Vehicle parts matching methods, devices, electronic equipment and storage media

Abstract

This invention discloses a method, apparatus, electronic device, and storage medium for matching vehicle parts. The method includes: acquiring vehicle diagnostic data of a target vehicle; inputting the vehicle diagnostic data into a first matching model and a second matching model respectively to obtain a first parts list output by the first matching model and a second parts list output by the second matching model; determining a first confidence level for each parts in the first parts list based on the first parts information of each parts in the first parts list; determining a second confidence level for each parts in the second parts list based on the second parts information of each parts in the second parts list; and determining the target parts required by the target vehicle based on the first confidence level and the second confidence level of each parts in the first and second parts lists. Using this invention, the matching accuracy of vehicle parts can be improved when a vehicle malfunctions.

Description

Technical Field

[0001] This invention relates to the field of vehicle diagnostic technology, and in particular to a method, apparatus, electronic device, and storage medium for matching vehicle parts. Background Technology

[0002] With the continuous increase in global car ownership, the demand for parts matching in the auto repair market is growing. Accurate and efficient parts matching has become a core element in improving repair efficiency and reducing user decision-making costs. Currently, the mainstream parts matching technologies in the auto repair market are mainly divided into two categories: one is the traditional solution based on database queries, and the other is the intelligent matching solution based on a single artificial intelligence model.

[0003] However, both approaches suffer from significant technical bottlenecks, making it difficult to meet matching needs in complex scenarios. For example, database queries rely on a single correspondence, struggling to handle complex situations such as part aliases, cross-model common parts, and model year changes, resulting in insufficient flexibility and generalization ability. Furthermore, individual models have limitations in knowledge coverage, training data quality, and inference capabilities. When faced with fuzzy queries, complex scenarios, or rare end-of-supply vehicle parts in the training data, a single model can easily produce biased, inaccurate, or even completely erroneous matching results. This leads to low accuracy in matching vehicle parts during vehicle malfunctions. Summary of the Invention

[0004] To address the aforementioned problems, embodiments of the present invention provide a vehicle parts matching method, apparatus, electronic device, and storage medium, which can improve the matching accuracy of vehicle parts when a vehicle malfunctions.

[0005] In a first aspect, embodiments of the present invention provide a method for matching vehicle parts, including: Obtain vehicle diagnostic data for the target vehicle; the vehicle diagnostic data includes at least one of the following: vehicle identification number, fault code, and problem description text; The vehicle diagnostic data is input into the first matching model and the second matching model respectively to obtain the first spare parts list output by the first matching model and the second spare parts list output by the second matching model; the model parameters of the first matching model and the model parameters of the second matching model are different. Based on the first component information of each component in the first component list, determine the first confidence level of each component in the first component list; Based on the second component information of each component in the second component list, determine the second confidence level of each component in the second component list; The target parts required for the target vehicle are determined based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list.

[0006] Secondly, embodiments of the present invention provide a vehicle parts matching device, the device comprising an acquisition unit and a processing unit; The acquisition unit is used to acquire vehicle diagnostic data of the target vehicle; the vehicle diagnostic data includes at least one of the following: vehicle identification number, fault code, and problem description text; The processing unit is used to input the vehicle diagnostic data into a first matching model and a second matching model respectively, to obtain a first spare parts list output by the first matching model and a second spare parts list output by the second matching model; the model parameters of the first matching model and the model parameters of the second matching model are different. Based on the first component information of each component in the first component list, determine the first confidence level of each component in the first component list; Based on the second component information of each component in the second component list, determine the second confidence level of each component in the second component list; The target parts required for the target vehicle are determined based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list.

[0007] Thirdly, embodiments of the present invention provide an electronic device, the electronic device including a processor and a memory, the processor being connected to the memory, the memory being used to store a computer program, and the processor being used to execute the computer program stored in the memory, so that the electronic device performs the method as described in the first aspect.

[0008] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that is executed by a processor to implement the method described in the first aspect.

[0009] Fifthly, embodiments of this application provide a computer program product, the computer program product including a non-transitory computer-readable storage medium storing a computer program, the computer being operable to perform the method as described in the first aspect.

[0010] Implementing the embodiments of this application has the following beneficial effects: In this embodiment, vehicle diagnostic data of the target vehicle is first acquired. This vehicle diagnostic data includes at least one of the following: vehicle identification number (VIN), fault code, and problem description text. Then, the vehicle diagnostic data is input into a first matching model and a second matching model, respectively, to obtain a first parts list output by the first matching model and a second parts list output by the second matching model. The model parameters of the first and second matching models are different. Based on the first parts information of each part in the first parts list, a first confidence level is determined for each part in the first parts list. Based on the second parts information of each part in the second parts list, a second confidence level is determined for each part in the second parts list. Finally, based on the first and second confidence levels of each part in the first and second parts lists, the target parts required by the target vehicle are determined. Therefore, by using two parts matching models with different model parameters, two different parts lists are output for the vehicle diagnostic data, and the target parts required by the target vehicle are determined based on the confidence level of each part. This differs from traditional techniques and improves the matching accuracy of vehicle parts when a vehicle malfunctions. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the background art, the drawings used in the embodiments of the present invention or the background art will be described below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the architecture of a vehicle parts matching system provided in an embodiment of this application; Figure 2 This is a flowchart of a vehicle parts matching method provided in an embodiment of this application; Figure 3 This is a module architecture diagram of a diagnostic device provided in an embodiment of this application; Figure 4 This is a schematic diagram of an interactive interface provided in an embodiment of this application; Figure 5 This is a schematic diagram of a spare parts display interface provided in an embodiment of this application; Figure 6 This is a schematic diagram of a target component display interface provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a vehicle parts matching device provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] The technical solutions of 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, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] The terms "first," "second," "third," and "fourth," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to these processes, methods, products, or devices.

[0015] In this document, the term "embodiment" means that a particular feature, result, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0016] The following describes the relevant content, concepts, technical issues, technical solutions, and beneficial effects involved in the embodiments of this application.

[0017] First, let me explain some of the technical terms used in this application: Long-tail vehicle parts: These are spare parts that are suitable for niche market models, small imported models, old and discontinued models, or special-purpose models. These spare parts correspond to the "long-tail demand" in the automotive aftermarket, which is characterized by small scale, scattered distribution, and narrow audience. They are at the end of the automotive spare parts supply chain.

[0018] Artificial Intelligence (AI) large-scale models refer to AI models trained on massive amounts of multi-source data, possessing a very large parameter scale, and capable of generalizing to multiple tasks. Their core characteristic is that through deep neural network architectures such as Transformer, they learn complex patterns and relationships in data, adapting to diverse scenarios such as natural language processing, image recognition, and industry knowledge reasoning without requiring extensive fine-tuning for a single task.

[0019] On-Board Diagnostics Interface (OBD): This is a standardized communication interface used in vehicles to connect diagnostic equipment to the vehicle's onboard computer. Its core function is to enable data exchange between the diagnostic equipment and the vehicle's electronic control system, reading key information such as real-time parameters, fault codes, and historical fault records during vehicle operation.

[0020] The Vehicle Identification Number (VIN) is a globally unique 17-character alphanumeric code, equivalent to a vehicle's "ID number." The VIN contains core vehicle identification information, and according to coding rules, it can be deciphered to key parameters such as the vehicle's country of origin, manufacturer, brand, model, year of production, assembly plant, engine model, and body type.

[0021] See Figure 1 , Figure 1 This is a schematic diagram of the architecture of a vehicle parts matching system provided in an embodiment of this application. Figure 1 As shown, the system includes diagnostic equipment and a target vehicle. The target vehicle is the vehicle to be matched with spare parts and is the service object of the system. It stores data such as vehicle identification number (VIN), fault codes, and vehicle model parameters. The diagnostic equipment is the core execution carrier of the system, with built-in functional units such as a vehicle data acquisition module, a dual-matching AI large model module, and a data integration module. The target vehicle and the diagnostic equipment communicate with each other through an OBD interface. The diagnostic equipment can read diagnostic data such as VIN code and fault codes from the target vehicle, providing a basic data source for subsequent spare parts matching.

