Vehicle-mounted electronic controller real-time fault diagnosis method and system based on edge computing

Abstract

The present application relates to the technical field of vehicle-mounted intelligent monitoring, in particular to a real-time fault diagnosis method and system for a vehicle-mounted electronic controller based on edge computing, in which, a multi-dimensional time sequence operation data matrix is reorganized in real time by using an edge computing architecture and spatial numerical decomposition is performed, the fluctuation amplitude of an electrical characteristic bottom layer calculation is directly and deeply calculated and a target abnormal state disturbance matrix is separated, the masking effect of conventional operation background noise on real hidden fault signals is effectively avoided, on this basis, a state observation sequence is reconstructed by extracting multi-dimensional features such as current spikes and voltage fluctuations, a state transition correlation between hidden features and potential failure risks is deeply mined by combining a dynamic probability deduction mechanism, the highest frequency operation occurrence probability under hidden classification is accurately interpreted, complex time sequence electrical fluctuations are quickly converted into intuitive current operation state judgments, customized warning messages are directly mapped and generated and a bus level terminal interaction is triggered, and the identification sensitivity and early warning real-time performance of a small abnormal state are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of vehicle-mounted intelligent monitoring technology, and in particular to a real-time fault diagnosis method and system for vehicle-mounted electronic controllers based on edge computing. Background Technology

[0002] The field of vehicle-mounted intelligent monitoring technology mainly involves the technical content of continuous collection, analysis and evaluation of vehicle operating status. This field monitors and manages the operation of key vehicle components and systems by integrating sensors, embedded control units and data processing modules. Its main purpose is to ensure that the vehicle's status is perceptible and information is available during operation. It is widely used in industries such as automotive electronics, intelligent connected vehicles and transportation equipment.

[0003] Among them, the real-time fault diagnosis method for vehicle electronic controllers based on edge computing refers to a method that uses data acquisition and processing mechanisms to analyze and judge the operating data of the controller in response to possible abnormalities or faults that may occur during the operation of the vehicle electronic controller. The traditional vehicle electronic controller fault diagnosis method refers to a technical solution that uses vehicle diagnostic systems and related testing equipment to acquire controller status information and combines rule judgment or model analysis methods to identify and locate fault types.

[0004] Traditional vehicle electronic controller fault diagnosis methods rely on on-board diagnostic systems and external testing equipment to obtain one-way operating status information. They rely on fixed rules and offline models to carry out passive post-event analysis. The existing operating mode shows obvious limitations when dealing with complex and ever-changing actual vehicle operating conditions. Due to the lack of a deep time-series correlation and fusion mechanism for the data acquisition of multi-dimensional electrical physical quantities, transient disturbances and normal operating condition noises inside the control system are easily superimposed, making it impossible to effectively isolate minute abnormal fluctuation characteristics. When faced with hidden transient current impacts or continuous slight voltage abnormalities, fixed rules are prone to missed detections and false alarms. At the same time, the process response based solely on threshold judgment has serious lag and cannot dynamically capture hidden risk states and provide immediate diagnostic results before the hardware suffers irreversible damage. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide a real-time fault diagnosis method for vehicle electronic controllers based on edge computing, comprising the following steps: To achieve the above objectives, the present invention adopts the following technical solution: a real-time fault diagnosis method for vehicle electronic controllers based on edge computing, comprising the following steps: S1: During the continuous operation cycle of the vehicle electronic controller, multiple data such as the command current, actual drive current and power supply ripple voltage of the vehicle electronic controller are obtained through edge computing nodes to construct a multi-dimensional time-series operation data matrix. S2: Based on the comparison results between the fluctuation amplitude of each data item in the multi-dimensional time-series running data matrix and the steady-state threshold, determine the initial abnormal state disturbance matrix, determine the spatial reconstruction error of the initial abnormal state disturbance matrix, and generate the target abnormal state disturbance matrix. S3: Based on the target abnormal state disturbance matrix, determine multiple characteristics such as current peak frequency, voltage fluctuation frequency and steady-state control deviation, and combine them to form a state observation sequence; S4: Based on the state observation sequence and preset probability parameters, determine the probability of occurrence of operation under various hidden state categories, and extract the name of the current running state with the first probability of occurrence arranged in descending order; S5: Generate a fault warning message based on the current operating status name, and use the fault warning message as the real-time fault diagnosis result of the vehicle electronic controller.

[0006] As a further aspect of the present invention, the multi-dimensional time-series operation data matrix includes a command current time-series item, an actual drive current time-series item, and a power supply ripple voltage time-series item; the target abnormal state disturbance matrix includes a command current disturbance component, an actual drive current disturbance component, and a power supply ripple voltage disturbance component; the state observation sequence includes current peak frequency feature elements, voltage fluctuation frequency feature elements, and steady-state control deviation feature elements; the current operating state name is specifically a hidden state identifier calculated by extracting the probability of occurrence of the first operating state; and the real-time fault diagnosis result is specifically a fault warning message calculated by determining the abnormal category of the current operating state name.

[0007] As a further aspect of the present invention, step S1 specifically comprises: S101: By monitoring multiple data points such as command current, actual drive current, and power supply ripple voltage of the vehicle electronic controller during continuous operation cycle through edge computing nodes, the corresponding timestamps are obtained and recorded. The time dimension of each data item is aligned according to the timestamps to obtain the initial time-series aligned data sequence. S102: Normalize each data item in the initial time-aligned data sequence to establish a normalized runtime sequence feature set; S103: Extract each normalized data item from the normalized runtime sequence feature set, and perform array structure splicing and combination according to the time sequence dimension to construct a multi-dimensional time sequence runtime data matrix.

