit representation of injection time has its corresponding implicit representation set. This implicit representation set is composed of the implicit representations of all injection times except the implicit representation currently being processed. When the computer system processes a specific implicit representation of injection time, it selects the target implicit representation set from the implicit representation set of this injection time. Assuming that the injection time has multiple implicit representations under different working conditions (such as idling, acceleration, high-speed driving, etc.), when processing the implicit representation of the injection time under the idling working condition, the implicit representations of the injection time under other working conditions (acceleration, high-speed driving, etc.) except the implicit representation under the idling working condition constitute the target implicit representation set. The implicit representations in these target implicit representation sets are the second implicit representations.

[0096] This operation of selecting the target involved implicit representation set helps the computer system to analyze the implicit representation differences of the same collated data item in different situations in a more detailed manner. For example, under normal circumstances, although the implicit representations of the injection time under different working conditions are different, there should be certain rules and similarities. If this rule is broken under a fault state, the computer system can find this abnormal change by comparing the currently processed injection time involved implicit representation (such as under idle conditions) with the second involved implicit representation in the target involved implicit representation set (the injection time involved implicit representation under conditions such as acceleration and high-speed driving), thereby more accurately evaluating the consistency of the implicit representation of the collated data item injection time under different working conditions, which is very critical for calculating the data implicit representation cost. Because the data implicit representation cost reflects the difference between the fault representation information corresponding to the same vehicle fault in the sample implicit representation, and by analyzing the implicit representation of the same collated data item under different conditions (i.e., comparing with the first involved implicit representation and the second involved implicit representation), the source and influencing factors of this difference can be deeply explored.

[0097] In step S1364, the commonality measurement is performed on the currently processed implicit representation and each first implicit representation, and the first commonality measurement value corresponding to each first implicit representation is obtained; the first commonality measurement value corresponding to each first implicit representation indicates the matching degree between each first implicit representation and the currently processed implicit representation. In this step, the computer system performs commonality measurement on the currently processed implicit representation and each previously determined first implicit representation. Taking the automobile engine ECU repair instrument data as an example, assuming that the currently processed item is an implicit representation corresponding to the injection time sorting data item (set as ), the first implied implicit representation includes the implied implicit representation corresponding to the intake volume (set to ), the throttle opening corresponding to the implicit representation (set to )wait.

[0098] The commonality measure is to measure the similarity or matching degree between them. There are many methods to measure commonality, such as calculating the cosine similarity between vectors. , the cosine similarity formula is: ,in is the dot product of the vectors, are the moduli of the two vectors respectively. The calculated cosine similarity value is the first commonality metric, which indicates the implicit representation of the intake volume Implicit representation of the injection timing The degree of match between them.

[0099] When the engine is operating normally, there is a certain internal connection between the injection time and the intake volume, which will be reflected in the commonality metric of their implicit representation. For example, if the control logic of the engine makes the injection time and the intake volume change in a certain proportion under normal operating conditions, then the first commonality metric value of their implicit representation should be high, indicating a high degree of match between the two. However, if the engine fails, such as the fuel injector is blocked or the intake pipe leaks, this synergistic relationship may be destroyed, which is reflected in the commonality metric as a decrease in the first commonality metric value.

[0100] In step S1365, the commonality measurement is performed for the currently processed implicit representation and each second implicit representation in the previously selected target implicit representation set. Still taking the injection time as an example, assuming that the currently processed one is the implicit representation of the injection time under the idle condition (set as ), the target implicit representation set is the implicit representation of the injection time under acceleration, high-speed driving and other working conditions (set to wait).

[0101] A similar method is also used to measure commonality, such as calculating the cosine similarity of vectors. The calculated cosine similarity is the second commonality measure, which indicates the implicit representation of the injection time under acceleration conditions. Implicit representation of the injection time at idle speed The degree of match between them.

[0102] Under normal circumstances, although the implicit representation of the injection time under different working conditions will be different, due to the basic operating logic of the engine, there are still certain similarities between them, which will be reflected in the second commonality metric. For example, under normal circumstances, the adjustment of the injection time under idle and acceleration conditions is based on certain rules, so their second commonality metric of implicit representation should be within a reasonable range. However, if some parts of the engine fail, such as a sensor failure in the injection control system, it may cause the adjustment of the injection time under different working conditions to lose the normal logical relationship, thereby causing the second commonality metric to deviate from the normal range.

[0103] In step S1366, the computer system needs to comprehensively consider the second commonality metric value and the first commonality metric value to determine the third algorithm cost. Continuing with the injection time as an example, assuming that the first commonality metric value of the implicit representation of the intake air volume and the implicit representation of the injection time has been obtained is The first commonality metric between the implicit representation of the throttle opening and the implicit representation of the injection time is , and the second common measure of injection time under acceleration conditions is The second common measurement value of injection time under high-speed driving conditions is wait.

[0104] One way to determine the cost of the third algorithm is through a predefined cost function, which may take into account the differences, weights, etc. of these commonality metrics. For example, the cost function can be: ,in is a pre-set baseline commonality metric, is the weight corresponding to different commonality measures. This weight can be set according to the importance of different data items in fault detection. For example, the relationship between injection time and intake volume may be more critical in some fault types, then The value may be relatively large.

[0105] The C value obtained by such calculation is the cost of the third algorithm corresponding to the currently processed implicit representation (such as the implicit representation of injection time under idle speed condition). This cost reflects the comprehensive difference between the current implicit representation and other related implicit representations (the first implicit representation and the second implicit representation) in terms of commonality measurement, thus reflecting the consistency of the implicit representation of the data to a certain extent.

[0106] In step S1367, after the computer system completes processing of all involved implicit representations in the sorted implicit representation set, it will obtain multiple third algorithm costs. For example, when processing multiple involved implicit representations (corresponding to different working conditions or different data samples) of the sorted data item injection time, a series of third algorithm cost values ​​will be obtained, set as wait.

[0107] The average cost of these third algorithms is calculated (where n is the number of costs of the third algorithm), and the value obtained is the data implicit representation cost. This data implicit representation cost represents the difference between the fault representation information corresponding to the same vehicle fault in the sample implicit representation in the ECU repair instrument data sample library.

[0108] If the cost of implicit data representation is high, it means that there are large differences between the implicit representations corresponding to the same fault type in different data samples or different working conditions. For example, in engine fault detection, if the cost of implicit data representation is high, it may mean that for the same fuel injection system fault, the implicit representation of ECU repair instrument data under different cars and different driving conditions is quite different, which will make it difficult for the fault detection algorithm to accurately identify the fault. By calculating and analyzing the cost of implicit data representation, it can provide an important basis for the subsequent optimization of the fault detection algorithm, such as adjusting the feature extraction method of the algorithm or optimizing the parameters in the algorithm, so as to improve the accuracy and reliability of fault detection.