[0022] See Figure 2 , Figure 2 This is a flowchart of a vehicle parts matching method provided in an embodiment of this application. Figure 2 As shown in the embodiments of this application, a vehicle parts matching method includes, but is not limited to, the following steps: Step S101: Obtain vehicle diagnostic data for the target vehicle; The vehicle diagnostic data includes at least one of the following: vehicle identification number, fault code, or problem description text; Step S102: Input the vehicle diagnostic data into the first matching model and the second matching model respectively to obtain the first spare parts list output by the first matching model and the second spare parts list output by the second matching model. The model parameters of the first matching model are different from those of the second matching model. Step S103: Determine the first confidence level of each component in the first spare parts list based on the first component information of each component in the first spare parts list; Step S104: Determine the second confidence level of each component in the second spare parts list based on the second spare parts information of each component in the second spare parts list; Step S105: Determine the target parts required for the target vehicle based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list.

[0023] Specifically, the model parameters include the model's training data, with the first and second matching models being heterogeneous. By leveraging the differences in model parameters, the two matching outputs are ensured to be complementary in terms of candidate part range, specifications, and applicability judgments, providing a diversified decision-making basis for subsequent intelligent integration. During training, domain-adaptive pre-training technology is employed, injecting professional knowledge such as vehicle parts atlases, repair manuals, vehicle model-part relationships, and fault code-part mapping relationships into the large model base. For the integration model, an additional training method based on contrastive learning is introduced to enable it to accurately grasp the evaluation criteria for part matching quality. Through model pruning and quantization techniques, the three large models are compressed to a scale suitable for offline operation on mobile devices or edge computing units. Knowledge distillation technology is used to significantly reduce the computational and storage overhead of the models while maintaining high accuracy. During application, structured query data packets are simultaneously input into the two heterogeneous matching AI models. Based on their internal knowledge, the two models independently perform reasoning and output matching result lists containing part numbers, names, specifications, and applicability descriptions, respectively.

[0024] Specifically, the first matching model is an AI model trained based on vehicle original parts atlas, official repair manuals, and standardized vehicle model-part association rules. Its model parameter optimization goal is to achieve accurate matching of parts with vehicle model year and configuration parameters. The second matching model is an AI model trained based on vehicle repair fault cases, actual parts transaction records, and technician experience summary data. Its model parameter optimization goal is to achieve efficient matching of parts with fault phenomena and repair scenarios. The model parameters of the second matching model and the first matching model are significantly different, specifically in the training dataset, feature extraction dimensions, and inference logic.

[0025] Specifically, the first part information consists of data output by the first matching model, including the part's basic identifier, technical parameters, vehicle model compatibility attributes, and information source. This includes part number, dimensions, compatible vehicle model year, compatible engine model, and information source type. The second part information consists of data output by the second matching model, including the part's basic identifier, technical parameters, fault compatibility attributes, and information source. This includes part number, compatible fault type, actual repair / replacement frequency, and information source type.

[0026] Specifically, the first confidence level is a quantitative indicator representing the compatibility between the first spare part and the target vehicle, calculated by weighting the information based on three dimensions: vehicle model year matching degree, number of parameter items, and information source type. The second confidence level is a quantitative indicator representing the compatibility between the second spare part and the target vehicle, calculated by weighting the information based on three dimensions: vehicle model year matching degree, number of parameter items, and information source type.

[0027] In one possible embodiment, vehicle diagnostic data of the target vehicle is acquired. This vehicle diagnostic data includes at least one of the following: vehicle identification number (VIN), fault codes, and problem description text. Specifically, structured diagnostic data such as the VIN and fault codes of the target vehicle are read through the vehicle's OBD interface. Simultaneously, a problem description text input by the user is received through a human-machine interface. This problem description text is a textual description of the vehicle's fault symptoms, such as cold start vibration or weak acceleration. The acquired data is then parsed and standardized. The VIN is parsed into core information such as vehicle model, year, and configuration version, and the fault codes are parsed into corresponding fault type descriptions, ultimately forming a unified diagnostic data set.

[0028] In one possible embodiment, vehicle diagnostic data is input into a first matching model and a second matching model, respectively, to obtain a first parts list output by the first matching model and a second parts list output by the second matching model; the model parameters of the first matching model and the second matching model are different. Specifically, the obtained standardized vehicle diagnostic data is synchronously input into the first matching model and the second matching model; since the model parameters of the two models are different, the first matching model infers and outputs a first parts list adapted to the target vehicle based on the original factory data, and the second matching model infers and outputs a second parts list matching the fault phenomena of the target vehicle based on fault case data. The parts in the two lists may overlap or differ.

[0029] In one possible embodiment, the first confidence level of each component in the first spare parts list is determined based on the first spare parts information of each component in the first spare parts list. Specifically, for each component in the first spare parts list, its corresponding first spare parts information is extracted, and scores are calculated for three dimensions: vehicle model year matching degree, number of parameter items, and information source type. The vehicle model year matching degree score is determined by comparing the consistency between the vehicle model year, engine model, and vehicle identification number parsing information of the component. The number of parameter items score is determined by statistically analyzing the completeness of the core parameters of the component. The information source type score is determined based on the authority of the source of the component information. The scores of the three dimensions are weighted and summed according to preset weights to obtain the first confidence level of each component. The preset weights can be 0.5 for vehicle model year matching degree, 0.2 for number of parameter items, 0.3 for information source type, etc.

[0030] In one possible embodiment, a second confidence level is determined for each component in the second spare parts list based on its second spare parts information. Specifically, for each component in the second spare parts list, its corresponding second spare parts information is extracted, and the same dimensional scoring framework as the first confidence level is adopted, but the scoring criteria for each dimension are differentiated. The vehicle model year matching score is determined based on the actual replacement frequency of the component in similar vehicle model failure cases; the parameter item quantity score is determined based on the completeness of the core parameters covering the repair scenario of the component; and the information source type score is determined based on the case verification pass rate corresponding to the component information. These scores are then weighted and summed according to the same weight ratio to obtain the second confidence level for each component.

[0031] In one possible embodiment, the target parts required by the target vehicle are determined based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list. Specifically, integrating the first confidence level of the first parts list and the second confidence level of the second parts list can extract the parts that intersect the two lists, weight and fuse their first and second confidence levels to increase the weight, highlighting the reliability of the dual-model consensus; alternatively, it can integrate the parts that are the union of the two lists, retaining the high-confidence parts recommended by the single model to fill in any gaps; or it can identify related parts combinations that need to be replaced together, adjust the confidence levels of the parts within the combination, and label the relationships; finally, parts with confidence levels higher than a preset threshold are selected to generate the target parts list required by the target vehicle.

[0032] In this embodiment, by using parallel matching of two models with differentiated parameters, combined with multi-dimensional confidence calculation and intelligent integration strategies, the knowledge blind spots of a single model are effectively avoided. This ensures that the matching results of spare parts are consistent with the original factory standards, while also meeting the needs of actual repair scenarios, thereby improving the accuracy and reliability of spare parts matching.

[0033] Furthermore, by enhancing intersection weights, filling gaps in union data, and identifying associated parts, accurate matching of multiple types of parts in complex fault scenarios is achieved, providing efficient and comprehensive decision support for vehicle maintenance.

[0034] Optionally, the first spare part information includes: compatible vehicle type, number of initial parameter items, number of actual parameter items, and information source type; step S103, determining the first confidence level of each spare part in the first spare part list based on the first spare part information of each spare part in the first spare part list, may include the following steps: Step S201: Obtain the vehicle type of the target vehicle; Step S202: Compare the vehicle type of the target vehicle with the compatible vehicle type of each part in the first part list to obtain the matching accuracy score of each part in the first part list. Step S203: Determine the parameter completeness score for each component in the first component list based on the initial number of parameter items and the actual number of parameter items for each component in the first component list; Step S204: Determine the source accuracy score for each part in the first spare parts list based on the information source type of each part in the first spare parts list; Step S205: Determine the first confidence level of each component in the first component list based on the matching accuracy score, parameter completeness score, and source accuracy score of each component in the first component list.

[0035] Specifically, the compatible vehicle type refers to the set of vehicle brands, series, models, and configuration versions that the part can be compatible with, as shown in the first part information. This is used to determine whether the part is compatible with the target vehicle. The initial number of parameter items is a baseline number of parameters set for a certain type of part based on the vehicle's original equipment manufacturer (OEM) parts standards and specifications. This is used to measure whether the part's parameters are complete. The actual number of parameter items is the number of parameters actually included in the first part information output by the first matching model. This reflects the actual coverage of the part's parameters.