[0008] As a further aspect of the present invention, step S2 specifically comprises: S201: The normalized command current, actual driving current and power supply ripple voltage in the multi-dimensional time-series running data matrix are numerically decomposed into matrix space structure by principal component analysis algorithm. The fluctuation amplitude corresponding to each data item is calculated and compared with the corresponding preset steady-state threshold to obtain the fluctuation amplitude comparison classification result. S202: Extract the fluctuation amplitude comparison and classification results, and classify the multiple data that have not reached the corresponding steady-state threshold value into the stable control baseline matrix, and classify the multiple data that have reached or exceeded the corresponding steady-state threshold into the initial abnormal state disturbance matrix. S203: Calculate the spatial reconstruction error between the data in the multidimensional time-series running data matrix, the baseline data in the stable control baseline matrix, and the disturbance data in the initial abnormal state disturbance matrix, and separate the target abnormal state disturbance matrix.

[0009] As a further aspect of the present invention, step S3 specifically comprises: S301: Obtain the normalized disturbance actual driving current in the target abnormal state disturbance matrix within the set time window, compare the normalized disturbance actual driving current with the preset current tolerance limit value corresponding to the vehicle electronic controller, count the number of times the current peak frequency is exceeded. S302: Obtain the normalized disturbance power supply ripple voltage and normalized disturbance command current in the target abnormal state disturbance matrix within the same time window, count the number of times the normalized disturbance power supply ripple voltage exceeds the preset voltage fluctuation limit, obtain the corresponding voltage fluctuation frequency, and at the same time calculate the absolute value of the difference between the normalized disturbance command current and the normalized disturbance actual drive current to obtain the steady-state control deviation characteristics. S303: Extract the voltage fluctuation frequency, the corresponding current peak frequency, and the corresponding steady-state control deviation characteristics, and arrange them in chronological order to generate a state observation sequence.

[0010] As a further aspect of the present invention, step S4 specifically comprises: S401: Summarize the preset state transition probability parameters and observation probability parameters, as well as the state observation sequence, to establish a probability operation data set; S402: Using the Hidden Markov Model, calculate the probability of the state observation sequence being in multiple preset hidden state classifications, and obtain the set of hidden classification operation probabilities; S403: In the set of hidden category operation probabilities, each operation probability is sorted in descending order, and the identifier of the corresponding hidden state category associated with the first operation probability in the sorted list is selected as the name of the current operation state.

[0011] As a further aspect of the present invention, step S5 specifically comprises: S501: Read the baseline classification information in the relevant status category determination rule library of the vehicle electronic controller, compare the current operating status name with the abnormal classification condition determination label in the status category determination rule library, determine whether the current operating status name meets the corresponding abnormal classification, and obtain the status abnormal determination result. S502: Based on the state anomaly determination result, confirm the anomaly category to which the current running state name belongs, and use it as the corresponding filling content for the anomaly warning information data. Encapsulate the filling content into the specified communication protocol format message data packet for information hierarchical combination and encapsulation to generate a fault early warning message. S503: The fault warning message to be processed is sent to the relevant vehicle display terminal for warning interaction through the edge computing node, and the communication link transmission status is confirmed by the feedback of the edge computing node on the transmission of the fault warning message, which serves as the real-time fault diagnosis result of the vehicle electronic controller.

[0012] A real-time fault diagnosis system for vehicle electronic controllers based on edge computing includes: The matrix construction module acquires multiple data points of the vehicle electronic controller, including command current, actual drive current, and power supply ripple voltage, through edge computing nodes during the continuous operation cycle of the vehicle electronic controller, and constructs a multi-dimensional time-series operation data matrix. The disturbance matrix partitioning module determines the initial abnormal state disturbance matrix based on the comparison results between the fluctuation amplitude of each data item in the multi-dimensional time-series running data matrix and the steady-state threshold, determines the spatial reconstruction error of the initial abnormal state disturbance matrix, and generates the target abnormal state disturbance matrix. The observation sequence combination module determines multiple characteristics such as current peak frequency, voltage fluctuation frequency and steady-state control deviation based on the target abnormal state disturbance matrix, and combines them to form a state observation sequence. The running state name determination module determines the probability of running occurrence under various hidden state categories based on the state observation sequence and preset probability parameters, and extracts the current running state name with the first running occurrence probability arranged in descending order. The diagnostic result generation module generates a fault warning message based on the current operating status name, and uses the fault warning message as the real-time fault diagnosis result of the vehicle electronic controller.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, an edge computing architecture is used to reconstruct a multi-dimensional time-series operational data matrix in real time and perform spatial numerical decomposition. This directly delves into the underlying electrical characteristics to calculate fluctuation amplitudes and separate the target abnormal state disturbance matrix, effectively avoiding the masking effect of normal operating background noise on real hidden fault signals. Based on this, multi-dimensional features such as current spikes and voltage fluctuations are extracted to reconstruct the state observation sequence. Combined with a dynamic probability inference mechanism, the state transition correlation between hidden features and potential failure risks is deeply mined, accurately interpreting the probability of the highest frequency operation under the hidden classification. Complex time-series electrical fluctuations are quickly transformed into intuitive current operating state judgments, directly mapping and generating customized warning messages and triggering bus-level terminal interactions, significantly improving the sensitivity of identifying minor abnormal states and the real-time performance of early warnings. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0017] Please see Figure 1 This invention provides a real-time fault diagnosis method for vehicle electronic controllers based on edge computing, comprising the following steps: S1: During the continuous operation cycle of the vehicle electronic controller, multiple data such as command current, actual drive current and power supply ripple voltage of the vehicle electronic controller are obtained through edge computing nodes to construct a multi-dimensional time-series operation data matrix. S2: Based on the comparison results between the fluctuation amplitude of each data item in the multi-dimensional time series running data matrix and the steady-state threshold, determine the initial abnormal state disturbance matrix, determine the spatial reconstruction error of the initial abnormal state disturbance matrix, and generate the target abnormal state disturbance matrix. S3: Based on the target abnormal state disturbance matrix, determine multiple characteristics such as current peak frequency, voltage fluctuation frequency and steady-state control deviation, and combine them to form a state observation sequence; S4: Based on the state observation sequence and preset probability parameters, determine the probability of occurrence of operation under various hidden state categories, and extract the name of the current running state with the first probability of occurrence arranged in descending order; S5: Generate a fault warning message based on the current operating status name, and use the fault warning message as the real-time fault diagnosis result of the vehicle electronic controller.