[0109] In one implementation, step S1366, combining the second commonality metric value corresponding to each second involved implicit representation and the first commonality metric values ​​corresponding to each of the plurality of first involved implicit representations, to determine a third algorithm cost corresponding to the currently processed involved implicit representation, includes:

[0110] Step S13661: Determine an estimated support coefficient corresponding to each second implicit representation by combining the second commonality metric value corresponding to each second implicit representation and the first commonality metric values ​​corresponding to each of the plurality of first implicit representations, wherein the estimated support coefficient corresponding to each second implicit representation indicates the possibility of association between each second implicit representation and the currently processed implicit representation;

[0111] Step S13662: Determine a fourth algorithm cost corresponding to each second implicit representation in combination with the estimated support coefficient corresponding to each second implicit representation, wherein the fourth algorithm cost represents the degree of distinction between each second implicit representation and the currently processed implicit representation;

[0112] Step S13663: Accumulate the fourth algorithm costs corresponding to each of the plurality of second involved implicit representations to obtain the third algorithm cost.

[0113] In step S13661, the operation is performed using the data of the automobile engine ECU repair instrument as an example. Assuming that the current processing is the implicit representation of the injection time under a certain working condition (such as the idling condition), set The first implicit representation corresponding to other sorted data items related to the injection time (such as intake volume, throttle opening, etc.) has been calculated before. The first commonality measure of (the first common metric value corresponding to the intake volume), (the first commonality metric corresponding to the throttle opening), etc.; at the same time, the second implicated implicit representation and The second commonality measure of (the second commonality metric corresponding to the acceleration condition), (The second commonality measurement value corresponding to high-speed driving conditions) etc.

[0114] In order to determine the estimated support coefficient corresponding to each second implicated implicit representation, the computer system adopts a calculation method based on the commonality metric. For example, it can be assumed that there is a predefined function relationship, which takes the first commonality metric and the second commonality metric as input and outputs the estimated support coefficient. Assume that this function is: ,in is the estimated support coefficient corresponding to the i-th second implicate implicit representation, is the second commonality metric value corresponding to the i-th second implicit representation or the first commonality metric value related to it (depending on the specific definition of the function, these commonality metric values ​​may need to be combined or converted as input), and j represents the index of all commonality metric values ​​involved in the calculation (including the first commonality metric value and the second commonality metric value).

[0115] Taking the second involved implicit representation of injection time under acceleration condition as an example, if the estimated support coefficient is calculated by the above function, , which means that according to the current calculation result of the commonality metric, the probability of association between the second implicated implicit representation of the injection time under acceleration condition and the implicated implicit representation of the injection time under idle condition is 0.3. This probability of association reflects the logical relationship between different conditions or different data items in the entire data implicit representation system. For example, when the engine is operating normally, since the injection time is adjusted based on a certain control logic under different conditions, under normal circumstances, this estimated support coefficient should be within a reasonable range. If the engine fails, such as a failure in the injection control system, this estimated support coefficient may change significantly and deviate from the normal range.

[0116] In step S13662, the computer system continues to operate with the above injection time example. For each second implied representation, the corresponding estimated support coefficient has been obtained. For example, the estimated support coefficient corresponding to the second implied representation of the injection time under the acceleration condition is The estimated support coefficient corresponding to the second implicit representation of the injection time under high-speed driving conditions is .

[0117] In order to determine the fourth algorithm cost, the computer system may use a calculation method related to the estimated support coefficient. For example, assuming that the calculation formula of the fourth algorithm cost is: ,in is the fourth algorithm cost corresponding to the ith second implicit representation. For the second implicit representation of injection time under acceleration conditions, its fourth algorithm cost is ; For the second implicit representation of injection time under high-speed driving conditions, the fourth algorithm cost is .

[0118] This fourth algorithm cost reflects the difference between each second implicated implicit representation and the currently processed implicated implicit representation (such as the implicated implicit representation of the injection time under idle conditions). From a numerical point of view, the greater the fourth algorithm cost, the greater the difference between the two. When the engine is operating normally, the difference between the second implicated implicit representation of the injection time under different conditions and the implicated implicit representation of the injection time under the current condition should be within a certain range. For example, since the adjustment of the injection time of the engine under different conditions follows certain rules, this difference will not be too large under normal circumstances. However, when the engine fails, such as the fuel injector is blocked or the sensor of the injection control system fails, this difference may increase significantly, which is manifested as a significant increase in the fourth algorithm cost.

[0119] In step S13663, the example of injection time is continued. The fourth algorithm cost corresponding to the second involved implicit representation of injection time under different working conditions has been calculated, such as under acceleration condition. and high-speed driving conditions .

[0120] The computer system accumulates and sums the fourth algorithm costs corresponding to all the second implicit representations to obtain the third algorithm cost. Assuming that there are only the second implicit representations under these two conditions, then the third algorithm cost is .

[0121] This third algorithm cost comprehensively reflects the comprehensive distinction between the currently processed implicit representation (such as the implicit representation of injection time under idle conditions) and other related second implicit representations (the implicit representation of injection time under other conditions). In the context of fault detection, this cost reflects the degree of influence of the difference between the representation of injection time under different conditions and the representation of injection time under the current condition on fault detection in terms of data implicit representation. If this cost is high, it means that there is a large inconsistency in the implicit representation of data, which may affect the fault detection algorithm's accurate judgment of injection time-related faults. For example, in a fault detection algorithm that classifies faults based on data implicit representation, if the third algorithm cost is too high, it may cause the algorithm to misjudge whether the injection time is within the normal range or whether there is a fault, because the relationship between the implicit representation of injection time under different conditions becomes complicated and difficult to accurately grasp.

[0122] In one implementation, step S132, arranging the multiple sample data items to obtain at least one arranged data item, includes:

[0123] Step S1321: Compare each sample data item with the pre-deployed ECU repair instrument data to obtain comparison results corresponding to multiple sample data items;

[0124] Step S1322: filtering out sample data items belonging to pre-deployed ECU repair instrument data from the plurality of sample data items in combination with the comparison result, to obtain filtered sample data items;

[0125] Step S1323: clear duplicate items from the filtered sample data items to obtain no less than one sorted data item.

[0126] For step S1321, the automobile ECU repair instrument data sample library contains numerous sample data items, which reflect various parameter information of the automobile electronic control unit (ECU) in different states (such as normal operation, fault state, etc.). The computer system needs to compare each sample data item with the pre-deployed ECU repair instrument data. The pre-deployed ECU repair instrument data is a reference standard, which contains data information related to the automobile ECU that is predetermined to be representative or standard. For example, for the ECU of a certain automobile engine, the pre-deployed ECU repair instrument data may contain information such as the normal value range of various engine parameters under standard operating conditions (such as specific ambient temperature, standard fuel quality, normal engine load, etc.), data format, and expected relationship between data.

[0127] Taking the sample data item of engine speed as an example, the computer system will match each sample data item of engine speed obtained from the sample library of ECU repair instrument data with the engine speed related information in the pre-deployed ECU repair instrument data. If the pre-deployed ECU repair instrument data stipulates that the normal range of engine speed at idle state is 700-900 rpm, and the engine speed value in the sample data item is 1200 rpm, then the computer system will obtain the comparison result of this sample data item with the pre-deployed data in terms of engine speed, that is, the engine speed in the sample data item exceeds the normal idle range. Similarly, for other sample data items, such as intake volume, injection time, etc., the computer system will also compare in the same way. For the sample data item of intake volume, if the pre-deployed ECU repair instrument data stipulates that the normal range of intake volume under specific working conditions is a certain value interval, and the intake volume value in the sample data item is not within this interval, the computer system will also record the corresponding comparison result. This comparison result can not only reflect whether the sample data item is within the normal range, but also whether the format of the data is consistent with the previously deployed data, and whether the relationship between the data is as expected. For example, in some cases, there is a certain proportional relationship between the injection time and the intake volume. If the numerical combination of the injection time and the intake volume in the sample data item does not meet this expected relationship, the computer system can also find this difference through comparison.