[0036] In one possible embodiment, the vehicle type of the target vehicle is obtained. This is achieved by parsing the vehicle identification number (VIN). According to preset VIN parsing rules, information such as the brand, series, production year, engine model, and configuration version of the target vehicle is extracted and integrated into standardized target vehicle type data.

[0037] In one possible embodiment, the vehicle type of the target vehicle is compared with the compatible vehicle type of each part in the first parts list to obtain a matching accuracy score for each part in the first parts list. Specifically, multi-dimensional comparison weights are set, with brand and series matching having the highest weight, followed by model year and engine model matching, and configuration version matching having a supplementary weight. If the compatible vehicle type of the part completely includes all the dimensional information of the target vehicle type, the matching accuracy score is full. If some dimensions do not match, the score is deducted according to the weight ratio of the mismatched dimensions. If core dimensions such as brand and series do not match, the matching accuracy score is zero.

[0038] In one possible embodiment, the parameter completeness score for each component in the first spare parts list is determined based on the initial number of parameter items and the actual number of parameter items for each component. Specifically, firstly, the parameter completeness ratio is calculated, i.e., parameter completeness ratio = actual number of parameter items ÷ initial number of parameter items; then, this ratio is mapped to a percentage score. If the ratio is 1, the parameter completeness score is full marks; if the ratio is between 0 and 1, the score decreases linearly with the ratio; if the ratio is 0 or the actual number of parameter items is zero, the parameter completeness score is zero; simultaneously, if there is a conflict between the actual parameters and the original factory standard parameters, a preset score is deducted from the above score.

[0039] In one possible embodiment, the source accuracy score for each part in the first parts list is determined based on the information source type of each part in the first parts list. Specifically, the authority level and corresponding score of the information source type are preset, where information types from the vehicle OEM's original parts catalog and official repair manuals correspond to the highest score; information types from the national automotive parts industry standard database correspond to the second highest score; information types from third-party non-authoritative parts catalogs correspond to the basic score; and information types without a clear source receive a score of zero. The corresponding score is automatically matched according to the information source type of the first parts to obtain the source accuracy score.

[0040] In one possible embodiment, the first confidence level of each component in the first parts list is determined based on its matching accuracy score, parameter completeness score, and source accuracy score. Specifically, weights for the three dimensions are pre-defined, with the matching accuracy score having the highest weight, followed by the parameter completeness score, and the source accuracy score having a supplementary weight. The first confidence level of each component in the first parts list is calculated using the weighted summation formula: First Confidence Level = Matching Accuracy Score × Weight 1 + Parameter Completeness Score × Weight 2 + Source Accuracy Score × Weight 3. This confidence level quantifies the degree of compatibility between the component and the target vehicle, as well as the reliability of the information.

[0041] In this embodiment, the calculation of the first confidence level is decomposed into three quantifiable scoring dimensions, and the scores of each dimension are based on the comparison of objective parts information and vehicle data. This avoids the bias of subjective reasoning by a single model and significantly improves the objectivity and accuracy of the first confidence level assessment results.

[0042] Optionally, step S105, determining the target parts required by the target vehicle based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list, may include the following steps: Step S301: Obtain the intersection of the first spare parts list and the second spare parts list; Step S302: Determine the first weighting coefficient for the intersecting parts; Step S303: Based on the first weighting coefficient, the first confidence level and the second confidence level of the intersection parts are weighted and calculated to obtain the first fusion confidence level of the intersection parts; Step S304: Determine the target component based on the first fusion confidence level and the first confidence threshold of the intersection components.

[0043] In one possible implementation, the intersection of the first and second parts lists is obtained. Specifically, the part type identifiers and core compatibility attributes of all parts in both lists are first extracted. A precise matching algorithm is then used to filter out parts with the same part type and overlapping core compatibility attributes, such as compatible fault type and compatible vehicle model year. These parts are identified as the intersection parts. Parts existing only in a single list are temporarily excluded from this process, ensuring that the intersection parts represent the consensus recommendations of both models for the target vehicle's faults.

[0044] In one possible embodiment, a first weighting coefficient for the intersecting parts is determined. The first weighting coefficient includes a first weight and a second weight. The first weight corresponds to a first confidence level, and the second weight corresponds to a second confidence level. This weighting coefficient is used to balance the weight ratio of the first confidence level and the second confidence level in the fusion calculation. Specifically, a fixed weighting coefficient can be set according to the training data characteristics and historical matching accuracy of the first matching model and the second matching model, or the weighting coefficient can be dynamically adjusted according to the fault type of the target vehicle. For example, when the fault of the target vehicle is a known typical fault, the first weight and the second weight in the first weighting coefficient can be set to be the same, that is, each accounts for 50%. When the fault of the target vehicle is more inclined to be a new type of fault and lacks original factory adaptation rules, the weight ratio corresponding to the second confidence level can be appropriately increased to improve the adaptability of the fusion confidence level to the actual repair scenario. The value range of each weight in the first weighting coefficient is limited to 0 to 1, and the sum of the first weight and the second weight is 1.

[0045] In one possible embodiment, the first confidence level and the second confidence level of the intersecting parts are weighted and calculated according to a first weighting coefficient to obtain the first fused confidence level of the intersecting parts. Specifically, for each intersecting part, its first confidence level corresponding to the first part list and its second confidence level corresponding to the second part list are retrieved and substituted into a preset weighted fusion formula: First fused confidence level = First confidence level × First weight + Second confidence level × Second weight, and the fused confidence level of the intersecting part is obtained by calculation. This fused confidence level integrates the evaluation results of the two models and, compared with the confidence level output by a single model, can more objectively reflect the degree of fit between the parts and the target vehicle fault.

[0046] In one possible embodiment, target parts are determined based on a first fusion confidence level and a first confidence threshold of the intersecting parts. Specifically, a first confidence threshold is preset, which is the minimum standard for determining whether a part has recommendation value, and its value can be determined based on the statistical analysis results of a large amount of historical matching data. The first fusion confidence level of each intersecting part is compared with the first confidence threshold, and intersecting parts with a first fusion confidence level higher than or equal to the first confidence threshold are selected and directly identified as the target parts required by the target vehicle. For intersecting parts with a first fusion confidence level lower than the first confidence threshold, their compatibility reliability is deemed insufficient, and they are temporarily not included in the target parts list.

[0047] In this embodiment, by extracting the intersection of parts recommended by the two models and performing weighted fusion calculation, the synergistic advantages of the two models can be fully utilized, effectively filtering out low-reliability parts recommended by a single model, and significantly improving the accuracy and reliability of the target parts matching results. At the same time, the flexible setting of the weighting coefficients can adapt to the matching requirements of different fault scenarios, further enhancing the versatility of the method.

[0048] Optionally, step S105, determining the target parts required by the target vehicle based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list, may include the following steps: Step S401: Obtain the union of the first spare parts list and the second spare parts list; Step S402: Determine the second weighting factor for the union of components; Step S403: Based on the second weighting coefficient, the first confidence and the second confidence of the union parts are weighted and calculated to obtain the second fusion confidence of the union parts; Step S404: Determine the target component based on the second fusion confidence level and the second confidence threshold of the union components.

[0049] In one possible embodiment, the union of the first and second parts lists is obtained. Specifically, all parts in both lists are first extracted, and duplicate checks are performed based on key information such as part number, part type, and core compatibility attributes. For parts of the same type that exist in both lists, only one record is retained. For parts that exist only in the first or only in the second list, all are retained. This results in a union set of parts containing all recommendations from both models, thus achieving comprehensive coverage of parts related to the target vehicle's malfunction.

[0050] In one possible embodiment, a second weighting coefficient for the parts is determined and set. The second weighting coefficient includes a third weight and a fourth weight, where the third weight corresponds to the first confidence level, and the fourth weight corresponds to the second confidence level. Specifically, the weights are set differently based on the technical characteristics of the first and second matching models: for the first matching model driven by original equipment manufacturer (OEM) data, if the target vehicle is a mainstream model currently on sale, the weight percentage corresponding to the first confidence level can be appropriately increased; for the second matching model driven by fault cases, if the target vehicle's fault is a rare fault or a fault lacking OEM adaptation rules, the weight percentage corresponding to the second confidence level can be appropriately increased; the value range of each weight in the second weighting coefficient is limited to 0 to 1, and the sum of the third and fourth weights is 1.

