A method for processing functional test data of an automobile headlamp driving board

Abstract

The application relates to the technical field of automobiles, in particular to a method for processing functional test data of an automobile headlamp driving board, which comprises the following steps: a hybrid index structure composed of a multi-dimensional combined key B+ tree and a fault transient feature hash table is constructed, wherein the multi-dimensional combined key B+ tree is composed of a combined key formed by a test batch, a relative time coordinate and a diagnostic parameter hierarchical level, and a leaf node stores an original data address and a fault local sensitive feature hash value; the fault transient feature hash table performs filtering and slope feature extraction on current and voltage waveforms in a preset window, generates a fault transient feature vector and establishes a mapping relationship between a hash value and a test batch and a relative time coordinate; after receiving a composite query, a target hash value is generated according to a fault mode, candidate data points are approximately matched, a B+ tree is searched in combination with a parameter interval, and hash similarity and diagnostic parameter checking are performed, so that target test data are extracted, and the retrieval efficiency and fault matching accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of automotive technology, and in particular to a method for processing functional test data of an automotive headlight driver board. Background Technology

[0002] With the rapid development of modern automotive electronics technology, the intelligence and integration of automotive headlight systems are constantly improving. As a core component for power regulation, brightness control, and fault protection, the headlight driver board's operational stability and reliability directly affect automotive lighting performance and driving safety. Therefore, systematic functional testing is required during the R&D verification and factory testing phases of the headlight driver board. During testing, testing equipment typically collects and records a large amount of test data at a high frequency. This data includes not only structured diagnostic parameters such as test batches, time series, temperature, current, and voltage, but also dynamic changes in current and voltage waveforms during transient processes such as startup, switching, overcurrent, and short circuits. This data is characterized by its large scale, multiple dimensions, heterogeneous types, and strong temporal correlation, containing rich operational status information and potential fault characteristics. How to efficiently store, organize, retrieve, and match this type of test data to support subsequent quality traceability, process optimization, and fault diagnosis has become a crucial problem that needs to be solved in the field of automotive electronics manufacturing and testing.

[0003] Existing large-scale data indexing technologies are mostly designed for single-field or low-dimensional structured data, and are usually suitable for retrieval by time, batch or single parameter. They are difficult to meet the needs of time interval, batch range, multi-dimensional diagnostic parameter constraints and joint query of specific fault modes at the same time. Especially for the transient characteristics of faults reflected by current and voltage waveforms, traditional indexing methods are difficult to effectively combine the approximate matching of unstructured waveform features with the interval retrieval of structured parameters. As a result, when locating specific complex fault scenarios in massive test data, the retrieval path is complex, there are many invalid data accesses, and the matching efficiency and accuracy are limited.

[0004] Therefore, it is necessary to propose a hybrid indexing and processing method for functional test data of automotive headlight driver boards to improve the retrieval efficiency and joint matching capability of multidimensional heterogeneous test data. Summary of the Invention

[0005] To address the technical problems in existing technologies, such as the multidimensional heterogeneity of automotive headlight driver board test data, the difficulty in integrating structured parameter range retrieval with fault waveform feature approximation matching, low efficiency of composite queries, and excessive access to invalid data, this application provides a method for processing functional test data of automotive headlight driver boards.

[0006] This application provides a method for processing functional test data of an automotive headlight driver board, including:

[0007] S1. Construct a hybrid index structure, including building a multidimensional composite key B+ tree and a fault transient feature hash table. The key value of the multidimensional composite key B+ tree is composed of test batch identifier, relative time coordinate, and hierarchical level of key diagnostic parameters. The leaf nodes store address pointers to the original data blocks and embed local sensitive feature hash values ​​representing fault transient features. Constructing the fault transient feature hash table involves filtering the current and voltage waveforms within a preset time window, extracting discrete slope sequences, generating fault transient feature vectors, calculating local sensitive feature hash values, and establishing a mapping relationship between local sensitive feature hash values ​​and test batch identifiers and relative time coordinates. S2. Receive a composite query request containing time interval, batch range, diagnostic parameter constraints and specific fault mode, extract the specific fault mode, reconstruct the target fault transient feature vector and calculate the target local sensitive feature hash value, approximate match in the fault transient feature hash table, and obtain the test batch identifier and relative time coordinate set corresponding to the candidate data point. S3. Based on the test batch identifier and relative time coordinate combined with hierarchical intervals, construct the query combination key range, and perform interval retrieval in the multidimensional combination key B+ tree to locate the target leaf node. Perform similarity verification based on the embedded hash value of the leaf node, and combine the constraint verification results of the actual diagnostic parameter measurement values ​​in the candidate data block to determine the target test data.

[0008] This application constructs a hybrid index structure combining a multidimensional composite key B+ tree and a hash table of transient fault features. Address pointers are stored in the leaf nodes of the B+ tree, and local sensitive feature hash values ​​are embedded within them. Upon receiving a composite query request, the target features are reconstructed, and hash approximation matching is performed to obtain a candidate set. Then, interval retrieval and embedded hash value verification are executed in the B+ tree to determine the target test data. This solves the problem that traditional single indexes cannot simultaneously satisfy the integration of structured parameter interval retrieval and unstructured fault waveform feature approximation matching. This combined mechanism of hash approximation matching for initial filtering, B+ tree multidimensional interval retrieval, and lightweight secondary verification greatly improves the retrieval efficiency of composite queries, significantly reduces system retrieval latency, and effectively reduces invalid data access during the retrieval of massive transient fault data, significantly reducing the number of disk reads.