[0128] In step S1322, after the comparison is completed, the computer system will screen the sample data items according to the comparison results. For those sample data items that belong to the pre-deployed ECU repair instrument data, the computer system will filter them out. For example, the pre-deployed ECU repair instrument data may contain some data templates or standard data values ​​for initialization or calibration purposes. Assuming that there are some sample data items in the ECU repair instrument data sample library that are directly copied from these initialization or calibration data, these sample data items may not have practical significance for fault detection because they are only standard reference values, not data that reflects the actual operating status of the car. Based on the comparison results obtained previously, the computer system can identify these sample data items and remove them from the numerous sample data items.

[0129] Continuing with the example of engine speed, if there is an engine speed value in the sample data item that is a pre-deployed standard idle speed value (e.g. 800 rpm), and the computer system finds through comparison that this data item is completely consistent with the pre-deployed standard data, then this data item will be filtered out. Because it is a standard reference value, it cannot provide useful information about the actual operating status of the engine (such as whether there is a fault). Similarly, other sample data items such as intake volume and injection time that completely match the pre-deployed data will also be filtered out.

[0130] After this step, the remaining sample data items (i.e., the sample data items after filtering) will be more focused on reflecting the real state and changes of the automobile ECU during actual operation, which will help to conduct more targeted fault detection and analysis in the future.

[0131] In step S1323, after filtering out the sample data items belonging to the pre-deployed ECU repair instrument data, there may still be some duplicate items in the remaining filtered sample data items. These duplicate items may be caused by some reasons in the data collection process (such as multiple collections of the same parameter or data redundancy in the collection process). The computer system needs to clean up the duplicate items of these filtered sample data items to obtain more concise and effective sorted data items.

[0132] Take the throttle opening data of a car as an example. In the sample data items after filtering, there may be multiple data items recording the throttle opening. These data items may be acquired at different time points or different collection cycles, but they are essentially all information about the throttle opening. The computer system will identify these repeated throttle opening sample data items and clean them up according to certain rules.

[0133] One possible cleaning rule is to retain the latest or most representative data items. For example, if there are multiple throttle opening sample data items, the computer system can choose to retain the data item acquired at the collection time closest to the current time, because this data item may better reflect the current actual state of the car. Alternatively, if some data items are collected under special operating conditions (such as sudden acceleration or sudden braking) and other data items are collected under normal operating conditions, the computer system can choose to retain the data items under special operating conditions or the data items under normal operating conditions according to the specific fault detection requirements.

[0134] For other data items that may have duplicates, such as engine temperature, oxygen sensor feedback voltage, etc., the computer system will also clean them up according to similar rules. After cleaning the duplicate items, the sorted data items will no longer contain a large amount of duplicate information. Each sorted data item has a unique meaning and can more effectively reflect the status information of the automobile ECU, which lays a good foundation for further analysis of the relationship between the ECU repair instrument data and faults.

[0135] In one implementation, in step S1323, after performing duplicate cleaning on the filtered sample data items to obtain at least one sorted data item, the method further includes:

[0136] Step S1324: determining that at least one main sorted data item among at least one sorted data item; the existence ratio of any one of the at least one main sorted data item in the ECU repair instrument data sample library is greater than a set ratio;

[0137] On this basis, step S133, determining that each sorted data item corresponds to at least one piece of data related to the ECU repair instrument, may include:

[0138] Step S1331: Determine that each main sorted data item corresponds to at least one piece of ECU repair instrument data.

[0139] After the computer system performs duplicate cleaning on the filtered sample data items to obtain sorted data items, it is necessary to determine the main sorted data items. The main sorted data items are data items with a high frequency of occurrence or a large proportion in the entire ECU repair instrument data sample library.

[0140] Take the ECU repair instrument data of automobile engines as an example. Assume that the sorted data items obtained through the previous steps include engine speed, intake temperature, injection time, etc. The computer system will count the number of occurrences or the proportion of each sorted data item in the ECU repair instrument data sample library. For example, if the sorted data item of engine speed exists in 90% of the data samples in the entire data sample library, and the set proportion is 80%, then the engine speed can be determined as a main sorted data item.

[0141] For intake air temperature, if it exists in 70% of the data samples in the data sample library, and the set ratio is 60%, then the intake air temperature can also be regarded as a major collation data item. These major collation data items are important because they exist in a large number of data samples, which means that they are more critical to describing the status of the vehicle ECU and possible faults.

[0142] In the actual operation and failure of a car, the information of certain parameters always appears frequently and may be more closely related to the failure. For example, the engine speed is closely related to the various operating states of the engine. Whether it is the stable speed during normal operation or the abnormal speed change under the fault state, it can provide important clues for fault detection. Similarly, changes in intake temperature will also have an important impact on the combustion process of the engine, and under different types of faults (such as intake system failure, incomplete combustion, etc.), the performance of intake temperature often has a certain regularity. Therefore, determining these main collated data items helps the computer system focus more on these key data information in subsequent fault detection analysis, and improves the efficiency and accuracy of fault detection.

[0143] In step S1331, after determining the main sorting data items, the computer system needs to determine the involved ECU repair instrument data corresponding to each main sorting data item. The involved ECU repair instrument data here refers to those data samples in the ECU repair instrument data sample library that contain specific main sorting data items.

[0144] Taking the main data item of engine speed as an example, the computer system will search for data samples containing engine speed information in the entire ECU repair instrument data sample library. These data samples may come from different cars and are collected under different driving conditions (such as idling, acceleration, high-speed driving, etc.). For example, one data sample may record the engine speed, intake volume, injection time and other information of the car at idle speed, and another data sample may record the engine speed, throttle opening, exhaust temperature and other information of the car at high speed. In these two data samples, since both contain engine speed information, they both belong to the ECU repair instrument data corresponding to the main data item of engine speed.

[0145] For the main collated data item of intake air temperature, the computer system will also search for data samples containing intake air temperature information in the data sample library. These data samples may cover intake air temperatures under different ambient temperatures, different engine running times, and other related ECU parameters. For example, in the data samples collected when starting the car in cold weather, the intake air temperature is low, and information such as the engine start time and initial injection strategy are also recorded; in the data samples collected during normal driving in hot weather, the intake air temperature is high, and ECU-related information such as vehicle speed and air conditioning usage status is included. The data samples containing intake air temperature information in these different situations are all ECU repair instrument data corresponding to the main collated data item of intake air temperature.