[0051] In one possible embodiment, the first and second confidence levels of the union of parts are weighted according to a second weighting coefficient to obtain a second fused confidence level for the union of parts. Specifically, for each item in the union of parts, the process is categorized as follows: for parts that exist in both lists, their corresponding first and second confidence levels are retrieved and substituted into the weighted fusion formula for calculation; for parts that exist only in a single list, their corresponding confidence level is directly assigned as the second fused confidence level. The weighted fusion formula is uniformly set as: second fused confidence level = first confidence level × third weight + second confidence level × fourth weight. Through the above calculation, the standardization of the confidence levels of all items in the union of parts is achieved.

[0052] In one possible embodiment, target parts are determined based on a second fusion confidence level and a second confidence threshold of the union of parts. Specifically, a second confidence threshold is preset, which is lower than a first confidence threshold, thereby effectively identifying potential compatible parts. The second fusion confidence level of each union of parts is compared with the second confidence threshold, and union of parts with a second fusion confidence level higher than or equal to the second confidence threshold are selected and identified as target parts required by the target vehicle. For union of parts with a second fusion confidence level lower than the second confidence threshold, their compatibility reliability is deemed insufficient, and they are temporarily excluded from the target parts list.

[0053] Optionally, for the union of parts that exist simultaneously in the first and second parts lists: the third and fourth weights are dynamically assigned based on the historical matching accuracy and data authority of the first and second matching models. For example, when the matching accuracy of the first matching model for original equipment manufacturer (OEM) vehicles is higher than that of the second matching model, the third weight is set to 60% and the fourth weight to 40%; when the recall rate of the second matching model for long-tail fault cases is higher than that of the first matching model, the third weight is set to 40% and the fourth weight to 60%. For unique union of parts that exist only in the first or second parts lists, basic weights are first set, with the weight of the corresponding model confidence set to 100% by default, and the weight of the other model confidence set to 0. That is, for parts that exist only in the first list, the initial third weight = 1 and the fourth weight = 0; for parts that exist only in the second list, the initial third weight = 0 and the fourth weight = 1. Then, a flexible weight correction rule is preset, introducing a confidence correction coefficient α, with a value range of 1.0-1.2. This coefficient is dynamically adjusted based on the model's historical performance in the corresponding scenario. If the model belonging to a unique type of spare part has a matching success rate greater than or equal to a preset standard (which can be 90%) for the same type of fault or vehicle model, then the weight corresponding to that model is multiplied by the correction coefficient α, at which point the weight can exceed the limit of 1. If the model's matching success rate is lower than the preset standard, then the weight remains unchanged at 1, without additional correction. The corrected weights are constrained so that the corrected weights do not exceed 1.2, and the weight of the other model remains 0, thus avoiding confidence distortion caused by excessive weight amplification.

[0054] Optionally, a second fusion confidence score is obtained by weighting the first and second confidence scores of the union of parts based on the second weighting coefficient. A unified weighted fusion formula is used: Second Fusion Confidence Score = First Confidence Score × Third Weight + Second Confidence Score × Fourth Weight. This calculation is performed for different types of parts within the union. For parts in the intersection union, the corresponding first and second confidence scores are retrieved and substituted into the formula for calculation, taking into account the evaluation results of both models and enhancing the reliability of the matching. For unique class union parts, if the part exists only in the first list and the model performance meets the requirements: the third weight = 1 × α, α ∈ [1.0, 1.2], the fourth weight = 0, and the second fusion confidence = the first confidence × α; if the part exists only in the second list and the model performance meets the requirements: the fourth weight = 1 × α, α ∈ [1.0, 1.2], the third weight = 0, and the second fusion confidence = the second confidence × α; if the model performance does not meet the requirements: the weight remains unchanged at 1, and the second fusion confidence is directly equal to the confidence of the corresponding model. This calculation method achieves both weight bias for high-reliability unique class parts and ensures the uniformity and comparability of the confidence calculation rules for all union parts. By introducing a confidence correction coefficient, the limitation that the weight of unique spare parts can only be 1 is broken, and the weight of high-reliability unique spare parts can be flexibly increased. This not only highlights the value of high-quality unique matching results, but also avoids the solidification of weight in the recommendation results of a single model. At the same time, the unified weighted calculation formula and weight upper limit constraint ensure the scientific nature of confidence calculation, effectively improve the accuracy of the selection of union spare parts, and further enhance the adaptability of the method to complex fault scenarios.

[0055] In this embodiment, by extracting the union of the recommended parts from the dual-model recommendations and performing weighted fusion calculations, the potential compatible parts recommended by a single model can be effectively covered, avoiding the omission of effective information due to dual-model consensus filtering. At the same time, the combination of differentiated weighting coefficient setting and hierarchical threshold screening not only ensures the reliability of the target parts, but also improves the method's adaptability to complex fault scenarios, providing more comprehensive decision support for vehicle maintenance.

[0056] Optionally, step S105, determining the target parts required by the target vehicle based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list, may include the following steps: Step S501: Obtain the associated parts combinations from the first parts list and the second parts list; Step S502: Determine the third weighting coefficient for each component in the associated component combination; Step S503: Based on the third weighting coefficient of each component in the associated component combination, the first confidence level and the second confidence level of each component in the associated component combination are weighted and calculated to obtain the third fusion confidence level of each component in the associated component combination. Step S504: Determine the target component based on the third fusion confidence level and the third confidence level threshold of each component in the associated component combination.

[0057] In one possible embodiment, associated parts combinations from a first parts list and a second parts list are obtained. Specifically, a preset parts association rule base is retrieved. This rule base is constructed based on original equipment manufacturer (OEM) joint repair specifications, fault case statistics, and technician experience, and includes typical associated combinations such as air flow meter-front oxygen sensor and spark plug-ignition coil. The first and second parts lists are scanned respectively, and parts combinations that meet the association rules are extracted. At the same time, the association combination information of the two lists is cross-compared. If the same associated combination is recommended by two models, it is marked as a high-priority associated combination. If only a single model recommends it, the association is verified in conjunction with the fault phenomena in the vehicle diagnostic data. After the verification is passed, it is included in the associated parts combination set. It should be noted that the parts in the associated parts combination belong to different types and have direct functional or fault-inducing associations, which is fundamentally different from the same type judgment criteria for overlapping parts.

[0058] In one possible embodiment, a third weighting coefficient is determined for each component in the associated component group. The third weighting coefficient includes a fifth weight and a sixth weight, whereby the fifth weight corresponds to the first confidence level and the sixth weight corresponds to the second confidence level. This weighting coefficient is used for differentiated allocation, that is, setting different weights based on the functional importance and failure contribution of the component in the associated group, rather than using a uniform weight ratio. Specifically, based on the pre-defined combination roles in the parts association rule base, the parts within the associated combination are divided into core parts and auxiliary parts. For core parts, the weight ratios of the first and second confidence levels in the third weighting coefficient can be appropriately increased. For example, the weight ratio of core parts is set to 40% for the fifth weight of the first confidence level and 60% for the sixth weight of the second confidence level to strengthen the adaptation priority of core components. For auxiliary parts, the weight ratio can be set to 50% for the fifth weight of the first confidence level and 50% for the sixth weight of the second confidence level to ensure the objectivity of the evaluation. The value range of each weight in the third weighting coefficient is limited to 0 to 1, and the sum of the fifth and sixth weights of a single part is 1, thereby achieving precise weight allocation for each part within the associated combination.

[0059] In one possible embodiment, based on the third weighting coefficient of each component in the associated component combination, the first confidence level and the second confidence level of each component in the associated component combination are weighted and calculated to obtain the third fused confidence level of each component in the associated component combination. Specifically, for each component in the associated combination, its first confidence level corresponding to the first component list and its second confidence level corresponding to the second component list are retrieved and substituted into the weighted fusion formula: Third fused confidence level = First confidence level × Fifth weight + Second confidence level × Sixth weight, to complete the fused confidence level calculation of a single component; for associated components that exist only in a single list, their corresponding confidence level is directly assigned as the third fused confidence level; through differentiated weighted calculation, the adaptation priority of each component in the associated combination can be accurately distinguished.