[0018] The multi-dimensional time-series operation data matrix includes command current time-series items, actual drive current time-series items, and power supply ripple voltage time-series items. The target abnormal state disturbance matrix includes command current disturbance components, actual drive current disturbance components, and power supply ripple voltage disturbance components. The state observation sequence includes current spike frequency feature elements, voltage fluctuation frequency feature elements, and steady-state control deviation feature elements. The current operating state name is specifically a hidden state identifier calculated by extracting the probability of occurrence of the first operating state. The real-time fault diagnosis result is specifically a fault warning message calculated by determining the abnormal category of the current operating state name.

[0019] Please see Figure 2 Step S1 is as follows: S101: By monitoring multiple data points such as command current, actual drive current, and power supply ripple voltage of the vehicle electronic controller during continuous operation cycle through edge computing nodes, the corresponding timestamps are obtained and recorded. The time dimension of each data item is aligned according to the timestamps to obtain the initial time-series aligned data sequence. The C++ embedded real-time computing software environment deployed within the edge computing node aligns multiple monitoring data points from the vehicle's electronic controller in the time dimension. The edge computing node calls its internally integrated analog-to-digital converter (ADC) hardware interface, which converts the continuously changing analog electrical signals from physical sensors into discrete, high-precision floating-point digital signals. The edge computing node establishes a low-level electrical connection with various physical sensors of the vehicle's electronic controller, setting a fixed sampling frequency of 1000 Hz (1 millisecond per sampling cycle). It performs a cyclic reading operation on the data within the continuous operating cycle, extracting the command current value, the actual drive current value, and the power supply ripple voltage value. For each data value acquired through analog-to-digital conversion, the edge computing node retrieves the current count value from its internal high-precision hardware system clock. This count value is used as a timestamp to record the time of a specific event and is bound and encapsulated with the corresponding data value, forming independent command current data records, actual drive current data records, and power supply ripple voltage data records. During the timestamp alignment phase, edge computing nodes extract the start timestamps from the instruction current timestamp sequence, the actual drive current timestamp sequence, and the power supply ripple voltage timestamp sequence, respectively, and select the largest time value among the three start timestamps as the reference start timestamp. They also extract the end timestamps from these three timestamp sequences and select the smallest time value as the reference end timestamp. Within the valid time interval defined by the reference start timestamp and reference end timestamp, a standard time axis sequence is generated with a standard time step of 1 millisecond. Each standard timestamp in the standard time axis sequence is traversed. If a data point that perfectly matches the standard timestamp is found in the original record, the corresponding data value is directly read; otherwise, if no data point that perfectly matches the standard timestamp is found in the original record, the linear interpolation algorithm module is invoked. In the original record, find the nearest previous valid timestamp to the left of the standard timestamp and its corresponding data value, and find the nearest next valid timestamp to the right of the standard timestamp and its corresponding data value. Calculate the difference between the standard timestamp value and the previous valid timestamp value, divide the next valid timestamp value by the difference between the previous and next valid timestamp values, and then multiply by the difference between their corresponding data values ​​to obtain the correction increment value. Add the correction increment value to the data value corresponding to the previous valid timestamp to obtain the completed data value. For all standard timestamps, the edge computing node obtains uniformly aligned command current values, actual drive current values, and power supply ripple voltage values, arranges and summarizes them in chronological order, and generates an initial timing-aligned data sequence.

[0020] S102: Normalize each data item in the initial time-aligned data sequence to establish a normalized runtime sequence feature set; The data is scaled proportionally to a specific floating-point space region. All instruction current values ​​in the initial timing-aligned data sequence are traversed, and the maximum and minimum instruction current values ​​are selected. Using the common practice of range normalization, the original instruction current values ​​are mapped to a normalized floating-point range to obtain normalized instruction current values. Following the same data traversal and extreme value selection logic, the maximum and minimum actual drive current values ​​are obtained from the initial timing-aligned data sequence, and the original actual drive current values ​​are also transformed into normalized actual drive current values ​​through range normalization. The maximum and minimum power supply ripple voltage values ​​are found from the initial timing-aligned data sequence, and the original power supply ripple voltage values ​​are normalized through range normalization to obtain normalized power supply ripple voltage values. After completing range normalization for all time nodes and all data types, the edge computing node encapsulates the normalized instruction current value, normalized actual drive current value, and normalized power supply ripple voltage value at the same time node into independent time feature vectors, which are stored in contiguous memory space in time sequence to establish a normalized runtime sequence feature set.