[0146] Determining the ECU repair instrument data corresponding to each main collation data item is crucial for in-depth analysis of the relationship between each main collation data item and the ECU fault. By analyzing these ECU repair instrument data, the computer system can understand the value range, change rules and mutual relationship between the main collation data items under different conditions (different cars, different working conditions, different environments, etc.), so as to better judge the impact of the main collation data item on fault detection. For example, for the main collation data item of engine speed, by analyzing its corresponding ECU repair instrument data, the computer system can determine the reasonable value range of the engine speed during normal engine operation, and the abnormal changes in the engine speed under different fault conditions (such as engine cylinder shortage, insufficient fuel supply, etc.), thereby providing a more accurate basis for the fault detection algorithm. Similarly, for the main collation data item of intake temperature, by analyzing its corresponding ECU repair instrument data, it is possible to understand how the change of intake temperature affects the combustion efficiency of the engine, as well as the abnormal performance of the intake temperature data under conditions such as intake temperature sensor failure, etc., which helps to improve the accuracy and reliability of fault detection.

[0147] In one implementation, step S11, obtaining an ECU repair instrument data sample library and a fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library, includes:

[0148] Step S111: acquiring a plurality of first ECU repair instrument data samples from a pre-deployed ECU repair instrument data set;

[0149] Step S112: copy each first ECU repair instrument data sample to obtain a second ECU repair instrument data sample corresponding to each first ECU repair instrument data sample;

[0150] Step S113: Building an ECU repair instrument data sample library by combining the plurality of first ECU repair instrument data samples and the second ECU repair instrument data samples corresponding to the plurality of first ECU repair instrument data samples;

[0151] Step S114: Determine the fault classification label corresponding to the ECU repair instrument data sample library in combination with the fault classification label corresponding to the previously deployed ECU repair instrument data set.

[0152] In step S111, the pre-deployed ECU repair instrument data set is a pre-existing set containing a large amount of data related to the automotive electronic control unit (ECU). The source of this data set may be diverse, such as test data accumulated by automobile manufacturers during vehicle development, actual vehicle ECU data collected by automobile repair shops during daily maintenance work, etc.

[0153] The computer system obtains multiple first ECU repair instrument data samples in this data set. Taking the ECU of a car engine as an example, these data samples may contain various parameter information of the engine under different working conditions. For example, a data sample may record the information of the engine under idle conditions, including engine speed (such as 700-900 rpm), intake volume (such as a specific air flow value), injection time (such as injection pulse width of several milliseconds), throttle opening (such as 10%-15% opening) and oxygen sensor feedback voltage (such as a value between 0.1-0.9 volts). Another data sample may be the relevant parameters of the engine under high-speed driving conditions. The engine speed may reach 3000-5000 rpm, the intake volume increases significantly, the injection time and throttle opening are also adjusted accordingly, and the oxygen sensor feedback voltage is also in different ranges. These engine ECU data samples under different working conditions are the first ECU repair instrument data samples obtained by the computer system from the pre-deployed ECU repair instrument data set.

[0154] These first ECU repair instrument data samples are the basis for the subsequent construction of the ECU repair instrument data sample library. They cover a wealth of automotive ECU operating status information, including normal operating status and possible fault status, providing original data materials for the training of fault detection algorithms.

[0155] In step S112, the computer system copies each first ECU repair instrument data sample acquired in step S111. The purpose of copying is to increase the number of data samples and to have more data for analysis and processing when subsequently building a data sample library.

[0156] Continuing with the engine ECU data as an example, assume that a first ECU repair instrument data sample obtained previously records various parameters of the engine under specific working conditions. Through the copy operation, the corresponding second ECU repair instrument data sample is obtained. This second ECU repair instrument data sample is exactly the same as the first ECU repair instrument data sample in terms of data content. For example, the engine speed in the first ECU repair instrument data sample is 1500 rpm, the intake volume is a certain value, the injection time is a specific pulse width, the throttle opening is 20%, and the oxygen sensor feedback voltage is 0.5 volts, then these parameter values ​​in the copied second ECU repair instrument data sample are also exactly the same.

[0157] This copy operation helps to improve the richness of the data in some cases, especially when the amount of data is relatively small or special processing is required for specific data samples. At the same time, it also provides more flexibility in the subsequent construction of the data sample library and the determination of fault classification labels.

[0158] In step S113, the computer system combines the previously acquired multiple first ECU repair instrument data samples and their corresponding second ECU repair instrument data samples to construct an ECU repair instrument data sample library. This data sample library becomes a collection of ECU related data that is centrally stored for fault detection algorithm training.

[0159] For example, suppose three first ECU repairer data samples are obtained from the pre-deployed ECU repairer data set, which are ECU data samples of the engine under idling, acceleration and high-speed driving conditions, and each first ECU repairer data sample is copied to obtain the corresponding second ECU repairer data sample. The computer system combines these six data samples (three first ECU repairer data samples and three second ECU repairer data samples) to build an ECU repairer data sample library. Each data sample in this data sample library contains rich engine ECU related parameter information, which varies under different working conditions and covers all aspects of engine operation, from speed, intake volume to injection control, throttle control and oxygen sensor feedback.

[0160] This ECU repair instrument data sample library is a comprehensive data set containing a variety of ECU operating status information, providing a rich data foundation for subsequent fault detection algorithm training. The differences and commonalities between different data samples will help the algorithm learn how to distinguish different fault types and normal operating conditions, thereby improving the accuracy of the fault detection algorithm.

[0161] In step S114, the pre-deployed ECU repair instrument data set has corresponding fault classification labels. These fault classification labels are identifiers of the ECU state (whether there is a fault and what kind of fault) represented by each data sample in the data set.

[0162] For example, for engine ECU data, if in the pre-deployed ECU repair instrument data set, the engine speed in a certain data sample continues to be abnormally high and the injection timing is also abnormal, and after actual testing or empirical judgment, the engine corresponding to this data sample has a fuel injection system failure, then the fault classification label of this data sample in the pre-deployed ECU repair instrument data set is "fuel injection system failure."

[0163] The computer system determines the fault classification label corresponding to each data sample in the ECU repair instrument data sample library according to the fault classification labels of the pre-deployed ECU repair instrument data set. Since the ECU repair instrument data sample library is constructed by the first ECU repair instrument data sample in the pre-deployed ECU repair instrument data set and the second ECU repair instrument data sample obtained by copying it, the fault classification label in the source data set can be directly matched to the data sample in the constructed data sample library.

[0164] For example, in a previously constructed ECU repairer data sample library containing six engine ECU data samples (three first ECU repairer data samples and three second ECU repairer data samples), if the fault classification label corresponding to the first first ECU repairer data sample in the previously deployed ECU repairer data set is "throttle fault", then the fault classification label of this data sample and its corresponding second ECU repairer data sample in the ECU repairer data sample library is also determined to be "throttle fault".

[0165] In this way, each data sample in the ECU repair instrument data sample library has a corresponding fault classification label, which is critical for the subsequent operation of transferring the data sample library to the preset AI algorithm for fault detection and calculating the algorithm cost. The fault classification label provides a basis for the algorithm to make judgments, enabling the algorithm to learn the relationship between different data features and fault types, thereby achieving accurate fault detection.