[0060] In one possible embodiment, target parts are determined based on the third fusion confidence level and the third confidence threshold of each part in the associated parts combination. Specifically, a third confidence threshold is preset, which is lower than the first confidence threshold of the intersection parts and higher than the second confidence threshold of the union parts, thereby balancing the mining efficiency and matching reliability of associated combinations. The third fusion confidence level of each part in the associated combination is compared with the third confidence threshold. If the third fusion confidence level of all parts in the combination is higher than the threshold, the entire associated combination is included in the target parts list and marked "Recommended for joint replacement" and the basis for the combination. If only some parts in the combination have a third fusion confidence level higher than the threshold, the qualified parts are retained, and their associated parts information is noted for maintenance personnel to refer to. If the third fusion confidence level of all parts in the combination is lower than the threshold, the associated combination is determined to have insufficient matching reliability and is temporarily not included in the target parts list.

[0061] In this embodiment, by accurately identifying the combination of related parts that need to be replaced together, and using differentiated weighting coefficients to calculate the fusion confidence, it can effectively cover the matching needs of multiple types of parts in complex vehicle fault scenarios, and avoid fault recurrence caused by missing related parts; at the same time, the hierarchical screening and labeling mechanism of related combinations can provide repair personnel with more guiding decision-making basis, significantly improving the efficiency of vehicle repair and the first-time repair rate.

[0062] Optionally, step S105, determining the target parts required by the target vehicle based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list, may include the following steps: Step S601: If the first confidence level of each part in the first part list is less than the fourth confidence level threshold, and the second confidence level of each part in the second part list is less than the fifth confidence level threshold, obtain the candidate part database of the target vehicle. Step S602: Based on the candidate parts database and vehicle diagnostic data, match the target parts required for the target vehicle.

[0063] In one possible embodiment, if the first confidence level of each component in the first component list is less than a fourth confidence threshold, and the second confidence level of each component in the second component list is less than a fifth confidence threshold, a candidate component database for the target vehicle is obtained. Specifically, a fourth confidence threshold and a fifth confidence threshold are preset, where the fourth confidence threshold is the minimum confidence standard for determining the validity of the first component list, and the fifth confidence threshold is the minimum confidence standard for determining the validity of the second component list, and both thresholds are lower than the confidence thresholds corresponding to intersection, union, and association combinations. The first confidence levels of all components in the first component list are iterated over; if all first confidence levels are less than the fourth confidence threshold, the output result of the first matching model is determined to be invalid. Simultaneously, the second confidence levels of all components in the second component list are iterated over; if all second confidence levels are less than the fifth confidence threshold, the output result of the second matching model is determined to be invalid. When both model outputs are deemed invalid, a fallback matching mechanism is automatically triggered, which retrieves a pre-set database of candidate spare parts for the target vehicle. This database is a comprehensive collection of spare parts data covering multiple brands, models, and fault types, including original parts data, third-party compliant parts data, and historical data related to difficult fault matching cases. Furthermore, the spare parts in the database are associated with structured information such as vehicle model compatibility attributes and fault matching tags, providing data for fallback matching.

[0064] In one possible embodiment, the target parts required by the target vehicle are matched based on a candidate parts database and vehicle diagnostic data. Specifically, the vehicle diagnostic data is deeply analyzed to extract core feature information such as the target vehicle's brand, model year, engine model, fault codes, and fault descriptions, and this information is converted into standardized search keywords. A retrieval algorithm combining fuzzy matching and exact matching is used to search the candidate parts database: the exact matching step filters out a subset of fully compatible parts based on deterministic features such as vehicle model and year obtained from vehicle identification code parsing; the fuzzy matching step filters out a subset of parts potentially related to the target fault based on features such as fault codes and fault descriptions through semantic similarity calculation and fault association rule matching. The two subsets are then merged and deduplicated to obtain a preliminary list of parts for fallback matching. Parts with questionable compatibility are then removed through manual verification or a secondary screening based on historical matching success rates, ultimately determining the target parts required by the target vehicle. Meanwhile, the relevant data from this last-resort matching, including vehicle diagnostic data, matching process, and final parts results, will be structured and stored in the model training knowledge base for subsequent iterative optimization of the first and second matching models.

[0065] In this embodiment, by setting a fallback guarantee process, the extreme scenario where the confidence levels of both model outputs are below standard is effectively solved, avoiding the situation where the target part matching result is empty, thus improving the stability and robustness of the entire matching method. At the same time, the multi-source data characteristics and hybrid retrieval algorithm of the candidate part database ensure the accuracy of the fallback matching result, while the backflow storage mechanism of the matching data can continuously optimize the performance of the two models, realizing the self-iteration and upgrading of the matching method.

[0066] Optionally, the following steps may also be included: Step S701: Obtain the vehicle parts catalog, repair manual, and vehicle model-parts association rule library for the target vehicle; Step S702: Train the first candidate matching model based on the vehicle parts atlas, repair manual, and vehicle model-part association rule library to obtain the first matching model; Step S703: Obtain vehicle repair fault cases and actual transaction records of spare parts for the target vehicle; Step S704: Determine the matching label for the parts based on vehicle repair failure cases and actual parts transaction records; Step S705: Train the second candidate matching model based on vehicle repair failure cases, actual parts transaction records and parts matching labels to obtain the second matching model.

[0067] Specifically, the first candidate matching model is a basic AI model that has not been trained on industry data. It possesses general data feature extraction and correlation reasoning capabilities and serves as the foundation for building the first matching model. The second candidate matching model is a basic AI model with the same or different architecture as the first candidate matching model. Initially, it does not possess specialized capabilities for vehicle parts matching and requires training with fault cases and transaction data to achieve functional optimization. The vehicle parts map is a structured, visualized data set of hierarchical relationships, assembly associations, and model matching rules for parts in various vehicle systems. It covers information such as original equipment manufacturer (OEM) codes, compatible vehicle models, and upstream and downstream related parts. The vehicle model-part association rule base is a set of rules built based on a large amount of OEM data to characterize the correspondence between vehicle model year, configuration version, and parts model. It includes forward matching rules and reverse verification rules. Parts matching tags are labeling information based on vehicle repair fault cases and actual parts transaction records. They are generated manually or automatically by algorithms to characterize the fault type-part model matching relationship. They include precise matching tags, high-probability matching tags, and exclusion matching tags.

[0068] In one possible embodiment, the vehicle parts atlas, repair manual, and model-parts association rule base for the target vehicle are obtained. Specifically, this is achieved through methods such as interfacing with the vehicle manufacturer's data interface and authorized access to standardized industry databases. The vehicle parts atlas needs to be structured and converted into a hierarchical data format recognizable by the model. The repair manual needs to extract core content such as fault-parts association sections and parts replacement specifications. The model-parts association rule base is used to formalize the rules, clarify the logical boundaries of compatibility between each model and its parts, and provide a data source for model training.

[0069] In one possible embodiment, a first candidate matching model is trained based on vehicle parts atlas, repair manuals, and a vehicle model-part association rule base to obtain a first matching model. Specifically, structured data is used as the training set and input into the first candidate matching model. The training process adopts a supervised learning approach, with the accurate matching of vehicle model features and parts models as the optimization objective. The model's parameter weights are adjusted through multiple iterations, enabling the model to learn and master the parts matching rules contained in the original equipment manufacturer (OEM) data. During the training process, a validation set and a test set are set. The validation set is used to monitor the model's fitting degree in real time to avoid overfitting. The test set is used to evaluate the model's matching accuracy on unknown vehicle model data. When the model's matching accuracy reaches a preset standard, training is terminated and the model parameters are saved, resulting in a first matching model with OEM data matching capabilities.

[0070] In one possible implementation, vehicle repair fault cases and actual parts transaction records for the target vehicle are obtained. Specifically, by integrating multi-source data such as historical work order data from auto repair companies, transaction records from parts e-commerce platforms, and publicly available fault diagnosis cases in the industry, vehicle repair fault cases and actual parts transaction records for the target vehicle are obtained. The repair fault cases must include complete information such as a description of the fault phenomenon, fault codes, the model of the parts replaced, and the effect of fault resolution. The actual parts transaction records are associated with corresponding vehicle model information and fault background, providing data for the generation of parts matching tags.