[0009] In one embodiment, the leaf node stores an address pointer to the original data block and embeds a hash value of a local sensitivity feature representing the transient characteristics of the fault, including: Convert the test batch identifier to a fixed-length hexadecimal string; Extract the absolute timestamp of the raw data and subtract it from the system's zero-point reference time to obtain the relative time coordinates; The measured values ​​of key diagnostic parameters are compared with preset segment boundary intervals, and the corresponding grading levels are assigned based on the comparison results. The fixed-length hexadecimal string, relative time coordinate, and hierarchical level are concatenated in order to form a key, which is then stored in the leaf node of the multidimensional key B+ tree.

[0010] This application achieves memory alignment and improves the comparison efficiency of B+ trees by converting batch identifiers into fixed-length hexadecimal strings and padding them; it reduces the storage overhead of time fields by extracting relative time coordinates instead of absolute time and reduces the probability of generating the same time coordinates within the same millisecond; and it reduces the large amount of intersection calculations required by traditional multi-condition independent indexes by concatenating fixed-length strings, relative time coordinates, and hierarchical levels into a composite key, significantly optimizing the efficiency of range queries under multi-dimensional conditions.

[0011] In one embodiment, calculating the local sensitive feature hash value and establishing the mapping relationship between the local sensitive feature hash value and the test batch identifier and relative time coordinate includes: A preset time window is formed by extracting a signal sequence of a fixed number of clock cycles before and after the time when the preset fault code occurs. The discrete difference quotient between current and voltage between adjacent sampling points is calculated as a discrete slope sequence; After the discrete slope sequence is dimensionless, it is arranged in chronological order to form a two-dimensional matrix. After discretization and dimensionality reduction, a predefined locality-sensitive hash function is used to calculate and generate locality-sensitive feature hash values.

[0012] This application calculates the discrete difference quotient as a discrete slope sequence by extracting the time window signal, and combines it with dimensionless processing to eliminate the difference in dimensions and amplitude between voltage and current, making the features more accurately reflect the dynamic changes of the waveform. By arranging the sequence into a two-dimensional matrix according to time sequence and performing discretization dimensionality reduction processing, the computational complexity is effectively reduced. The feature hash value is generated by using the locality-sensitive hash function, which retains the characteristic that similar inputs produce similar outputs, laying the foundation for subsequent efficient Hamming distance approximate matching.

[0013] In one embodiment, the extraction of a specific fault mode includes: Extract the start timestamp, end timestamp, test batch identifier boundary code, diagnostic parameter constraints, and fault mode identifier from the composite query request to determine the corresponding time interval, test batch range, and specific fault mode.

[0014] This application determines the corresponding time interval and batch range by accurately extracting the start timestamp, end timestamp, batch identifier boundary code, diagnostic parameter constraints, and fault mode identifier from the composite query request. This provides accurate basic parameters for subsequently limiting the target retrieval time interval, defining the test batch range, and matching precise fault waveform templates, ensuring the accuracy of the multi-dimensional joint query operation boundary.

[0015] In one embodiment, reconstructing the target fault transient feature vector and calculating the target local sensitivity feature hash value includes: The preset fault mode standard waveform template is invoked according to the specific fault mode, and the target fault transient feature vector is generated according to the same feature processing rules as the construction of the fault transient feature hash table, and the target local sensitivity feature hash value is calculated. The time interval, test batch range, diagnostic parameter constraints, specific fault modes, and target local sensitive feature hash values ​​are encapsulated into a query task package and added to the task queue of the retrieval scheduling controller.

[0016] This application reconstructs the target fault transient feature vector and calculates the hash value by calling a pre-collected or cleaned preset standard waveform template according to a specific fault mode, thus ensuring the baseline accuracy of the target hash value calculation. The query conditions and target hash value are encapsulated into a task package and added to the task queue of the retrieval scheduler controller. Through a multi-threaded safety mechanism, the asynchronous execution of the background retrieval worker thread is supported, which greatly improves the system throughput and retrieval response speed under high concurrency conditions.

[0017] In one embodiment, obtaining the test batch identifier and relative time coordinate set corresponding to the candidate data points includes: Invoke the target local sensitive feature hash value generated based on the specific failure mode; In the fault transient feature hash table, approximate collision matching is performed based on a preset hash distance threshold to parse out the corresponding test batch identifier bound to the underlying record of the hit local sensitive feature hash value and the relative time coordinate mapping pair. The mapping pairs are filtered according to the time interval and the test batch range to obtain the test batch identifier and relative time coordinate set corresponding to the candidate data points.

[0018] This application uses approximate collision matching based on a preset hash distance threshold in the fault transient feature hash table to parse out the corresponding mapping pairs bound to the underlying records of the hit hash values, and then filters according to the time interval and batch range. This greatly reduces the scope of the entire table traversal, enables the rapid location of candidate hash buckets that meet the distance requirements, achieves efficient preliminary filtering of candidate data points at the memory level, and reduces the data scale of subsequent precise retrieval.