[0021] S103: Extract each normalized data item from the normalized runtime sequence feature set, and perform array structure splicing and combination according to the time sequence dimension to construct a multi-dimensional time-series runtime data matrix; One-dimensional data sequences are stacked into a multi-dimensional matrix structure arranged in rows. Normalized instruction current values ​​corresponding to all time nodes within the set are extracted and arranged from left to right in ascending order of timestamps, generating a first row of data arrays with a single-dimensional horizontal span. Normalized actual drive current values ​​corresponding to all time nodes within the set are extracted and arranged from left to right in ascending order of timestamps, generating a second row of data arrays. Normalized supply ripple voltage values ​​corresponding to all time nodes within the set are extracted and arranged from left to right following the ascending time axis rule, generating a third row of data arrays. In the memory data structure, the first, second, and third row data arrays are vertically aligned and combined sequentially from top to bottom. By using a three-dimensional vector tiling method where rows represent data dimension types and columns represent time progression nodes, a two-dimensional floating-point grid system containing three data rows and several time node columns is constructed in memory, establishing a multi-dimensional time-series running data matrix.

[0022] Please see Figure 3 Step S2 is as follows: S201: The principal component analysis algorithm is used to perform numerical decomposition of the normalized command current, actual drive current and power supply ripple voltage matrix in the multi-dimensional time-series running data matrix, calculates the fluctuation amplitude corresponding to each data item, and compares it with the corresponding preset steady-state threshold to obtain the fluctuation amplitude comparison classification result. The system utilizes a built-in C++ matrix operation library to perform orthogonal transformations on the multidimensional time-series data matrix using principal component analysis (PCA), projecting correlated three-dimensional variables into uncorrelated two-dimensional principal component spaces. For each row of data, the mean of the first, second, and third rows is calculated by summing the results and dividing by the total number of columns in that row. For each data point in the multidimensional time-series data matrix, a decentralized translation operation is performed by subtracting the mean of the row containing each value, generating a decentralized data matrix. This decentralized data matrix is ​​then multiplied by its transpose to generate an intermediate product matrix. Each element in the intermediate product matrix is ​​divided by the number of columns minus one to obtain the eigenvalue covariance matrix. Eigenvalue decomposition is performed on the eigenvalue covariance matrix to obtain multiple eigenvalues ​​and their corresponding eigenvectors. All eigenvalues ​​are sorted in descending order, and the eigenvectors corresponding to the top two eigenvalues ​​are concatenated column-wise to form the projection transformation matrix. The decentralized data matrix and the projection transformation matrix are multiplied to obtain a two-dimensional reduced feature matrix. The fluctuation amplitude is then calculated using the following formula: The process involves extracting a preset steady-state threshold, offline statistical analysis of all fluctuation amplitudes of the on-board electronic controller during 100,000 consecutive fault-free operations under a standard bench test environment, calculating the arithmetic mean of these 100,000 baseline fluctuation amplitudes as the baseline mean, calculating the baseline standard deviation, and adding three times the baseline standard deviation to the baseline mean to derive the preset steady-state threshold value of 0.35. The process iterates through the fluctuation amplitudes of all calculated time points. When the fluctuation amplitude at a given time point is less than 0.35, the corresponding Boolean judgment state is marked as stable; when the fluctuation amplitude at a given time point is greater than 0.35, the corresponding Boolean judgment state is marked as abnormal fluctuation; and when the fluctuation amplitude at a given time point is equal to 0.35, the corresponding Boolean judgment state is marked as abnormal fluctuation. Finally, the Boolean judgment states and corresponding timestamp information of all time points are integrated to generate a fluctuation amplitude comparison and classification result.

[0023] S202: In the comparison and classification results of fluctuation amplitude, multiple data that have not reached the corresponding steady-state threshold values ​​are assigned to the stable control baseline matrix, and multiple data that have reached or exceeded the corresponding steady-state threshold values ​​are assigned to the initial abnormal state disturbance matrix. Initialize two two-dimensional floating-point grid data structures in memory: the stable control baseline matrix and the initial abnormal state perturbation matrix. Initially, set the number of rows to three and the number of columns to zero. Sequentially read the time node identifiers and their associated Boolean judgment states from the fluctuation amplitude comparison and classification results. When a Boolean judgment state of "stable" is detected for a certain time node, return to the multi-dimensional time-series running data matrix for indexing and location, and extract a complete column of raw normalized values ​​corresponding to that time node. Append this extracted data vector to the last column of the stable control baseline matrix, increasing the total column count of the stable control baseline matrix by 1. When a Boolean judgment state of "abnormal fluctuation" is detected for a certain time node, return to the multi-dimensional time-series running data matrix to extract a complete column of raw normalized values ​​corresponding to that time node. Append this column of values ​​to the last column of the initial abnormal state perturbation matrix, increasing the total column count of the initial abnormal state perturbation matrix by 1. Continue execution until all time nodes in the fluctuation amplitude comparison and classification results have been traversed.

[0024] S203: Calculate the spatial reconstruction error between the data in the multi-dimensional time-series running data matrix, the baseline data in the stable control baseline matrix, and the disturbance data in the initial abnormal state disturbance matrix, and separate the target abnormal state disturbance matrix. Extract the stable control baseline matrix. Multiply this baseline matrix by its transpose to obtain the baseline product matrix. Calculate the inverse of the baseline product matrix. Multiply the transpose of the baseline matrix by the inverse matrix. Then multiply the intermediate matrix by the original baseline matrix to obtain the baseline projection mapping matrix. Initialize a blank target anomaly state perturbation matrix in memory. Extract the column perturbation data vector from the initial anomaly state perturbation matrix. Perform matrix multiplication between the baseline projection mapping matrix and this column perturbation data vector to obtain the baseline projection reconstruction vector. Subtract the values ​​at the same dimension positions in the baseline projection reconstruction vector from the values ​​at each dimension positions in the original column perturbation data vector to obtain a residual vector containing three-dimensional features. Calculate the spatial reconstruction error using the following formula: A baseline reconstruction error set was collected for historical vehicle electronic controllers of the same model when subjected to external electromagnetic interference but without substantial hardware failure. All error values ​​in this set were sorted from largest to smallest, and the 90th percentile value was used as the reconstruction error screening threshold of 0.85. The currently acquired spatial reconstruction error value was compared with 0.85. If the spatial reconstruction error value was greater than 0.85, the disturbance data in that column was determined to contain abnormal characteristics and was added to the target abnormal state disturbance matrix; if the spatial reconstruction error value was less than 0.85, the data in that column was discarded; if the spatial reconstruction error value was equal to 0.85, the data in that column was discarded. This process was repeated iteratively on the remaining columns of the initial abnormal state disturbance matrix to separate the target abnormal state disturbance matrix.