[0071] In one possible implementation, parts matching labels are determined based on vehicle repair failure cases and actual parts transaction records. Specifically, for repair failure cases and transaction records, parts matching labels are generated using a combination of manual annotation and automatic algorithm annotation. The manual annotation stage is performed by technicians with professional repair experience, who annotate typical failure cases with corresponding parts matching labels. The automatic algorithm annotation stage uses association rule mining algorithms to analyze atypical cases, generating high-probability matching labels and excluded matching labels. All generated parts matching labels are manually reviewed and verified to ensure that the correspondence between the labels and failure cases is accurate, ultimately forming a training dataset with standardized labels.

[0072] In one possible embodiment, a second candidate matching model is trained based on vehicle repair failure cases, actual parts transaction records, and parts matching tags to obtain a second matching model. Specifically, repair failure cases with parts matching tags and parts transaction records are used as the training set and input into the second candidate matching model. The training process adopts a combination of contrastive learning and supervised learning, with the actual matching rules of failure features and parts models as the optimization objective. This strengthens the model's ability to learn the correlation between failure phenomena and parts replacement, while introducing failure resolution effect data to optimize the model's ability to identify and filter parts with low matching rates. During the training process, a validation set and a test set are also set. The model's learning rate and number of iterations are adjusted through the validation set, and the recall and precision of the model on unknown failure cases are evaluated through the test set. When the model performance reaches the preset standard, the training is terminated and the model parameters are saved, resulting in a second matching model with the ability to match actual repair scenarios.

[0073] In this embodiment, a dual-model differentiated matching capability is constructed by training a first matching model based on original factory standardized data and a second matching model based on actual repair case data. The data sources and optimization objectives of the two models are complementary, which not only ensures the consistency of the parts matching results with the original factory standards, but also meets the needs of actual repair scenarios, laying a solid model foundation for subsequent confidence calculation and result integration.

[0074] In a specific embodiment, the vehicle parts matching method of this solution is based on a layered distributed local architecture. It constructs a full-process technical solution encompassing dual-model parallel matching, integrated model intelligent adjudication, continuous optimization through a feedback mechanism, and multi-level fallback services. This addresses the limitations of existing technologies, such as the constraints of single-model decision-making, lack of self-evolution, and weak ability to handle complex scenarios. All data processing in this embodiment is completed locally on the diagnostic device, balancing matching efficiency, data security, and privacy protection. It can be widely applied in scenarios such as auto repair shops, portable diagnostic devices, and auto parts e-commerce platforms.

[0075] Specifically, see Figure 3 , Figure 3 This is a modular architecture diagram of a diagnostic device provided in an embodiment of this application. The diagnostic device adopts a layered distributed architecture and consists of five core modules: a vehicle data acquisition and preprocessing module, a dual-matching AI large model module, a data integration large model adjudication and optimization module, a feedback and self-learning optimization module, and a multi-level fallback and interactive output module.

[0076] The vehicle data acquisition and preprocessing module serves as the data entry point, responsible for the acquisition, parsing, and standardized packaging of multi-source vehicle diagnostic data.

[0077] This includes multi-source data acquisition, such as collecting static identification information of the target vehicle, including the Vehicle Identification Number (VIN), brand, model, year, and engine model, through the vehicle's OBD standard interface; reading current and historical fault codes and fault statuses, such as P0171 fuel system too lean and P0507 idle speed too high; and receiving problem description text input by the repair technician through the human-machine interface, including fault symptoms such as cold start shaking and weak acceleration, as well as supplementary information such as recent air filter replacement.

[0078] The data preprocessing and standardization process includes, for example, parsing the collected fault codes and converting non-standard fault codes into industry-standard descriptions; normalizing vehicle model year information to eliminate inconsistencies such as "2020 model" and "2020 1.5T". Natural Language Understanding (NLU) technology is used to perform semantic analysis on the user-input problem description text, extracting key entities such as "cold start" and "shaking," and converting them into structured data. All of the above information is integrated into a unified JSON query data package. The data package fields include vehicle identification number (vin_code), brand model (brand_model), fault code (fault_code), fault description (fault_description), and maintenance history (maintenance_history), providing standardized input for subsequent model matching.

[0079] The dual-matching AI large model module consists of two heterogeneous domain large models: the first matching model and the second matching model. The two models are deployed locally and have independent parameters.

[0080] The training process for the first matching model is as follows: Vehicle parts atlases, official OEM repair manuals, and vehicle model-parts association rule bases are acquired for the target vehicle. Domain-adaptive pre-training technology is used to inject the aforementioned standardized OEM data into the base model, optimizing the model's ability to learn the mapping relationships between vehicle model, model year, fault codes, and OEM parts. During training, validation and test sets are set. Training is terminated and parameters are saved when the model achieves an OEM parts matching accuracy of over 95% on the test set.

[0081] The training process for the second matching model is as follows: Vehicle repair fault cases and actual parts transaction records are acquired. A combination of manual and automatic algorithmic annotation is used to generate fault feature-part model matching labels, including precise matching labels and high-probability matching labels. A hybrid training method of contrastive learning and supervised learning is employed to enable the model to learn the correlation between faults and parts in actual repair scenarios. When the model's recall rate for faulty parts on the test set reaches over 92%, training is terminated and the parameters are saved.

[0082] Then, the preprocessed standardized query data package is simultaneously input into both the first and second matching models. The first matching model outputs a first list of spare parts based on the original manufacturer's data, which includes information such as part number, part name, and original manufacturer's compatibility instructions. The second matching model outputs a second list of spare parts based on the fault case data, which includes information such as part number, part name, and actual repair / replacement frequency. The two models reason independently, and their outputs are complementary.

[0083] For the data integration large model adjudication and optimization module, this module does not participate in the initial matching, but only performs multi-dimensional analysis and intelligent fusion of the output results of the dual matching model.

[0084] In the multi-dimensional quality assessment, the integrated model retrieves both a first and a second list of spare parts and quantifies each spare part based on three dimensions: vehicle model year matching, number of parameter items, and information source type. For vehicle model year matching, the model compares the fit of the spare part with the VIN parsing information of the target vehicle; a perfect match scores 100 points, while a mismatch in core dimensions scores 0 points. For the number of parameter items, the model assesses the completeness of core parameters such as part number, compatible fault type, and replacement cycle; complete parameters score 100 points, and missing items are penalized proportionally. For information source type, original factory data scores 100 points, authoritative case library data scores 80 points, and non-authoritative sources score 50 points. The integrated model calculates the confidence level of each spare part based on preset weights, such as 50% for vehicle model year matching, 20% for the number of parameter items, and 30% for the information source type, and identifies intersecting, union, and related spare parts combinations from the two lists.

[0085] When implementing the differentiated fusion strategy, there are three methods: intersection component fusion, union component fusion, and related component fusion. One of these methods can be used, or two of them can be combined. When combined, the first confidence threshold, the second confidence threshold, and the third confidence threshold can be the same.

[0086] For example, for the fusion of overlapping parts, parts of the same type that exist in both lists are extracted and their confidence scores are weighted and fused, with each weight accounting for 50% and the priority increased by 10%-15%, highlighting the reliability of the dual-model consensus. For the fusion of unionized parts, all parts from the two lists are integrated, duplicate entries are removed, and high-confidence parts recommended by a single model are retained to fill in the gaps. For the fusion of related components, different types of component combinations that need to be replaced together are identified, such as air flow meter and pre-oxygen sensor. Differentiated weights are assigned according to the importance of the components in the combination, the fusion confidence is calculated and marked "recommended to replace together".

[0087] Finally, the integrated model filters out parts with a fusion confidence level higher than a preset threshold, sorts them from high to low confidence level, supplements them with information such as part compatibility descriptions and confidence level calculation basis, and generates a final list including the target parts.

[0088] The feedback and self-learning optimization module is used to perform feedback data collection and storage, data review and sample labeling, incremental learning and model optimization.

[0089] During feedback data collection and storage, the system automatically records the entire process data for each match, including standardized query data packets, original outputs of the dual-matching model, integrated model fusion decision records, and a list of target spare parts. Simultaneously, it receives interactive feedback from maintenance technicians, including result acceptance markers, error correction information for faulty parts, and suggestions for adding new adaptation rules. All data is structured and stored in a local feedback database.