[0019] In one embodiment, constructing the query composite key range includes: Parse the upper and lower limits of diagnostic parameter constraints in the composite query request and map them to the corresponding hierarchical level range; The obtained test batch identifier, relative time coordinate, and grade range are combined to generate a range of query combination keys.

[0020] In one embodiment, performing range retrieval to locate the target leaf node in a multidimensional composite key B+ tree includes: Using the range of query composite keys as routing credentials, the multidimensional composite key B+ tree is traversed downwards from the root node and range search is performed to locate the target leaf node.

[0021] In one embodiment, determining the target test data includes: Extract the hash value of the local sensitive features embedded in the target leaf node; The similarity of the target's local sensitive feature hash value with the local sensitive feature hash value embedded in the target's leaf node is compared. If the similarity between the hash values ​​of local sensitive features meets the preset similarity condition, and the actual diagnostic parameter measurement values ​​in the candidate data block meet the diagnostic parameter constraints, then the data block reading instruction is triggered to extract the target test data.

[0022] In one embodiment, the predefined locality-sensitive hash function is the SimHash function.

[0023] The technical solution of this application has the following beneficial technical effects: This application employs a hybrid indexing mechanism combining a multidimensional composite key B+ tree with a fault transient feature hash table. This mechanism concatenates test batches, relative time coordinates, and diagnostic parameter hierarchies into a composite key for rapid location of structured conditions. Simultaneously, it uses a local sensitive feature hash table to approximate the matching of current and voltage fault waveforms, filtering candidate data points. During the composite query process, a coarse screening is first performed using fault feature hashing, followed by interval retrieval using a B+ tree, and verification is conducted using embedded hash values ​​in leaf nodes and diagnostic parameter constraints. This reduces invalid data access and improves the retrieval efficiency of target test data and the accuracy of fault feature matching. Attached Figure Description

[0024] Figure 1 A flowchart of a method for processing functional test data of an automotive headlight driver board; Figure 2 Assign byte histograms to each component of the composite key storage structure; Figure 3 A standardized time-series vector curve for fault transient characteristics. Detailed Implementation

[0025] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0026] Figure 1This is a flowchart illustrating a method for processing functional test data of an automotive headlight driver board according to an embodiment of this application. Figure 1 As shown, the method for processing functional test data of automotive headlight driver board includes steps S1 to S3, which are described in detail below.

[0027] S1. Constructing a hybrid index containing B+ tree and hash table mapping fault transient characteristics.

[0028] Specifically, a hybrid index structure is constructed, including building a multidimensional composite key B+ tree and a fault transient feature hash table. The key value of the multidimensional composite key B+ tree is composed of test batch identifier, relative time coordinate, and hierarchical level of key diagnostic parameters. The leaf nodes store address pointers to the original data blocks and embed local sensitive feature hash values ​​representing fault transient features. The fault transient feature hash table is constructed by filtering the current and voltage waveforms within a preset time window, extracting discrete slope sequences, generating fault transient feature vectors, calculating local sensitive feature hash values, and establishing a mapping relationship between local sensitive feature hash values ​​and test batch identifiers and relative time coordinates.

[0029] Furthermore, the original time series data of the current and voltage waveforms of the automotive headlight driver board are read, and the data within the preset time window is low-pass filtered to remove high-frequency noise interference. The filtered waveform data is then subjected to first-order difference operation to extract the discrete slope sequence of current and voltage. The discrete slope sequence is then concatenated in time order and subjected to feature processing to generate a fault transient feature vector. Finally, the local sensitive hash algorithm is used to calculate the local sensitive feature hash value.

[0030] A fault transient feature hash table is constructed in memory, and the local sensitive feature hash value is used as an index to establish a one-to-many mapping relationship between it and the corresponding test batch identifier, relative time coordinate and address pointer. At the same time, the local sensitive feature hash value is divided into multiple hash segments according to the preset bit segmentation rules, and a segmented hash bucket index is constructed with hash segment identifier and segment value to support subsequent candidate hash bucket location based on Hamming distance.

[0031] For key diagnostic parameters, a hierarchical level is generated according to the preset binning rules, and the test batch identifier, relative time coordinate, and hierarchical level are concatenated into a multi-dimensional composite key. The multi-dimensional composite key is used as the index key of the B+ tree, and the original data block address pointer and the corresponding local sensitive feature hash value are stored in the leaf node of the B+ tree. When the same composite key corresponds to multiple original data blocks, the set of address pointers and the corresponding local sensitive feature hash value are stored in the corresponding leaf node.

[0032] In an optional embodiment, a multidimensional composite key B+ tree is constructed. The key value of the multidimensional composite key B+ tree is composed of test batch identifier, relative time coordinate, and hierarchical levels of key diagnostic parameters. The leaf nodes store address pointers to the original data blocks and embed local sensitive feature hash values ​​representing transient fault characteristics, including: Convert the test batch identifier to a fixed-length hexadecimal string; Extract the absolute timestamp of the raw data and subtract it from the system's zero-point reference time to obtain the relative time coordinates; The measured values ​​of key diagnostic parameters are compared with preset segment boundary intervals, and the corresponding grading levels are assigned based on the comparison results. The fixed-length hexadecimal string, relative time coordinate, and hierarchical level are concatenated in order to form a key, which is then stored in the leaf node of the multidimensional key B+ tree.