[0166] In one implementation, step S14, combining sample classification prediction information, performs prediction error analysis on the ECU repair instrument data sample library to obtain the defect detection cost, including:

[0167] Step S141: determining a plurality of data item groups, wherein any one of the data item groups includes a first data item and a second data item, the first data item is any one of the first ECU repair instrument data samples, the second data item is a data item corresponding to the first data item in a third ECU repair instrument data sample, and the third ECU repair instrument data sample is a second ECU repair instrument data sample corresponding to any one of the first ECU repair instrument data samples;

[0168] Step S142: determining the estimated fault type group corresponding to each data item group; the estimated fault type group corresponding to each data item group includes the fault classification estimation information corresponding to the first data item group in each data item group and the fault classification estimation information corresponding to the second data item group in each data item group;

[0169] Step S143: combining the estimated fault type group corresponding to each data item group, performing estimated fault error analysis on the ECU repair instrument data sample library to obtain the defect detection cost.

[0170] In step S141, the ECU repair instrument data of the automobile engine is taken as an example, and the operation is performed in the previously constructed ECU repair instrument data sample library. Assume that there is a first ECU repair instrument data sample in the ECU repair instrument data sample library, which contains data items such as engine speed, intake volume, injection time, etc. This data sample is copied to obtain the corresponding second ECU repair instrument data sample.

[0171] For the data item of engine speed in the first ECU repair instrument data sample (i.e., the first data item), the corresponding data item of engine speed in the corresponding second ECU repair instrument data sample (i.e., the third ECU repair instrument data sample) (i.e., the second data item) constitutes a data item group. Similarly, for data items such as intake volume and injection time, the data items in the two corresponding data samples can also be found, thereby forming a corresponding data item group.

[0172] The construction of this data item group is based on the correspondence between the first ECU repair instrument data sample and the second ECU repair instrument data sample obtained by copying it. The data item group constructed in this way can reflect the situation of the same data item in the original data sample and the copied data sample, providing a basis for subsequent analysis of the changes in data in different samples.

[0173] For example, in a specific automobile engine ECU data scenario, the engine speed in the first ECU repair instrument data sample is 1500 rpm (first data item), and the corresponding engine speed in the second ECU repair instrument data sample is 1500 rpm (second data item), which constitutes a data item group about the engine speed. For the intake volume, if the intake volume in the first ECU repair instrument data sample is 20 cubic meters / hour (first data item), the intake volume in the second ECU repair instrument data sample is 20 cubic meters / hour (second data item), which constitutes another data item group about the intake volume. These data item groups will be used in subsequent steps to analyze the differences in the estimated fault types of the same data item in different data samples.

[0174] In step S142, after constructing the data item group, the computer system needs to determine the estimated fault type group corresponding to each data item group. In the previous step, the ECU repair instrument data sample library has been passed into the preset AI algorithm for fault detection, and the sample classification estimation information has been obtained.

[0175] Taking the previously constructed engine speed data item group as an example, assuming that the preset AI algorithm performs fault prediction on the first ECU repair instrument data sample (the sample where the first data item is located) containing an engine speed of 1500 rpm, the fault classification prediction information obtained is that the engine is operating normally (this is the fault classification prediction information corresponding to the first data item). For the second ECU repair instrument data sample (the sample where the second data item is located) containing an engine speed of 1500 rpm, the fault classification information predicted by the preset AI algorithm is that the engine is operating normally (this is the fault classification prediction information corresponding to the second data item). Then for this engine speed data item group, the corresponding estimated fault type group is [engine operating normally, engine operating normally]. Taking the intake volume data item group as an example, if the fault classification prediction information corresponding to the intake volume in the first ECU repair instrument data sample is that the intake system is normal (the fault classification prediction information corresponding to the first data item), and the fault classification prediction information corresponding to the intake volume in the second ECU repair instrument data sample is that the intake volume is slightly high but still normal (the fault classification prediction information corresponding to the second data item), then the estimated fault type group corresponding to this intake volume data item group is [intake system is normal, the intake volume is slightly high but still normal].

[0176] The purpose of determining the estimated fault type group corresponding to each data item group is to subsequently analyze whether there are differences in the fault type estimates of the preset AI algorithm in different data samples of the same data item. This difference will be closely related to the calculation of the defect detection cost.

[0177] In step S143, after determining the estimated fault type group corresponding to each data item group, the computer system starts to perform estimated fault error analysis on the ECU repair instrument data sample library to obtain the defect detection cost.

[0178] Take the estimated fault type group [engine normal operation, engine normal operation] corresponding to the engine speed data item group as an example. In this case, the estimated fault type corresponding to the engine speed in the two data samples is the same, indicating that on this data item, the fault prediction of the preset AI algorithm for the two samples is consistent and there is no difference. However, for the estimated fault type group [intake system normal, intake volume slightly higher but still normal] corresponding to the intake volume data item group, although the two estimated results both indicate that there is no serious fault in the intake system, there is a certain difference (one is normal, the other is slightly higher but still normal). This difference reflects the inconsistency of the preset AI algorithm in the two related data samples when predicting the fault of the intake volume data item. In order to quantify this inconsistency and calculate the cost of defect detection, the computer system can adopt a variety of methods. One possible method is to define a difference metric function. Assume that for each estimated fault type group corresponding to the data item group, different estimated fault types are converted into numerical forms according to a certain rule, for example, "engine normal operation" is represented as 0, "intake volume slightly higher but still normal" is represented as 0.1, etc. Then, for each data item group, the difference between the two estimated fault type values ​​can be calculated. For example, for the intake air volume data item group, the difference is 0.1-0=0.1.

[0179] For all data item groups, the computer system calculates the respective differences and then integrates them in some way, such as averaging or weighted average, to obtain the defect detection cost. If the differences of all data item groups are very small, it means that the preset AI algorithm has a high consistency between different data samples when predicting faults in the data sample library of the ECU repair instrument, and the defect detection cost is low; conversely, if there are many data item groups with large differences, it indicates that the algorithm's estimation consistency is poor and the defect detection cost is high.

[0180] For example, suppose the computer system calculates the difference between multiple groups of data items and finds the average value to obtain a defect detection cost of 0.05. This value indicates that overall, the fault types predicted by the preset AI algorithm for the data in the ECU repair instrument data sample library have certain differences between different data samples, but the degree of difference is relatively small. If this value is 0.2, it means that the degree of difference is large, which may mean that the preset AI algorithm has some problems in fault prediction and needs further optimization, such as adjusting the algorithm parameters or improving the algorithm structure, etc., to improve the accuracy and consistency of fault prediction, thereby reducing the defect detection cost.

[0181] In one implementation, step S143, combining the estimated fault type group corresponding to each data item group, performs estimated fault error analysis on the ECU repair instrument data sample library to obtain the defect detection cost, including:

[0182] Step S1431: performing fault type group distinction analysis on the estimated fault type group corresponding to each data item group to obtain the estimated difference corresponding to each data item group;

[0183] Step S1432: Averaging multiple estimated differences corresponding to multiple data item groups to obtain defect detection costs.

[0184] In step S1431, the fault type group distinction analysis is performed for the estimated fault type group corresponding to each data item group. Taking the automobile engine ECU repair instrument data as an example, the operation is performed based on the previously constructed data item group and the determined estimated fault type group.