[0090] In the data review and sample labeling process, professional technicians review the data in the feedback database and label it as "correctly matched sample", "incorrectly matched sample" and "newly added knowledge sample" to ensure the quality of the training data.

[0091] In incremental learning and model optimization, high-quality samples are periodically extracted from the reviewed feedback database to construct an incremental training dataset. Incremental learning techniques are then used to fine-tune the parameters of the first matching model, the second matching model, and the data integration model. During the fine-tuning process, the original core knowledge of the model is preserved, and only erroneous logic is corrected and new knowledge is added, thus achieving continuous iteration of the model's matching capabilities.

[0092] The multi-level fallback and interactive output module is used to ensure service continuity, visualize results, and protect data security.

[0093] This involves a multi-level fallback mechanism. When the target spare parts list is empty, or the fusion confidence of all spare parts is lower than the fallback threshold, the multi-level fallback process is automatically triggered. The first level of fallback is: sending a standardized query data packet to a third-party spare parts database through the Model Context Protocol (MCP), parsing the returned results and verifying their suitability, and supplementing the target spare parts list with valid matching results. The second level of fallback is: if there are no valid results in the third-party database, a manual processing work order containing full diagnostic data is automatically generated and synchronized to the back-end manual service system, where professional technicians complete the spare parts matching and return the results.

[0094] Regarding data security and privacy protection, all sensitive data involving vehicle VIN, fault codes, and maintenance history are processed locally on the diagnostic equipment without being uploaded to the cloud, fundamentally avoiding the risk of data leakage.

[0095] For result output and display, the final list of target spare parts is presented to the user through a local interactive interface. The interface includes information such as part number, part name, confidence score, compatibility instructions, and joint replacement suggestions to assist repair technicians in making quick decisions. At the same time, information on spare parts added during the fallback process is synchronized to the feedback database for subsequent model training.

[0096] In specific embodiments, the vehicle parts matching method of the present invention can be widely applied to multiple scenarios in the automotive repair market, as illustrated below with typical application scenario diagrams.

[0097] For example, in a fault diagnosis scenario at an auto repair shop, a technician receives a 2020 model vehicle that exhibits symptoms such as shaking when starting from cold and weak acceleration. The technician uses a portable diagnostic device connected to the vehicle's OBD interface, reads fault code P0171 indicating a lean fuel system, and enters a description of the fault symptoms on the device's interface.

[0098] Accordingly, the diagnostic equipment's data acquisition and preprocessing module collects data such as VIN code, fault code P0171, and fault description, which is then standardized and packaged into a query data package. Parallel inference is achieved through a dual-matching model module. The first matching model, based on original factory data, outputs the air flow meter (part number 37980-5AA-A01) and the front oxygen sensor (part number 36532-5AA-A01), among other parts. The second matching model, based on fault case data, outputs the air flow meter, fuel injectors, and other parts. The integrated model module identifies the intersecting parts as the air flow meter, calculates the fusion confidence score, and prioritizes them. It also identifies the air flow meter and the front oxygen sensor as a related part combination and marks them as a joint replacement recommendation. The diagnostic equipment interface displays a list of target parts, which the technician uses to replace the air flow meter and the front oxygen sensor, thus resolving the vehicle fault. Finally, the technician marks the device as "matched correctly," and the feedback data is stored in a local database for subsequent incremental model training.

[0099] In this embodiment, the invention helps repair technicians quickly locate faulty parts, reduce the probability of misdiagnosis, and improve repair efficiency and first-time repair rate.

[0100] For example, in an outdoor rescue scenario using portable diagnostic equipment, rescuers carry the equipment outdoors to provide assistance for a rare vehicle model, such as an imported niche brand. The vehicle cannot start, the diagnostic equipment reads a fault code, but there is no matching record in the internal model.

[0101] Correspondingly, vehicle diagnostic data is collected through the data acquisition and preprocessing module to generate standardized query data packages. After inference by the dual-matching model module, the confidence levels of the output spare parts are all below the fallback threshold, triggering a multi-level fallback mechanism. First, the first-level fallback strategy is executed, querying a third-party imported spare parts database via the MCP protocol to obtain potential matching parts. Then, the second-level fallback strategy is executed; since the third-party database has no valid results, a manual work order is generated and synchronized to the backend expert system. Experts remotely analyze the data and return matching part suggestions. Finally, rescue personnel replace the parts according to the suggestions, and the vehicle restarts. Newly added rare model spare parts matching data is synchronized to the feedback database, enriching the system's knowledge base.

[0102] In this embodiment, the problem of matching spare parts for rare and niche car models is solved, ensuring the continuity of outdoor rescue services and expanding the applicable boundaries of diagnostic equipment.

[0103] For example, in an intelligent recommendation scenario for an auto parts e-commerce platform, a user enters their vehicle's VIN code and fault code P0171 into the platform, hoping to purchase compatible parts. The platform's backend data collection and preprocessing module receives the user's VIN code and fault code, parses and standardizes them; a dual-matching model module outputs a parts list in parallel; and an integration model module generates a target parts list, annotating the parts' compatible vehicle models, replacement suggestions, user reviews, and other information; the target parts list is displayed on the platform's front end, allowing users to directly place orders; user feedback after purchase, such as "good fit" or "trouble resolved after installation," is synchronized to a feedback database; and incremental learning optimizes the model to improve the accuracy of subsequent recommendations.

[0104] For example, see Figure 4 , Figure 4 This is a schematic diagram of an interactive interface provided in an embodiment of this application. In the interactive interface, the left side first prompts the user to input a vehicle malfunction problem, and the right side displays the vehicle identification number and fault code entered by the user. The target spare part can be obtained based on the user's input.

[0105] Further, see Figure 5 , Figure 5 This is a schematic diagram of a parts display interface provided in an embodiment of this application. If there are three target parts, namely parts 1, parts 2, and parts 3, they can be displayed through... Figure 5 The information layout shown displays the target spare parts. The right side of the display interface can also display spare parts parameter information, such as quantity, specifications, amount, etc.

[0106] In this embodiment, the e-commerce platform is provided with accurate intelligent recommendation capabilities for spare parts, reducing the probability of users purchasing the wrong parts and improving the user shopping experience.

[0107] For example, in the scenario of matching parts for long-tail vehicles, the target vehicle is a 2012 imported niche brand model with fewer than a thousand units in China. The vehicle exhibits a fault phenomenon of "power interruption during high-speed driving and P0011 fault code displayed on the dashboard." The original parts manual for this model has stopped being updated, and mainstream parts databases only store a small amount of basic parts information, which is a typical scenario for matching parts for long-tail vehicles.

[0108] First, technicians connected a portable diagnostic device to the vehicle's OBD interface to read the vehicle's VIN code, fault code P0011, and fault status. They then input a fault description via the human-machine interface: "2012 imported XX model, high-speed power interruption, reporting fault code P0011, no other repair history." Next, the VIN code was parsed to extract core information such as the vehicle's country of origin, manufacturer, model, production year, and engine model. Since this model is a niche imported version, some configuration information was missing from the VIN code parsing results. Natural language understanding technology was then used to perform entity recognition and semantic standardization on the fault description text, extracting the core entities "2012 imported XX model," "high-speed power interruption," and "P0011 fault code," and converting the colloquial description into structured data. All data was then integrated to generate a standardized JSON query data package, supplemented with the "long-tail model" tag.

[0109] Furthermore, the first matching model is trained based on standardized data such as vehicle parts atlas and original factory repair manuals. After inputting the query data package, it retrieves the general part corresponding to fault code P0011 as "variable valve timing solenoid valve". However, since this model is a long-tail model, the model lacks the adaptation parameters between this model and the general part. The output first parts list only contains one part, "variable valve timing solenoid valve". After calculating the first confidence level, if the first confidence level is 65%, it is lower than the preset threshold of 70%.

[0110] The second matching model is trained based on a massive number of repair failure cases and spare parts transaction records. After inputting a query data package, it retrieves three repair cases of other niche models on the same platform as the vehicle model through a fuzzy matching algorithm. In the cases, the parts corresponding to fault code P0011 are "variable valve timing solenoid valve" and "camshaft position sensor". The second spare parts list is output, containing two parts. After calculating the second confidence level, the second confidence level of "variable valve timing solenoid valve" is 72%, and the second confidence level of "camshaft position sensor" is 68%.