[0033] In the automotive headlight driver board functional test data processing system, in order to achieve the retrieval of massive test data, the generation process of multi-dimensional combination keys needs to go through parameter setting and conversion calculation. Test batch identifiers of varying lengths are uniformly converted into 16-byte fixed-length hexadecimal strings. For example, the batch number "BAT-2023-10" is converted into "4241542D323032332D31300000000000". If the length is insufficient, zeros are padded at the end to achieve memory alignment and improve the comparison efficiency of B+ trees.

[0034] Set a fixed zero-point reference time, such as 2023-01-01 00:00:00.000, extract the absolute timestamp of the original test data, and record it with a time precision of not less than the sampling clock period, such as using a microsecond-level relative time coordinate. The difference between the two is used to obtain an unsigned long integer relative time coordinate. An example value can be represented as 24849025125000 microseconds. This coordinate variable compresses the storage overhead of the time field and reduces the probability that multiple transient data blocks within the same millisecond will generate the same time coordinate.

[0035] For key diagnostic parameters, such as LED driver output current, multiple segmented boundary ranges are preset. For example, the output current range is divided into four levels: [0, 500mA), [500mA, 1000mA), [1000mA, 1500mA), and [1500mA, +∞), which correspond to classification levels 01, 02, 03, and 04, respectively.

[0036] When the actual measured value is 1250mA, the automatic classification level is assigned as 03; the bytes allocated to each component of the composite key storage structure are as follows: Figure 2As shown, the batch identifier occupies 16 bytes, the time coordinate occupies 8 bytes, and the grade level field occupies 1 byte. Following the order of "hexadecimal string + relative time coordinate + grade level," these three fields are merged through string concatenation and shift operations. In the underlying storage, this forms a 25-byte binary fixed-length key consisting of the 16-byte batch identifier field, the 8-byte relative time coordinate field, and the 1-byte grade level field. For example, this key can be displayed as a string like "4241542D323032332D31300000000000_24849025125000_03". This key serves as an index credential stored in the leaf nodes of the B+ tree, optimizing the efficiency of range queries under multi-dimensional conditions. If multiple original data blocks correspond to the same key, the different original data blocks are distinguished by the set of address pointers in the leaf nodes.

[0037] In an optional embodiment, constructing the fault transient feature hash table involves filtering the current and voltage waveforms within a preset time window to extract discrete slope sequences, generating a fault transient feature vector, calculating local sensitive feature hash values, and establishing a mapping relationship between the local sensitive feature hash values ​​and the test batch identifier and relative time coordinates, including: A preset time window is formed by extracting a signal sequence of a fixed number of clock cycles before and after the time when the preset fault code occurs. The discrete difference quotient between current and voltage between adjacent sampling points is calculated as a discrete slope sequence; After the discrete slope sequence is dimensionless, it is arranged in chronological order to form a two-dimensional matrix. After discretization and dimensionality reduction, a predefined locality-sensitive hash function is used to calculate and generate locality-sensitive feature hash values.

[0038] Using the timestamp triggered by a preset fault code, such as overcurrent fault code E001, as the center, set the clock cycle parameter to 10. Extract signal sequences of 500 clock cycles forward and backward to construct a time span of 10. The preset time window contains 1001 sampling points; the standardized discrete fault characteristic time series curves within the window are as follows: Figure 3 As shown, the rate of change of current and voltage between adjacent sampling points is calculated using the difference quotient formula, generating a current discrete slope sequence and a voltage discrete slope sequence of length 1000.

[0039] Since the dimensions and amplitudes of current and voltage differ significantly (e.g., voltage is 9-16V and current is 0-2A), Z-score standardization is performed on both sequences to make the mean of the processed sequences 0 and the standard deviation 1. After dimensionless standardization, the standardized current slope sequence and voltage slope sequence are arranged in chronological order as row vectors to construct a two-dimensional feature matrix of size 2×1000.

[0040] To reduce computational complexity, the two-dimensional feature matrix is ​​divided into 128 consecutive time segments along the time dimension, allowing adjacent segments to differ in length by no more than one sampling point. The mean value of the feature values ​​within each segment is calculated to obtain a 2×128 dimensionality-reduced feature matrix. Subsequently, the SimHash locality-sensitive hash function is used to perform hash mapping on the dimensionality-reduced feature matrix, generating a 64-bit locality-sensitive feature hash value, such as 0x3F2A8B4C9E1D7F55. This hash value has the characteristic that similar inputs produce similar hash results. This hash value is then used as a key to insert into the fault transient feature hash table, realizing the storage and comparison retrieval of waveform features.

[0041] When inserting into the fault transient feature hash table, the local sensitive feature hash value is divided into multiple hash segments according to the preset bit segmentation rules. The segment bucket key is constructed by the hash segment number and hash segment value. The complete local sensitive feature hash value and its corresponding test batch identifier, relative time coordinate and address pointer are written into the record set corresponding to the segment bucket key to build the segment hash bucket index. This is so that the candidate hash bucket can be located according to the target local sensitive feature hash value and Hamming distance approximate matching can be performed.