[0185] Assume that there is a data item group about engine speed, and its corresponding estimated fault type group is [engine normal operation, engine normal operation]. For this estimated fault type group, the computer system performs fault type group distinction analysis. Since both estimated fault types are normal engine operation, in this case, the computer system may determine the estimated difference through some predefined rules. For example, a simple numerical measurement method can be defined. When the two estimated fault types are exactly the same, the estimated difference is 0, indicating that in this data item group, for the engine speed, there is no difference in the preset AI algorithm's estimates in the two related data samples.

[0186] Consider another intake volume data item group, whose corresponding estimated fault type group is [intake volume slightly high but normal, intake system normal]. The computer system needs to parse these two different estimated fault types to determine the estimated difference. This may involve the semantic understanding of fault types and the quantification of the differences between them. For example, the difference can be determined based on the hierarchical relationship and semantic similarity of the fault types in the entire fault classification system. If in the predefined fault classification system, "intake volume slightly high but normal" and "intake system normal" both indicate that there is no serious fault in the intake system, but because "intake volume slightly high but normal" indicates that the intake volume is in a relatively special state, it is somewhat different from the pure "intake system normal". The computer system can quantify this difference into a numerical value, such as 0.2, according to pre-set rules, indicating that the estimated difference corresponding to this data item group is 0.2.

[0187] This kind of fault type group differentiation analysis needs to take into account the complexity and diversity of automotive ECU fault types. Different fault types may differ in severity, scope of impact, and probability of fault occurrence. For example, for fuel injection system failures, the two estimated fault types of "injection time is slightly longer but still works normally" and "partial blockage of the injector nozzle leads to abnormal injection" are both related to the injection system, but their differences need to be accurately quantified. The computer system may need to refer to a large amount of automotive ECU fault knowledge, maintenance experience, and pre-established fault classification models to perform this analysis operation.

[0188] In step S1432, after obtaining the estimated difference corresponding to each data item group, the computer system needs to average the multiple estimated differences corresponding to the multiple data item groups to obtain the defect detection cost.

[0189] Continuing with the above-mentioned engine speed and intake volume data item groups as an example, assume that in addition to these two data item groups, there are other data item groups, such as the throttle opening data item group corresponding to an estimated difference of 0.1, the oxygen sensor feedback voltage data item group corresponding to an estimated difference of 0.15, and so on.

[0190] The computer system averages the estimated differences corresponding to all these data item groups. Assuming there are n data item groups, and the corresponding estimated differences are d1, d2,…, dn, the defect detection cost calculation formula is: Defect detection cost = (d1+d2+…+dn) / n.

[0191] For example, if there are 5 data item groups, and the estimated differences are 0, 0.2, 0.1, 0.15, and 0.05, respectively, then the defect detection cost = (0+0.2+0.1+0.15+0.05) / 5 = 0.1. This defect detection cost of 0.1 represents the average difference in each data item group when the preset AI algorithm performs fault prediction on different data samples in the entire ECU repair instrument data sample library.

[0192] This defect detection cost reflects the degree of distinction between the fault classification prediction information corresponding to the same vehicle fault in the sample classification prediction information in the ECU repair instrument data sample library. If the defect detection cost is high, it means that when the preset AI algorithm performs fault classification prediction on the data sample, the prediction results between different data samples are quite different, that is, the consistency of the algorithm is poor. For example, in an automobile repair scenario, if the defect detection cost is high, it may mean that when faced with similar automobile ECU data, the preset AI algorithm may give fault prediction results with large differences, which is not conducive to accurate fault detection. On the contrary, if the defect detection cost is low, such as close to 0, it means that the algorithm has better prediction consistency between different data samples, and can more stably classify and predict automobile ECU faults.

[0193] In one implementation, step S12, the ECU repair instrument data sample library is passed into a preset AI algorithm for fault detection, and an implicit representation of samples corresponding to the ECU repair instrument data sample library and sample classification estimation information corresponding to the ECU repair instrument data sample library are obtained, including:

[0194] Step S121: transferring the ECU repair instrument data sample library to the fault implicit representation extraction operator in the preset AI algorithm to extract the fault implicit representation and obtain the sample implicit representation;

[0195] Step S122: The implicit representation of the sample is passed into the fault prediction equation in the preset AI algorithm for fault prediction to obtain sample classification prediction information.

[0196] In step S121, the ECU repair instrument data sample library is a collection of many automobile ECU related data samples, which cover various parameter information of automobile electronic control units (ECUs) in different states (such as normal operation, different fault states, etc.). The preset AI algorithm is a pre-designed and constructed artificial intelligence algorithm that aims to analyze and process the ECU repair instrument data to realize the fault detection function. The fault implicit representation extraction operator is an important component of the preset AI algorithm. Taking the neural network algorithm as an example, this operator is similar to the function of the hidden layer part in the neural network. When the computer system passes the ECU repair instrument data sample library to the fault implicit representation extraction operator in the preset AI algorithm, it performs complex mathematical transformation operations on each data sample. For example, for the data sample of the automobile engine ECU, it contains multiple parameters such as engine speed, intake volume, injection time, throttle opening, etc. The fault implicit representation extraction operator will process according to the intrinsic relationship between these parameters and the preset algorithm rules. Assume that in the fault implicit representation extraction operator of the neural network structure, the parameters of each data sample are used as the values ​​of the input nodes, and after the weighted summation of the hidden layer neurons and the action of the activation function, a new vector representation is obtained. This vector representation is the sample implicit representation. For example, the original parameter values ​​of an engine ECU data sample are engine speed 1500 rpm, intake volume 20 cubic meters / hour, injection time 3 milliseconds, and throttle opening 20%. After being processed by the fault implicit representation extraction operator, the obtained sample implicit representation may be a multidimensional vector, such as [0.2, 0.5, -0.1, 0.3] (the values ​​here are only examples). Each element in this vector does not directly correspond to the original engine parameters, but is an implicit representation after algorithm abstraction and transformation, which contains the complex relationship between these original parameters and the potential connection with the engine failure mode.

[0197] Different fault states may present different characteristics in the sample implicit representation. For example, when the engine has a blocked injector fault, the sample implicit representation obtained by the fault implicit representation extraction operator may be significantly different from the sample implicit representation in the normal state in some dimensions. This difference can be used in subsequent fault detection analysis to identify whether the engine has a blocked injector fault and other possible fault types.

[0198] In step S122, after obtaining the implicit representation of the sample, the computer system will pass it to the fault prediction equation in the preset AI algorithm. The fault prediction equation is also an important part of the preset AI algorithm, and its function is to estimate the fault classification of each data sample based on the implicit representation of the sample.

[0199] Continuing with the above engine ECU data as an example, after the fault prediction module receives the implicit representation of the sample (such as [0.2, 0.5, -0.1, 0.3]), it processes it according to the predefined classification rules and algorithm logic. In the preset AI algorithm, different fault type categories may be pre-set, such as injection system fault, intake system fault, ignition system fault, and no fault.