[0025] Please see Figure 4 Step S3 is as follows: S301: Obtain the normalized disturbance actual driving current in the target abnormal state disturbance matrix within the set time window, compare the normalized disturbance actual driving current with the preset current tolerance limit value of the on-board electronic controller, count the number of times the current peak frequency is exceeded. The time window span is set to 50 consecutive data points. Starting from the first column of the target abnormal state disturbance matrix, 50 consecutive data columns are extracted to form the current time window. The second row of the data array is located, and 50 normalized disturbance actual drive current values ​​are extracted. The maximum thermal breakdown current (amperes) of the power transistor, as specified in the vehicle electronic controller hardware design manual, is retrieved. This ampere value is multiplied by a safety derating factor of 0.8 to obtain the physical limit safety threshold amperes. To standardize the units, the physical limit safety threshold amperes is used as the original actual drive current value and proportionally converted in the aforementioned range normalization process, resulting in a preset current tolerance limit of 0.92. The cumulative counter is initialized and cleared. The normalized disturbance actual drive current values ​​are read one by one and compared with 0.92 for comparison. If the normalized disturbance actual drive current value is greater than 0.92, the current value of the accumulator counter is incremented by 1; if the normalized disturbance actual drive current value is less than 0.92, the accumulator counter value remains unchanged; if the normalized disturbance actual drive current value is equal to 0.92, the accumulator counter value remains unchanged. After completing the comparison operation of 50 data points within the time window, the accumulator counter value is recorded as the current peak frequency value of the corresponding time window.

[0026] S302: Obtain the normalized disturbance power supply ripple voltage and normalized disturbance command current in the target abnormal state disturbance matrix within the same time window, count the number of times the normalized disturbance power supply ripple voltage exceeds the preset voltage fluctuation limit, obtain the corresponding voltage fluctuation frequency, and at the same time calculate the absolute value of the difference between the normalized disturbance command current and the normalized disturbance actual drive current to obtain the steady-state control deviation characteristics. Locate the third row of the data array in the target abnormal state disturbance matrix and extract 50 normalized disturbance power supply ripple voltage values. Extract the ripple voltage data sample set of the on-board electronic controller under fault-free operation after range normalization processing, calculate the highest peak value of the value fluctuation in this sample set, and add a fixed safety tolerance bias parameter of 0.15 on top of this peak value to obtain the preset voltage fluctuation limit value of 0.65. Initialize the voltage over-limit counter and set it to 0. Compare the normalized disturbance power supply ripple voltage value with 0.65 one by one. If the voltage value is greater than 0.65, increment the voltage over-limit counter value by 1; if the voltage value is less than 0.65, do not increment the over-limit counter value; if the voltage value is equal to 0.65, do not increment the over-limit counter value. After traversing 50 data points, record the integer value as the voltage fluctuation frequency value. Extract the normalized perturbation command current value from the first row of the data array and the normalized perturbation actual drive current value from the second row of the data array. Subtract the normalized perturbation actual drive current value from the normalized perturbation command current value to obtain the floating-point value of the current control deviation. Determine the polarity of the floating-point value of the current control deviation: if it is less than 0, multiply the floating-point value of the current control deviation by -1 to obtain the absolute value of the current deviation; if it is greater than 0, leave it unchanged as the absolute value of the current deviation; if it is equal to 0, leave it unchanged as the absolute value of the current deviation. Accumulate the 50 absolute values ​​of the current deviation calculated within the time window to obtain the total absolute value of the current deviation. Divide the total absolute value of the current deviation by 50 to obtain the steady-state control deviation characteristic value.

[0027] S303: Extract the frequency of voltage fluctuations, the corresponding frequency of current spikes, and the corresponding steady-state control deviation characteristics, and arrange them in chronological order to generate a state observation sequence; Initialize a dynamically expandable unidirectional queue structure in memory. For the first time window interval, extract the frequency values ​​of current spikes, voltage fluctuations, and steady-state control deviations. Package these three feature values ​​into a comprehensive feature data block, append a start timestamp, and push it to the head of the dynamically expanding queue. Edge computing nodes generate the next time window interval by sliding backward along the time axis. Re-execute the data extraction, boundary judgment, count accumulation, and deviation calculation operations to obtain the feature values ​​for the second time window, package them into a comprehensive feature data block, and append it to the tail of the dynamically expanding queue. Continue to advance the time window until the last time node is covered. All comprehensive feature data blocks form a linear queue combination, confirming a state observation sequence containing time evolution attributes.