[0042] S2. Receive composite query requests and reconstruct target features to perform approximate collision matching.

[0043] Specifically, it receives a composite query request containing time interval, batch range, diagnostic parameter constraints, and specific fault modes, extracts the specific fault modes, reconstructs the target fault transient feature vector and calculates the target local sensitive feature hash value, approximates matching in the fault transient feature hash table, and obtains the test batch identifier and relative time coordinate set corresponding to the candidate data points.

[0044] Furthermore, the system receives composite query requests containing time intervals, batch ranges, diagnostic parameter constraints, and specific fault modes through the service interface, and parses them to obtain the corresponding query conditions. Based on the specific fault mode, it calls a preset fault mode standard waveform template, and follows the same processing rules as when constructing the fault transient feature hash table. It then filters, extracts discrete slopes, and reconstructs features from the current and voltage waveforms in the standard template to generate the target fault transient feature vector and calculates the target local sensitive feature hash value. Based on the target local sensitive feature hash value, it generates segmented bucket keys, locates candidate hash buckets in the segmented hash bucket index of the fault transient feature hash table, and calculates the Hamming distance for the complete hash values ​​within the candidate hash buckets. Records that meet a preset distance threshold are used as approximate matching records, and their corresponding test batch identifiers and relative time coordinate mapping pairs are extracted. These are then filtered according to the time interval and test batch range to obtain a set of candidate data points.

[0045] In an optional embodiment, a composite query request containing a time interval, batch range, diagnostic parameter constraints, and a specific fault mode is received; the specific fault mode is extracted; the target fault transient feature vector is reconstructed; and the target local sensitivity feature hash value is calculated, including: Extract the start timestamp, end timestamp, test batch identifier boundary code, diagnostic parameter constraints, and fault mode identifier from the composite query request to determine the corresponding time interval, test batch range, and specific fault mode. The preset fault mode standard waveform template is invoked according to the specific fault mode, and the target fault transient feature vector is generated according to the same feature processing rules as the construction of the fault transient feature hash table, and the target local sensitivity feature hash value is calculated. The time interval, test batch range, diagnostic parameter constraints, specific fault modes, and target local sensitive feature hash values ​​are encapsulated into a query task package and added to the task queue of the retrieval scheduling controller.

[0046] When a client initiates a composite query request, the structured query parameters are parsed to extract the multi-dimensional filtering conditions contained therein. For the time dimension, the start timestamp attribute with an accuracy no less than the sampling clock cycle is extracted, such as "start_time": 1696118400000000 and the end timestamp attribute, such as "end_time": 1696122000000000. The difference between the two is calculated and verified against the baseline zero point to lock the time interval of the target retrieval and ensure that the query operation is limited to a specific time period.

[0047] For the batch range, read the test batch identifier boundary code in the request body, such as the starting test batch identifier "BAT-A-001" and the ending test batch identifier "BAT-A-999", and use the lexicographical comparison algorithm to define the test batch range set.

[0048] Simultaneously, the specific fault mode string identifier in the request is parsed, such as "Fault_Type": "Short_Circuit_Transient", and mapped to the preset fault mode standard waveform template library predefined by the internal system. The preset fault mode standard waveform template library is constructed by pre-collecting or importing test samples of automotive headlight driver boards labeled with fault types. The current and voltage waveform segments of each fault mode are time-aligned, abnormal samples are removed, and averaged to generate the corresponding standard fault waveform template. After establishing a mapping relationship between the fault mode identifier and the standard fault waveform template, it is stored in the template library.

[0049] In addition, diagnostic parameter constraints, such as limiting the temperature range to -40℃ to 85℃ and the input voltage to 12V±10%, will also be parsed simultaneously. Among them, the key diagnostic parameter constraints that participate in the construction of multidimensional composite keys are used to generate graded level ranges; other diagnostic parameter constraints that do not participate in the construction of composite keys are retained in the query task package as subsequent precise filtering conditions.

[0050] After completing the extraction of all fields, format validation, reconstruction of the target fault transient feature vector, and calculation of the target local sensitive feature hash value, the time interval, test batch range, diagnostic parameter constraints, specific fault modes, and target local sensitive feature hash value are serialized and encapsulated into a structured query task package. This task package contains a unique request ID, priority identifier, and values ​​of each constraint field. Through a multi-threaded safety mechanism, such as using a lock-free queue or blocking queue mechanism, it is pushed into the task queue of the retrieval scheduler controller, waiting for the background retrieval worker thread to asynchronously dequeue and execute the specific retrieval process.

[0051] In an optional embodiment, approximate matching is performed in the fault transient feature hash table to obtain the test batch identifier and relative time coordinate set corresponding to the candidate data point, including: Invoke the target local sensitive feature hash value generated based on the specific failure mode; In the fault transient feature hash table, approximate collision matching is performed based on a preset hash distance threshold to parse out the corresponding test batch identifier bound to the underlying record of the hit local sensitive feature hash value and the relative time coordinate mapping pair. The mapping pairs are filtered based on the time interval and test batch range to obtain the test batch identifier and relative time coordinate set corresponding to the candidate data points.