[0200] The fault prediction function may use a probability-based method to predict faults. For example, it will calculate the probability that this sample implicitly represents each fault type category. Assume that after calculation, the probability that this sample implicitly represents a fuel injection system fault is 0.1, the probability that it belongs to an intake system fault is 0.05, the probability that it belongs to an ignition system fault is 0.03, and the probability that it is not faulty is 0.82. Then the sample classification prediction information obtained is a vector of the probabilities of each fault type, that is, [0.1, 0.05, 0.03, 0.82].

[0201] This sample classification prediction information reflects the probability that the data sample belongs to different fault types according to the current sample implicit representation. This information has important reference value for automobile maintenance and fault diagnosis. For example, if the sample classification prediction information of a data sample shows that the probability of belonging to the injection system fault is high (such as 0.6), while the probability of belonging to other fault types and no fault is low, then the maintenance personnel can focus on checking whether there is a problem with the injection system.

[0202] At the same time, in actual applications, the preset AI algorithm may continuously learn and optimize the classification rules of the fault prediction equation. For example, by collecting more automotive ECU data samples and combining them with actual fault diagnosis results to adjust the method of calculating probability or optimize the classification threshold. This can improve the accuracy of fault prediction and make the sample classification prediction information more reliably reflect the actual fault status of the automotive ECU.

[0203] To summarize, sub-steps S121-S122 of step S12 sequentially transfer the ECU repair instrument data sample library to the fault implicit representation extraction operator and fault prediction operator in the preset AI algorithm through the computer system to obtain sample implicit representation and sample classification prediction information. These two results provide important basis and foundation for subsequent fault detection algorithm training (such as calculating data implicit representation cost, defect detection cost, etc.) and final fault detection.

[0204] Based on the foregoing embodiments, the embodiments of the present application provide a data processing device for an automobile ECU repair instrument. The units included in the device and the modules included in each unit can be implemented by a processor in a computer device; of course, they can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0205] Figure 3 A schematic diagram of the composition structure of a data processing device for an automobile ECU repair instrument provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the automobile ECU repair instrument data processing device 200 includes:

[0206] The data acquisition module 210 is used to acquire the ECU repair instrument data;

[0207] The fault detection module 220 is used to input the ECU repair instrument data into a fault detection algorithm for fault detection, and obtain a fault detection result corresponding to the ECU repair instrument data.

[0208] Among them, the fault detection algorithm is obtained by optimizing the preset AI algorithm in combination with the first algorithm cost and the second algorithm cost, the first algorithm cost is at least one of the data implicit representation cost and the defect detection cost, the data implicit representation cost represents the distinction between the fault characterization information corresponding to the same vehicle fault in the ECU repair instrument data sample library in the sample implicit representation, the defect detection cost represents the distinction between the fault classification prediction information corresponding to the same vehicle fault in the ECU repair instrument data sample library in the sample classification prediction information, the sample implicit representation and the sample classification prediction information are obtained by passing the ECU repair instrument data sample library into the preset AI algorithm for fault detection, and the second algorithm cost is calculated in combination with the sample classification prediction information and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library.

[0209] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiment of the present application can be used to execute the method described in the above method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding. It should be noted that in the embodiment of the present application, if the above-mentioned automobile ECU repair instrument data processing method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0210] An embodiment of the present application provides a computer system, including a memory and a processor, wherein the memory stores a computer program that can be executed on the processor, and when the processor executes the program, some or all of the steps in the above method are implemented.

[0211] Figure 4 A hardware entity diagram of a computer system provided in an embodiment of the present application is shown in FIG. Figure 4 As shown, the hardware entity of the computer system 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can be run on the processor 1001, and the processor 1001 implements the steps in the method of any of the above embodiments when executing the program.

Claims

1. A data processing method for an automobile ECU repair instrument, characterized in that: The method comprises: Get ECU repair instrument data; The ECU repair instrument data is input into a fault detection algorithm for fault detection, and a fault detection result corresponding to the ECU repair instrument data is obtained; Wherein, the fault detection algorithm is obtained by optimizing the preset AI algorithm in combination with the first algorithm cost and the second algorithm cost, the first algorithm cost is at least one of the data implicit representation cost and the defect detection cost, the data implicit representation cost represents the degree of distinction between the fault characterization information corresponding to the same vehicle fault in the sample implicit representation in the ECU repair instrument data sample library, the defect detection cost represents the degree of distinction between the fault classification prediction information corresponding to the same vehicle fault in the ECU repair instrument data sample library in the sample classification prediction information, the sample implicit representation and the sample classification prediction information are obtained by transferring the ECU repair instrument data sample library into the preset AI algorithm for fault detection, and the second algorithm cost is calculated in combination with the sample classification prediction information and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library; The fault detection algorithm is trained by the following steps: Obtain the ECU repair instrument data sample library and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library; The ECU repair instrument data sample library is passed into the preset AI algorithm for fault detection, and the sample implicit representation corresponding to the ECU repair instrument data sample library and the sample classification estimation information corresponding to the ECU repair instrument data sample library are obtained; In combination with the sample implicit representation, an implicit representation error analysis is performed on the ECU repair instrument data sample library to obtain the data implicit representation cost; Combined with the sample classification prediction information, the prediction error analysis is performed on the ECU repair instrument data sample library to obtain the defect detection cost; Determine the second algorithm cost by combining the sample classification prediction information and the fault classification label; The preset AI algorithm is trained in combination with the first algorithm cost and the second algorithm cost to obtain the fault detection algorithm.

2. The method according to claim 1, characterized in that The implicit representation of the sample is combined with the implicit representation, and the implicit representation error of the ECU repair instrument data sample library is analyzed to obtain the implicit representation cost of the data, including: Splitting each ECU repair instrument data sample in the ECU repair instrument data sample library into data items to obtain a plurality of sample data items corresponding to the ECU repair instrument data sample library; Arrange the plurality of sample data items to obtain at least one arranged data item; Determine that each of the sorted data items corresponds to at least one piece of involved ECU repair instrument data, wherein the at least one piece of involved ECU repair instrument data corresponding to each of the sorted data items is an ECU repair instrument data sample having each of the sorted data items in the ECU repair instrument data sample library; In combination with at least one involved ECU repair instrument data corresponding to each of the sorted data items, determine at least one involved implicit representation corresponding to each of the sorted data items, and a set of sorted implicit representations corresponding to the at least one sorted data items; any one involved implicit representation corresponding to each of the sorted data items is fault characterization information corresponding to each of the sorted data items in any one involved ECU repair instrument data corresponding to each of the sorted data items; the set of sorted implicit representations includes at least one involved implicit representation corresponding to each of the at least one sorted data items; In combination with the at least one involved implicit representation corresponding to each of the sorted data items, an involved implicit representation set corresponding to each involved implicit representation is determined; the involved implicit representation set corresponding to any one of the involved implicit representations includes involved implicit representations other than any one of the involved implicit representations in the at least one involved implicit representation corresponding to the target sorted data item; the target sorted data item is a sorted data item corresponding to any one of the involved implicit representations in the at least one sorted data items; In combination with the involved implicit representation set corresponding to each involved implicit representation and the sorted implicit representation set, implicit representation similarity analysis is performed on the at least one sorted data item to obtain the data implicit representation cost.