[0028] Please see Figure 5 Step S4 is as follows: S401: Summarize the preset state transition probability parameters, observation probability parameters, and state observation sequences to establish a probability operation data set; The state transition probability parameter is set based on the frequency statistical distribution of the transitions between state intervals in the historical operation database of the on-board electronic controller, and the observation probability parameter is set based on the feature scatter point mapping distribution extracted from the known fault classification in the pre-executed bench environment test. The internal database module is used to establish a probability calculation dataset. Historical maintenance and diagnostic logs of the on-board electronic controller are extracted, and the equipment state change trajectory before and after 10,000 valid fault occurrences is statistically analyzed. The frequency of the final transition from the stable operating state category to the transient current impact state category is calculated and divided by the total frequency of stable operating state categories to obtain the probability value of transition from stable to current impact. Similarly, using the historical frequency ratio, all transition probability values ​​between the stable operating state category, transient current impact state category, continuous voltage anomaly state category, and control logic failure state category, including self-maintenance and mutual cross-changes, are calculated and filled into a four-row, four-column square matrix to form a state transition probability matrix. A continuous voltage anomaly fault is physically implanted into the on-board electronic controller on a test bench. Steady-state control deviation characteristic values ​​are extracted from 1000 sets of corresponding state observation sequences. The arithmetic mean is calculated as the Gaussian distribution mean parameter, and the variance of this dataset is calculated as the Gaussian distribution variance parameter. The observation probability parameter under this hidden state category is calculated using the probability density principle of the Gaussian distribution. The calculation formula is: For each of the four categories, bench test samples were collected to calculate the corresponding mean and variance parameters, and this process was used to obtain the corresponding observation probability parameters. The state transition probability matrix, observation probability parameters, and state observation sequence set were then concatenated and merged to establish a probability calculation data set.

[0029] S402: Using the Hidden Markov Algorithm, calculate the probability of occurrence of the state observation sequence in various preset hidden state categories, and obtain the set of hidden category operation probabilities; Hidden state classification includes stable operation state classification, transient current impact state classification, continuous voltage abnormality state classification, and control logic failure state classification; A C++-based machine learning probabilistic inference software library is used to calculate the hidden state sequence using a Hidden Markov Model. The hidden state classification is predefined in the underlying enumeration variables as four categories: stable operation state, transient current impact state, continuous voltage anomaly state, and control logic failure state. For the first-order comprehensive feature data block in the state observation sequence, the prior initial probability value of this classification is multiplied by the observation probability parameter value obtained from the operation corresponding to the first feature data block to obtain the forward joint probability value for the classification. This process is repeated to obtain the first step forward joint probability value for the four state classifications. For the second-order data block and any target hidden state classification, the forward joint probability values ​​for stable classification, impact classification, voltage classification, and logic failure classification are multiplied by the transition probability parameter value for the target classification state. These four products are then summed to obtain the intermediate cumulative probability value. This intermediate cumulative probability value is then multiplied by the observation probability parameter value of the second-order feature data block under the current target hidden state classification to obtain the second step target forward joint probability value. The algorithm iteratively performs multiplication and addition operations on the four target classifications, following the state axis, until the last feature data block is covered. The four final values ​​are extracted as probabilities of occurrence and packaged to obtain a set of hidden classification probabilities.

[0030] S403: In the set of probability of occurrence of hidden category, sort the probability of occurrence of each operation in descending order, select the first probability of occurrence of the operation as the identifier of the corresponding hidden state category, and use it as the name of the current operation state; Extract the four floating-point probabilities of occurrence from the hidden category probability set, and perform a memory bubble sort operation. Extract the probability value at the first position and the probability value at the second position. If the probability value at the first position is less than the probability value at the second position, swap their positions and hidden state category labels in memory; if the probability value at the first position is greater than the probability value at the second position, keep the original arrangement unchanged; if the probability value at the first position is equal to the probability value at the second position, keep the original arrangement unchanged. Extract the probability value currently at the second position and the probability value at the third position, and perform a size comparison and conditional swap operation. Repeat this loop after one round of comparison until no position swaps occur in a round of comparison, forming a descending sequence. Read the maximum occurrence probability value of the first position in the descending sequence, and extract the string text content as the current running state name by querying the original enumeration label identifier bound to the global variable throughout its entire lifecycle.

[0031] Please see Figure 6 Step S5 is as follows: S501: Read the baseline classification information in the rule library for determining the relevant status category of the vehicle electronic controller, compare the current operating status name with the abnormal classification condition judgment label in the rule library for determining the status category, determine whether the current operating status name meets the corresponding abnormal classification, and obtain the status abnormal judgment result. The system retrieves the state category determination rule base file residing in the non-volatile read-only memory area of ​​the on-board electronic controller via the internal integrated circuit interconnect bus. It then reads the baseline classification information list from the data dictionary node. During factory programming, system safety risk states such as transient current anomalies, continuous voltage anomalies, and control logic failures are defined as anomaly classification condition determination labels, while text representing stable states is removed. The edge computing node extracts the string data sequence of the current operating state name and the string data sequence of the anomaly classification condition determination label from the baseline classification information list. It compares the first character of the current operating state name with the first character of the first label in the rule base using binary low-level encoding. If they match, the pointer moves to the next character for comparison. If all character encodings and sequence lengths are completely identical, a successful match is determined. If any character encoding is inconsistent, a match is determined, and the system switches to the next label to start the first character comparison. If the lengths are inconsistent, a match is determined, and the system switches to the next label. If any anomaly classification condition determination label is successfully matched, the edge computing node triggers internal diagnostic alarm logic, generating a state anomaly determination result representing a Boolean truth value. If no label is matched after traversing the list, a state anomaly determination result representing a Boolean false value is generated.