[0052] The built-in preset fault mode standard waveform template library is called to extract the calibration current and voltage waveform data of specific fault modes such as "short circuit transient". Following the same processing logic as the feature construction, namely fixed period truncation, difference quotient calculation, dimensionless and discretization dimensionality reduction, a standard transient feature matrix of size 2×128 is reconstructed.

[0053] Using the same locality-sensitive hash function as when constructing the fault transient feature hash table, such as the same SimHash hash seed and projection vector configuration, the standard transient feature matrix is ​​hashed to generate a 64-bit target locality-sensitive feature hash value, such as 0x3F2A8B4C9E1D7F50. After obtaining the target locality-sensitive feature hash value, Hamming distance is calculated in the fault transient feature hash table for approximate collision matching.

[0054] Set a fault tolerance threshold, for example, a Hamming distance threshold of 3, which allows a maximum of 3 different bits; reduce the full table traversal range by using segmented hash bucket indexes, locate candidate hash buckets, and calculate the Hamming distance between complete hash values ​​within the candidate hash buckets to filter candidate records that meet the distance threshold requirements, where the number of hash fragments is greater than the preset hash distance threshold; extract key-value pairs pre-bound to hash values ​​from the underlying records of these hit hash buckets, i.e., the corresponding test batch identifiers, such as "BAT-2023-10" and relative time coordinates, such as 24849025125000; then filter the key-value pairs based on the time interval and test batch range in the composite query request, and store the filtered data in a deduplicated mapping pair set to complete the initial filtering of candidate data points.

[0055] S3. Extract target test data based on combination key range retrieval and feature-accurate verification.

[0056] Specifically, based on the test batch identifier and relative time coordinates combined with hierarchical intervals, a range of query combination keys is constructed. An interval search is performed in the multidimensional combination key B+ tree to locate the target leaf node. Similarity verification is performed based on the hash value embedded in the leaf node. Combined with the constraint verification results of the actual diagnostic parameter measurement values ​​in the candidate data block, the target test data is determined.

[0057] Based on the received diagnostic parameter constraints and the aforementioned binning rules, the corresponding target classification level range is determined. The candidate test batch identifier and relative time coordinate obtained in the previous step are combined with the target classification level range to generate a query key range. Based on the query key range, an interval search is performed in the multidimensional key B+ tree to locate the target leaf node that meets the range conditions.

[0058] Read the embedded local sensitive feature hash value from the target leaf node and compare it with the target local sensitive feature hash value using Hamming distance. If the Hamming distance meets the preset secondary verification threshold, read the actual diagnostic parameter measurement value from the metadata field of the corresponding leaf node or the original data block and compare it precisely with the diagnostic parameter constraints in the composite query request. For other diagnostic parameter constraints that did not participate in the construction of the composite key, read the corresponding actual measurement value and compare it. Only when both the hash secondary verification and the precise comparison of the diagnostic parameters are satisfied is it confirmed as the target test data, and the corresponding original data block of the automotive headlight driver board function test is read according to the address pointer stored in the leaf node.

[0059] In an optional embodiment, a query composite key range is constructed based on the test batch identifier and relative time coordinates combined with hierarchical intervals, and an interval retrieval is performed in the multidimensional composite key B+ tree to locate the target leaf node, including: Parse the upper and lower limits of diagnostic parameter constraints in the composite query request and map them to the corresponding hierarchical level range; The obtained test batch identifier, relative time coordinate, and grade range are combined to generate a range of query combination keys; Using the range of query composite keys as routing credentials, the multidimensional composite key B+ tree is traversed downwards from the root node and range search is performed to locate the target leaf node.

[0060] In the process of composite condition retrieval, in order to achieve addressing in the B+ tree, discrete query constraints are transformed into continuous composite key boundaries. The specific upper and lower limits of the diagnostic parameter constraints in the composite query request are parsed. For example, if the request is to limit the LED driver output current to between 600mA and 1200mA, the quantity constraint is mapped to the preset segmented boundary model. According to the previously defined interval rules, it is determined that 600mA falls into the grade level "02" and 1200mA falls into the grade level "03", generating the grade level interval [02,03]. The candidate test batch identifiers obtained in the previous steps are uniformly converted into a 16-byte fixed-length format and candidate relative time coordinates. These are then concatenated with the upper and lower boundaries of the grade range using a string or byte format of "batch + time + grade". This generates a query combination key range that includes a lower limit key, such as "424154...00_24849025125000_02", and an upper limit key, such as "424154...00_24849025125000_03".

[0061] After generating the range of query key combinations, this range is used as an internal routing credential to initiate the retrieval process of the multidimensional key B+ tree. The retrieval pointer starts from the root node of the B+ tree and uses the prefix comparison mechanism of the key combination to perform a binary search layer by layer downwards. For example, by comparing batch identifiers, a specific branch is located; by comparing time coordinates, the node range is narrowed down; according to the hierarchical level interval, a directional traversal is performed to lock the first target leaf node that meets the lower limit key requirement; after successful location, all subsequent leaf nodes within the interval are sequentially scanned through the doubly linked list pointers between B+ tree leaf nodes until a node exceeding the upper limit key is encountered. This interval retrieval mechanism reduces the number of random I / O reads from disk or memory, ensuring the location efficiency under high concurrency.