3. The method according to claim 2, characterized in that The step of combining the involved implicit representation set corresponding to each involved implicit representation and the sorted implicit representation set, performing implicit representation similarity analysis on the at least one sorted data item, and obtaining the data implicit representation cost, comprises: Processing multiple involved implicit representations in the sorted implicit representation set one by one; Determine, in combination with the sorted implicit representation set, a plurality of first involved implicit representations corresponding to the currently processed involved implicit representation; the plurality of first involved implicit representations are involved implicit representations in the sorted implicit representation set other than the currently processed involved implicit representation; Selecting a target implicit representation set corresponding to the currently processed implicit representation from the implicit representation set corresponding to each implicit representation; the target implicit representation set includes a plurality of second implicit representations; Performing commonality measurement on the currently processed implicit representation and each first implicit representation to obtain a first commonality measurement value corresponding to each first implicit representation; the first commonality measurement value corresponding to each first implicit representation indicates the matching degree between each first implicit representation and the currently processed implicit representation; Performing commonality measurement on the currently processed implicit representation and each second implicit representation to obtain a second commonality measurement value corresponding to each second implicit representation; the second commonality measurement value corresponding to each second implicit representation indicates the degree of match between each second implicit representation and the currently processed implicit representation; Determine a third algorithm cost corresponding to the currently processed implicit representation by combining the second commonality metric value corresponding to each second implicit representation and the first commonality metric values ​​corresponding to each of the plurality of first implicit representations; At the end of the processing, the multiple third algorithm costs obtained during the processing are averaged to obtain the data implicit representation cost.

4. The method according to claim 3, characterized in that The determining of a third algorithm cost corresponding to the currently processed implicit representation by combining the second commonality metric value corresponding to each second implicit representation and the first commonality metric values ​​corresponding to each of the plurality of first implicit representations comprises: Determine an estimated support coefficient corresponding to each second implicit representation by combining the second commonality metric value corresponding to each second implicit representation and the first commonality metric values ​​corresponding to each of the plurality of first implicit representations, wherein the estimated support coefficient corresponding to each second implicit representation indicates the degree of possibility of association between each second implicit representation and the currently processed implicit representation; Determine, in combination with the estimated support coefficient corresponding to each second involved implicit representation, a fourth algorithm cost corresponding to each second involved implicit representation, wherein the fourth algorithm cost represents the degree of distinction between each second involved implicit representation and the currently processed involved implicit representation; Accumulating fourth algorithm costs corresponding to each of the plurality of second involved implicit representations to obtain the third algorithm cost; The arranging the multiple sample data items to obtain at least one arranged data item includes: Comparing each sample data item with the previously deployed ECU repair instrument data to obtain comparison results corresponding to the multiple sample data items; In combination with the comparison result, the sample data items belonging to the pre-deployed ECU repair instrument data are filtered out from the plurality of sample data items to obtain the filtered sample data items; Duplicate items are cleaned up for the filtered sample data items to obtain the at least one sorted data item.

5. The method according to claim 4, characterized in that After performing duplicate clearing on the filtered sample data items to obtain the at least one sorted data item, the method further includes: Determining at least one main sorted data item among the at least one sorted data item; the existence ratio of any one of the at least one main sorted data item in the data sample library of the ECU repair instrument is greater than a set ratio; The step of determining that each sorted data item corresponds to at least one piece of data related to the ECU repair instrument includes: Make sure that each main sorted data item corresponds to at least one piece of related ECU repair instrument data.

6. The method according to claim 5, characterized in that The obtaining of the ECU repair instrument data sample library and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library includes: Acquire multiple first ECU repair instrument data samples from a pre-deployed ECU repair instrument data set; Copy each first ECU repair instrument data sample to obtain a second ECU repair instrument data sample corresponding to each first ECU repair instrument data sample; Combining the plurality of first ECU repair instrument data samples and the second ECU repair instrument data samples corresponding to the plurality of first ECU repair instrument data samples, constructing the ECU repair instrument data sample library; Determine the fault classification label corresponding to the ECU repair instrument data sample library in combination with the fault classification label corresponding to the pre-deployed ECU repair instrument data set; The method of transferring the ECU repair instrument data sample library into the preset AI algorithm for fault detection, obtaining the sample implicit representation corresponding to the ECU repair instrument data sample library, and the sample classification estimation information corresponding to the ECU repair instrument data sample library, includes: The ECU repair instrument data sample library is transferred to the fault implicit representation extraction operator in the preset AI algorithm to extract the fault implicit representation, so as to obtain the sample implicit representation; The sample implicit representation is transmitted to the fault prediction module in the preset AI algorithm for fault prediction to obtain the sample classification prediction information.

7. The method according to claim 6, characterized in that The method of combining the sample classification prediction information and performing prediction error analysis on the ECU repair instrument data sample library to obtain the defect detection cost includes: Determine a plurality of data item groups, wherein any one of the data item groups includes a first data item and a second data item, wherein the first data item is any one of the first ECU repair instrument data samples, and the second data item is a data item corresponding to the first data item in a third ECU repair instrument data sample, and the third ECU repair instrument data sample is a second ECU repair instrument data sample corresponding to the any one of the first ECU repair instrument data samples; Determine an estimated fault type group corresponding to each data item group; the estimated fault type group corresponding to each data item group includes the fault classification estimation information corresponding to the first data item group in each data item group and the fault classification estimation information corresponding to the second data item group in each data item group; Combined with the estimated fault type group corresponding to each data item group, the estimated fault error analysis is performed on the ECU repair instrument data sample library to obtain the defect detection cost.

8. The method according to claim 7, characterized in that The step of combining the estimated fault type group corresponding to each data item group and performing estimated fault error analysis on the ECU repair instrument data sample library to obtain the defect detection cost includes: Performing fault type group distinction analysis on the estimated fault type group corresponding to each data item group to obtain the estimated difference corresponding to each data item group; The defect detection cost is obtained by averaging a plurality of estimated differences corresponding to the plurality of data item groups.

9. A data processing device for an automobile ECU repair instrument implementing the method according to claim 1, characterized in that: The device comprises: Data acquisition module, used to obtain ECU repair instrument data; A fault detection module, used for transferring the ECU repair instrument data into a fault detection algorithm for fault detection, and obtaining a fault detection result corresponding to the ECU repair instrument data; Among them, the fault detection algorithm is obtained by optimizing the preset AI algorithm in combination with the first algorithm cost and the second algorithm cost, the first algorithm cost is at least one of the data implicit representation cost and the defect detection cost, the data implicit representation cost represents the distinction between the fault characterization information corresponding to the same vehicle fault in the ECU repair instrument data sample library in the sample implicit representation, the defect detection cost represents the distinction between the fault classification prediction information corresponding to the same vehicle fault in the ECU repair instrument data sample library in the sample classification prediction information, the sample implicit representation and the sample classification prediction information are obtained by passing the ECU repair instrument data sample library into the preset AI algorithm for fault detection, and the second algorithm cost is calculated in combination with the sample classification prediction information and the fault classification label corresponding to each ECU repair instrument data sample in the ECU repair instrument data sample library.

Images