[0032] S502: Based on the status anomaly determination result, confirm the anomaly category to which the current running status name belongs, use it as the corresponding fill content for the anomaly warning information data, encapsulate the fill content into the specified communication protocol format message data packet for information hierarchical combination and encapsulation, and generate a fault early warning message. When the status anomaly determination result flag is a Boolean true value, the edge computing node extracts the successfully matched anomaly category name. The string text is then converted character by character into the corresponding hexadecimal byte array sequence using a standard character encoding mapping table. The system's pre-set anomaly warning information data empty template file is retrieved, and the generated hexadecimal byte array is overwritten into the payload segment offset address, completing the loading of the corresponding fill content. The frame structure elements of the vehicle local area network bus message data packet are constructed in memory. A frame start byte sequence, frame identification code data sequence, and data length code sequence are generated, and the anomaly warning information data containing the corresponding fill content is appended as the data segment payload body. The decimal values ​​of all bytes from the frame start sequence to the data segment payload body sequence are extracted and summed using continuous addition. The sum is divided by the pre-defined generator polynomial base integer value of the communication protocol, and the remainder is extracted. A cyclic redundancy check (CRC) byte sequence is generated according to the bit-field inversion and padding logic, and this sequence is concatenated with the frame end byte sequence to form a binary underlying pulse digital encoding sequence, which is established as the fault warning message.

[0033] S503: The edge computing node sends the fault warning message to be processed to the relevant vehicle display terminal for warning interaction, and records the communication link transmission status of the edge computing node in response to the fault warning message transmission, which serves as the real-time fault diagnosis result of the vehicle electronic controller. The hardware driver port communication baud rate is configured to 500 kilobits per second. Through pulse width modulation and high / low level switching, the pending fault warning message, in binary underlying pulse digital encoding sequence form, is pushed bit-by-bit to the vehicle-mounted physical twisted-pair transmission bus network. The message is transmitted to the bound vehicle-mounted display terminal. The terminal performs redundancy verification locally. If there are no errors, the driver displays a red fault warning icon and the associated anomaly classification text flashing on the display screen, forcibly pulling the bus level low to generate a dominant response feedback confirmation digital encoding signal. After the push is complete, the edge computing node starts a 2-millisecond response time window countdown. If the dominant response feedback signal is captured and successfully read before the countdown reaches zero, the edge computing node sets the value of its internal communication link status register to logic 1, determining that the transmission is successful; if the signal is not captured when the countdown reaches zero, the register value is set to logic 0, determining that the transmission has failed; if the signal is captured exactly at the moment the countdown reaches zero, it is determined that the effective time window limit has been exceeded, the register value is set to logic 0, determining that the transmission has failed. The edge computing node extracts the current operating status name text, concatenates it with the communication link status register value (logic 1 or logic 0) and the current system absolute timestamp information, and writes it into a read-only protected log sector for storage and persistence. This is then declared in the software interface, establishing it as the real-time fault diagnosis result of the vehicle electronic controller. Through the complete execution of this step and the preceding processes, the system achieves real-time early warning, closed-loop communication response, and persistent recording of underlying abnormal states. This provides reliable data support and underlying judgment criteria for fault prediction and health management of the vehicle electronic controller, significantly improving the safety and maintainability of the equipment under complex operating conditions.

[0034] Please see Figure 7 A real-time fault diagnosis system for vehicle electronic controllers based on edge computing includes: The matrix construction module acquires multiple data points, including command current, actual drive current, and power supply ripple voltage of the vehicle electronic controller through edge computing nodes during the continuous operation cycle of the vehicle electronic controller, and constructs a multi-dimensional time-series operation data matrix. The disturbance matrix partitioning module determines the initial abnormal state disturbance matrix based on the comparison between the fluctuation amplitude of each data item in the multi-dimensional time-series running data matrix and the steady-state threshold, determines the spatial reconstruction error of the initial abnormal state disturbance matrix, and generates the target abnormal state disturbance matrix. The observation sequence combination module determines multiple characteristics such as current spike frequency, voltage fluctuation frequency and steady-state control deviation based on the target abnormal state disturbance matrix, and combines them to form a state observation sequence. The running state name determination module determines the probability of running occurrence under various hidden state categories based on the state observation sequence and preset probability parameters, and extracts the current running state name with the first running occurrence probability arranged in descending order. The diagnostic result generation module generates a fault warning message based on the current operating status name, and uses the fault warning message as the real-time fault diagnosis result of the vehicle electronic controller.

[0035] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A real-time fault diagnosis method for vehicle electronic controllers based on edge computing, characterized in that, Includes the following steps: S1: During the continuous operation cycle of the vehicle electronic controller, multiple data such as the command current, actual drive current and power supply ripple voltage of the vehicle electronic controller are obtained through edge computing nodes to construct a multi-dimensional time-series operation data matrix. S2: Based on the comparison results between the fluctuation amplitude of each data item in the multi-dimensional time-series running data matrix and the steady-state threshold, determine the initial abnormal state disturbance matrix, determine the spatial reconstruction error of the initial abnormal state disturbance matrix, and generate the target abnormal state disturbance matrix. S3: Based on the target abnormal state disturbance matrix, determine multiple characteristics such as current peak frequency, voltage fluctuation frequency and steady-state control deviation, and combine them to form a state observation sequence; S4: Based on the state observation sequence and preset probability parameters, determine the probability of occurrence of operation under various hidden state categories, and extract the current running state name with the first running probability arranged in descending order; S5: Generate a fault warning message based on the current operating status name, and use the fault warning message as the real-time fault diagnosis result of the vehicle electronic controller.

2. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 1, characterized in that, The multidimensional time-series operation data matrix includes command current time-series items, actual drive current time-series items, and power supply ripple voltage time-series items. The target abnormal state disturbance matrix includes command current disturbance components, actual drive current disturbance components, and power supply ripple voltage disturbance components. The state observation sequence includes current spike frequency feature elements, voltage fluctuation frequency feature elements, and steady-state control deviation feature elements. The current operating state name is specifically a hidden state identifier calculated by extracting the probability of occurrence of the first operating state. The real-time fault diagnosis result is specifically a fault warning message calculated by determining the abnormal category of the current operating state name.

3. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 1, characterized in that, Step S1 is as follows: S101: By monitoring multiple data points such as command current, actual drive current, and power supply ripple voltage of the vehicle electronic controller during continuous operation cycle through edge computing nodes, the corresponding timestamps are obtained and recorded. The time dimension of each data item is aligned according to the timestamps to obtain the initial time-series aligned data sequence. S102: Normalize each data item in the initial time-aligned data sequence to establish a normalized runtime sequence feature set; S103: Extract each normalized data item from the normalized runtime sequence feature set, and perform array structure splicing and combination according to the time sequence dimension to construct a multi-dimensional time sequence runtime data matrix.

4. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 1, characterized in that, Step S2 is as follows: S201: The normalized command current, actual driving current and power supply ripple voltage in the multi-dimensional time-series running data matrix are numerically decomposed into matrix space structure by principal component analysis algorithm. The fluctuation amplitude corresponding to each data item is calculated and compared with the corresponding preset steady-state threshold to obtain the fluctuation amplitude comparison classification result. S202: Extract the fluctuation amplitude comparison and classification results, and classify the multiple data that have not reached the corresponding steady-state threshold value into the stable control baseline matrix, and classify the multiple data that have reached or exceeded the corresponding steady-state threshold into the initial abnormal state disturbance matrix. S203: Calculate the spatial reconstruction error between the data in the multidimensional time-series running data matrix, the baseline data in the stable control baseline matrix, and the disturbance data in the initial abnormal state disturbance matrix, and separate the target abnormal state disturbance matrix.

5. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 1, characterized in that, Step S3 is as follows: S301: Obtain the normalized disturbance actual driving current in the target abnormal state disturbance matrix within the set time window, compare the normalized disturbance actual driving current with the preset current tolerance limit value corresponding to the vehicle electronic controller, count the number of times the current peak frequency is exceeded. S302: Obtain the normalized disturbance power supply ripple voltage and normalized disturbance command current in the target abnormal state disturbance matrix within the same time window, count the number of times the normalized disturbance power supply ripple voltage exceeds the preset voltage fluctuation limit, obtain the corresponding voltage fluctuation frequency, and at the same time calculate the absolute value of the difference between the normalized disturbance command current and the normalized disturbance actual drive current to obtain the steady-state control deviation characteristics. S303: Extract the voltage fluctuation frequency, the corresponding current peak frequency, and the corresponding steady-state control deviation characteristics, and arrange them in chronological order to generate a state observation sequence.

6. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 1, characterized in that, Step S4 is as follows: S401: Summarize the preset state transition probability parameters and observation probability parameters, as well as the state observation sequence, to establish a probability operation data set; S402: Using the Hidden Markov Model, calculate the probability of the state observation sequence being in multiple preset hidden state classifications, and obtain the set of hidden classification operation probabilities; S403: In the set of hidden category operation probabilities, each operation probability is sorted in descending order, and the identifier of the corresponding hidden state category associated with the first operation probability in the sorted list is selected as the name of the current operation state.

7. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 1, characterized in that, Step S5 is as follows: S501: Read the baseline classification information in the relevant status category determination rule library of the vehicle electronic controller, compare the current operating status name with the abnormal classification condition determination label in the status category determination rule library, determine whether the current operating status name meets the corresponding abnormal classification, and obtain the status abnormal determination result. S502: Based on the state anomaly determination result, confirm the anomaly category to which the current running state name belongs, and use it as the corresponding filling content for the anomaly warning information data. Encapsulate the filling content into the specified communication protocol format message data packet for information hierarchical combination and encapsulation to generate a fault early warning message. S503: The fault warning message to be processed is sent to the relevant vehicle display terminal for warning interaction through the edge computing node, and the communication link transmission status is confirmed by the feedback of the edge computing node on the transmission of the fault warning message, which serves as the real-time fault diagnosis result of the vehicle electronic controller.

8. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 6, characterized in that, The state transition probability parameter is set based on the frequency statistical distribution of mutual transitions between state intervals in the historical operation database of the vehicle electronic controller, and the observation probability parameter is set based on the feature scatter point mapping distribution extracted from the known fault classification in the pre-executed bench environment test.

9. The real-time fault diagnosis method for vehicle electronic controllers based on edge computing according to claim 6, characterized in that, The hidden state classification includes stable operation state classification, transient current impact state classification, continuous voltage abnormality state classification, and control logic failure state classification.

10. A real-time fault diagnosis system for vehicle electronic controllers based on edge computing, characterized in that, The system is used to implement the real-time fault diagnosis method for vehicle electronic controllers based on edge computing as described in any one of claims 1-9, and the system includes: The matrix construction module acquires multiple data points of the vehicle electronic controller, including command current, actual drive current, and power supply ripple voltage, through edge computing nodes during the continuous operation cycle of the vehicle electronic controller, and constructs a multi-dimensional time-series operation data matrix. The disturbance matrix partitioning module determines the initial abnormal state disturbance matrix based on the comparison results between the fluctuation amplitude of each data item in the multi-dimensional time-series running data matrix and the steady-state threshold, determines the spatial reconstruction error of the initial abnormal state disturbance matrix, and generates the target abnormal state disturbance matrix. The observation sequence combination module determines multiple characteristics such as current spike frequency, voltage fluctuation frequency and steady-state control deviation based on the target abnormal state disturbance matrix, and combines them to form a state observation sequence. The running state name determination module determines the probability of running occurrence under various hidden state categories based on the state observation sequence and preset probability parameters, and extracts the current running state name with the first running occurrence probability arranged in descending order. The diagnostic result generation module generates a fault warning message based on the current operating status name, and uses the fault warning message as the real-time fault diagnosis result of the vehicle electronic controller.

Images