[0062] In an optional embodiment, similarity verification is performed based on the embedded hash value of the leaf node, and the target test data is determined by combining the constraint verification results of the actual diagnostic parameter measurements in the candidate data block, including: Extract the hash value of the local sensitive features embedded in the target leaf node; The similarity of the target's local sensitive feature hash value with the local sensitive feature hash value embedded in the target's leaf node is compared. If the similarity between the hash values ​​of local sensitive features meets the preset similarity condition, and the actual diagnostic parameter measurement values ​​in the candidate data block meet the diagnostic parameter constraints, then the data block reading instruction is triggered to extract the target test data.

[0063] After identifying candidate leaf nodes through B+ tree interval retrieval, a lightweight hash-based secondary verification mechanism is implemented to remove false positives and avoid unnecessary large-volume data reads. Starting from a fixed memory offset of the currently traversed target leaf node, the 64-bit locality-sensitive feature hash value embedded during construction is read (e.g., 0x3F2A8B4C9E1D7F55). A bitwise XOR operation is performed between this extracted hash value and the target locality-sensitive feature hash value reconstructed in the previous query request (e.g., 0x3F2A8B4C9E1D7F50). The number of 1s in the XOR result is counted using the Popcount instruction, and the Hamming distance between the two hash values ​​is calculated. This Hamming distance approximates the similarity between the transient feature vectors corresponding to the two locality-sensitive feature hash values. When using the SimHash function, the Hamming distance can be used to approximate the angular similarity of the original feature vectors.

[0064] In the comparison phase, a secondary verification threshold variable is pre-configured, for example, specifying that the Hamming distance is less than or equal to 2, and this secondary verification threshold is not greater than the preset distance threshold used in the aforementioned approximate matching phase. If the comparison result shows that the Hamming distance is 1, satisfying the preset secondary verification threshold, it indicates that the original waveform corresponding to the leaf node has a high similarity to the target fault transient data. Subsequently, the actual diagnostic parameter measurement values ​​in the metadata field or original data block corresponding to the leaf node are read, and the actual diagnostic parameter measurement values ​​are precisely compared with the diagnostic parameter constraints in the composite query request; only when both the waveform hash secondary verification and the actual diagnostic parameter boundary verification are satisfied is the data block determined as the target test data.

[0065] At this point, based on the address pointer stored in the leaf node, a data block read instruction is triggered to completely extract the target test data block containing detailed voltage and current waveforms from the underlying SSD or time-series database. Conversely, if the calculated Hamming distance is greater than the secondary verification threshold, or the actual diagnostic parameter measurement value does not meet the diagnostic parameter constraints in the composite query request, then this node is determined not to meet the target query conditions, and the candidate node is discarded at the memory level, blocking I / O access requests initiated to the lower-level memory. This strategy of determining whether to perform deep reading through hash pre-verification and precise parameter filtering reduces the overall system retrieval latency and improves the consistency between the composite query results and the original diagnostic parameter constraints.

[0066] The experimental hardware platform used a server equipped with a 32-core processor, 64GB of memory, and a high-speed solid-state drive. The test dataset contained 50 million historical test data entries for automotive headlight driver boards, covering hundreds of batch identifiers and various preset fault modes. The experiment consisted of three groups: the baseline control group used a traditional multi-condition independent index retrieval system; the ablation experimental group used a multi-dimensional composite key B+ tree with leaf node removal and a hash-based secondary verification mechanism; and the complete solution group adopted all the technical features of this application. During the test, 1000 concurrent composite query requests were sent, and the average query latency and average disk I / O access count were recorded.

[0067] The baseline control group had an average query latency of 850 milliseconds and an average of approximately 1200 disk accesses per query. The ablation experimental group had an average query latency reduced to 320 milliseconds and 450 disk accesses per query. The complete solution group performed best, with an average query latency reduced to 85 milliseconds and 12 disk accesses per query. In terms of false positive data read rate, both the baseline control group and the ablation experimental group had a false positive data read rate of over 80%, while the complete solution group did not observe any significant invalid data block reads under the test conditions. The complete solution group was able to improve retrieval speed and data read effectiveness by accurately filtering computational overhead with smaller hash checksum parameters.

[0068] Comparing the baseline control group and the ablation experiment group reveals that the multidimensional composite key, by concatenating batch identifiers, relative time coordinates, and hierarchical levels, reduces the large amount of intersection calculations required by traditional multi-condition independent indexes, thus optimizing the efficiency of interval queries. Comparing the ablation experiment group and the complete solution group demonstrates the value of the leaf node local sensitive feature hash secondary verification mechanism and the actual diagnostic parameter boundary filtering mechanism. Because the complete solution group compares Hamming distances at the memory level and intercepts false matching nodes generated by hash collisions, while precisely filtering the actual diagnostic parameter measurements of candidate data blocks, it shields redundant data block read instructions initiated to the lower-level memory, resulting in a decrease in disk access. This pre-verification mechanism and precise parameter filtering mechanism further shorten the system retrieval latency and reduce invalid reads during the retrieval of massive transient fault data, building upon the ablation experiment group.

[0069] It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application shall be determined by the appended claims.

Claims

1. A method for processing functional test data of an automotive headlight driver board, characterized in that, include: S1. Construct a hybrid index structure, including constructing a multidimensional composite key B+ tree and a fault transient feature hash table. The key value of the multidimensional composite key B+ tree is composed of test batch identifier, relative time coordinate and hierarchical level of key diagnostic parameters. The leaf nodes store address pointers to the original data blocks and embed local sensitive feature hash values ​​representing fault transient features. The process involves: S1) Constructing a fault transient feature hash table by filtering current and voltage waveforms within a preset time window to extract discrete slope sequences, generating fault transient feature vectors, calculating local sensitive feature hash values, and establishing a mapping relationship between local sensitive feature hash values ​​and test batch identifiers and relative time coordinates; S2) Receiving a composite query request containing time intervals, batch ranges, diagnostic parameter constraints, and specific fault modes, extracting specific fault modes, reconstructing the target fault transient feature vector, calculating the target local sensitive feature hash value, approximating it in the fault transient feature hash table, and obtaining the test batch identifier and relative time coordinate set corresponding to the candidate data points; S3) Constructing a query combination key range based on the test batch identifier and relative time coordinates combined with hierarchical intervals, performing interval retrieval in a multidimensional combination key B+ tree to locate the target leaf node, performing similarity verification based on the hash value embedded in the leaf node, and combining the constraint verification results of the actual diagnostic parameter measurement values ​​in the candidate data blocks to determine the target test data.

2. The method according to claim 1, characterized in that, The leaf nodes store address pointers to the original data blocks and embed hash values ​​of local sensitive features representing transient fault characteristics, including: Convert the test batch identifier to a fixed-length hexadecimal string; Extract the absolute timestamp of the raw data and subtract it from the system's zero-point reference time to obtain the relative time coordinates; The measured values ​​of key diagnostic parameters are compared with preset segment boundary intervals, and the corresponding grading levels are assigned based on the comparison results. The fixed-length hexadecimal string, relative time coordinate, and hierarchical level are concatenated in order to form a key, which is then stored in the leaf node of the multidimensional key B+ tree.

3. The method according to claim 1, characterized in that, The calculation of the local sensitive feature hash value and the establishment of the mapping relationship between the local sensitive feature hash value and the test batch identifier and relative time coordinate include: A preset time window is formed by extracting a signal sequence of a fixed number of clock cycles before and after the time when the preset fault code occurs. The discrete difference quotient between current and voltage between adjacent sampling points is calculated as a discrete slope sequence; After the discrete slope sequence is dimensionless, it is arranged in chronological order to form a two-dimensional matrix. After discretization and dimensionality reduction, a predefined locality-sensitive hash function is used to calculate and generate locality-sensitive feature hash values.

4. The method according to claim 1, characterized in that, The extraction of specific fault modes includes: Extract the start timestamp, end timestamp, test batch identifier boundary code, diagnostic parameter constraints, and fault mode identifier from the composite query request to determine the corresponding time interval, test batch range, and specific fault mode.

5. The method according to claim 4, characterized in that, The process of reconstructing the target fault transient feature vector and calculating the target local sensitive feature hash value includes: The preset fault mode standard waveform template is invoked according to the specific fault mode, and the target fault transient feature vector is generated according to the same feature processing rules as the construction of the fault transient feature hash table, and the target local sensitivity feature hash value is calculated. The time interval, test batch range, diagnostic parameter constraints, specific fault modes, and target local sensitive feature hash values ​​are encapsulated into a query task package and added to the task queue of the retrieval scheduling controller.

6. The method according to claim 1, characterized in that, The process of obtaining the test batch identifier and relative time coordinate set corresponding to the candidate data points includes: Invoke the target local sensitive feature hash value generated based on the specific failure mode; In the fault transient feature hash table, approximate collision matching is performed based on a preset hash distance threshold to parse out the corresponding test batch identifier bound to the underlying record of the hit local sensitive feature hash value and the relative time coordinate mapping pair. The mapping pairs are filtered based on the time interval and test batch range to obtain the test batch identifier and relative time coordinate set corresponding to the candidate data points.

7. The method according to claim 1, characterized in that, The construction of the query composite key range includes: Parse the upper and lower limits of diagnostic parameter constraints in the composite query request and map them to the corresponding hierarchical level range; The obtained test batch identifier, relative time coordinate, and grade range are combined to generate a range of query combination keys.

8. The method according to claim 1, characterized in that, The step of performing an interval search to locate the target leaf node in a multidimensional composite key B+ tree includes: Using the range of query composite keys as routing credentials, the multidimensional composite key B+ tree is traversed downwards from the root node and range search is performed to locate the target leaf node.

9. The method according to claim 1, characterized in that, The determination of the target test data includes: Extract the hash value of the local sensitive features embedded in the target leaf node; The similarity of the target's local sensitive feature hash value with the local sensitive feature hash value embedded in the target's leaf node is compared. If the similarity between the hash values ​​of local sensitive features meets the preset similarity condition, and the actual diagnostic parameter measurement values ​​in the candidate data block meet the diagnostic parameter constraints, then the data block reading instruction is triggered to extract the target test data.

10. The method according to claim 3, characterized in that, The predefined locality-sensitive hash function is the SimHash function.

Images