[0111] The integrated model receives the parts list and query data package output by the dual-matching model, identifies the intersection parts as "variable valve timing solenoid valve", and adjusts the evaluation dimension weights for long-tail vehicle scenarios: the matching degree weight of the same platform vehicle cases is increased to 40%, the original factory data adaptation weight is reduced to 30%, and the parameter integrity weight is reduced to 30%. Therefore, the average confidence of the intersection parts "variable valve timing solenoid valve" of the dual models is 68.5%, and the corrected basic confidence is 75% after combining the matching degree of the same platform cases.

[0112] For the intersecting component "variable valve timing solenoid valve", a third weight of 40% and a fourth weight of 60% are set, and the second fusion confidence level is calculated using the formula: 65%×40%+72%×60%=69.2%. Since the second matching model retrieves cases from the same platform and the model performance meets the standard, a long-tail model correction coefficient α=1.1 is introduced to correct the fusion confidence level: 69.2%×1.1=76.12%. For the unique component "camshaft position sensor" in the second component list, the confidence level is directly optimized using the correction coefficient: 68%×1.1=74.8%. The integrated model filters out parts with a fusion confidence level higher than the threshold of 70%, generating a target part list: "Variable valve timing solenoid valve (confidence level 76.12%)" and "Camshaft position sensor (confidence level 74.8%)", and marks "It is recommended to replace the variable valve timing solenoid valve first, as it has a high matching degree with cases on the same platform".

[0113] For example, see Figure 6 , Figure 6 This is a schematic diagram of a target spare parts display interface provided in an embodiment of this application. The displayed parameters may include the confidence level and remarks of the spare parts. For example, for spare parts 1, the confidence level is 80%. When spare parts 1 and 2 are jointly replaceable, a note is added, such as: "It is recommended to replace them together with spare parts 2." For spare parts 2, the confidence level is 75%, and the note is: "It is recommended to replace them together with spare parts 1." For spare parts 3, the confidence level is 72%, and no specific remarks are provided.

[0114] Since the confidence levels of the target parts list output by the integrated model are all above the threshold, there is no need to trigger a fallback mechanism. If the confidence level of the target parts list does not meet the threshold, the system will automatically query the niche vehicle parts database via the MCP protocol and initiate a manual expert review process.

[0115] Technicians replaced the "variable valve timing solenoid valve" according to the target parts list, resolving the vehicle malfunction. The vehicle was then marked "correct match" on the diagnostic equipment. All data from this matching process, including VIN code, fault code, matching result, and feedback information, was structured and stored in a local feedback database, labeled as a "high-quality matching sample for long-tail models." Subsequently, incremental learning techniques were used to inject this sample into the dual-matching model and the integrated model, optimizing the model's matching capability for similar long-tail models.

[0116] This embodiment addresses the part matching scenario for long-tail vehicle models. It compensates for the knowledge gaps of a single model through parallel inference using a dual-model approach, and optimizes confidence calculation using differentiated weighting and correction coefficients from the integrated model, ultimately achieving accurate matching. Compared to traditional solutions that directly return "no matching result," this method effectively solves the part matching problem for long-tail vehicle models, improving the matching success rate in complex scenarios. Simultaneously, through feedback and self-learning mechanisms, it continuously enriches the model's knowledge base on long-tail vehicle models, forming a positive cycle of matching, feedback, and optimization.

[0117] The above embodiments effectively avoid the knowledge blind spots of a single model through a collaborative mechanism of parallel matching using dual models and intelligent adjudication using an integrated model, significantly improving the accuracy and reliability of parts matching. Through closed-loop feedback and incremental learning mechanisms, the system achieves continuous self-evolution, adapting to the rapid iteration of new vehicle models and parts. A multi-level fallback mechanism ensures service continuity in complex scenarios, solving the pain point of traditional vehicle parts matching where services stop immediately upon encountering difficulties. Furthermore, the architecture design, which processes all data locally, balances matching efficiency and data security, demonstrating practical value and promising prospects for widespread adoption.

[0118] In summary, in this embodiment, vehicle diagnostic data of the target vehicle is first obtained. This vehicle diagnostic data includes at least one of the following: vehicle identification number (VIN), fault code, and problem description text. Then, the vehicle diagnostic data is input into a first matching model and a second matching model, respectively, to obtain a first spare parts list output by the first matching model and a second spare parts list output by the second matching model. The model parameters of the first matching model and the second matching model are different. Furthermore, based on the first spare parts information of each spare parts in the first spare parts list, a first confidence level is determined for each spare parts in the first spare parts list. Based on the second spare parts information of each spare parts in the second spare parts list, a second confidence level is determined for each spare parts in the second spare parts list. Finally, based on the first confidence level and the second confidence level of each spare parts in the first and second spare parts lists, the target spare parts required by the target vehicle are determined. Therefore, by using two spare parts matching models with different model parameters, two different spare parts lists are output for the vehicle diagnostic data, and the target spare parts required by the target vehicle are determined based on the confidence level of each spare parts. This differs from existing technologies and can improve the matching accuracy of vehicle spare parts when a vehicle malfunctions.

[0119] The methods of the embodiments of the present invention have been described in detail above, and the apparatus of the embodiments of the present invention is provided below.

[0120] See Figure 7 , Figure 7 This is a schematic diagram of the structure of a vehicle parts matching device provided in an embodiment of this application. Figure 7As shown, the vehicle parts matching device 800 includes an acquisition unit 801 and a processing unit 802. The acquisition unit 801 is used to acquire vehicle diagnostic data of the target vehicle. The vehicle diagnostic data includes at least one of the following: vehicle identification number, fault code, and problem description text. The processing unit 802 is used to input the vehicle diagnostic data into a first matching model and a second matching model respectively to obtain a first parts list output by the first matching model and a second parts list output by the second matching model. The model parameters of the first matching model and the second matching model are different. Based on the first parts information of each parts in the first parts list, a first confidence level of each parts in the first parts list is determined. Based on the second parts information of each parts in the second parts list, a second confidence level of each parts in the second parts list is determined. Based on the first confidence level of each parts in the first parts list and the second confidence level of each parts in the second parts list, the target parts required by the target vehicle are determined.

[0121] In specific implementations, the acquisition unit 801 and the processing unit 802 in this application embodiment can also execute other implementation methods described in the vehicle parts matching method of this application embodiment, which will not be repeated here.

[0122] See Figure 8 , Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. This electronic device can be the diagnostic device described above. For example... Figure 8 As shown, the electronic device 900 includes a transceiver 901, a processor 902, and a memory 903, which are connected via a bus 904. The memory 903 stores computer programs and data, and can transmit the data stored in the memory 903 to the processor 902. The electronic device 900 can be the aforementioned vehicle parts matching device 800, and the processor 902 can be the aforementioned acquisition unit 801 and processing unit 802. In this embodiment, the processor 902 is used to read the computer program in the memory 903 and execute some or all of the steps of the aforementioned vehicle parts matching method.

[0123] This application also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement some or all of the steps of any of the vehicle parts matching methods described in the above method embodiments.

[0124] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the vehicle parts matching methods described in the above method embodiments.

[0125] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0126] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0127] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or modules may be electrical or other forms.

[0128] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0129] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software program modules.

[0130] If the integrated module is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0131] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A vehicle parts matching method characterized by, include: Obtain vehicle diagnostic data for the target vehicle; the vehicle diagnostic data includes at least one of the following: vehicle identification number, fault code, and problem description text; The vehicle diagnostic data is input into the first matching model and the second matching model respectively to obtain the first spare parts list output by the first matching model and the second spare parts list output by the second matching model; the model parameters of the first matching model and the model parameters of the second matching model are different. Based on the first component information of each component in the first component list, determine the first confidence level of each component in the first component list; Based on the second component information of each component in the second component list, determine the second confidence level of each component in the second component list; The target parts required for the target vehicle are determined based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list.

2. The method of claim 1, wherein, The first spare parts information includes: compatible vehicle type, number of initial parameter items, number of actual parameter items, and information source type; determining the first confidence level of each spare part in the first spare parts list based on the first spare parts information of each spare part in the first spare parts list includes: Obtain the vehicle type of the target vehicle; The vehicle type of the target vehicle is compared with the compatible vehicle type of each part in the first parts list to obtain the matching accuracy score of each part in the first parts list. Based on the number of initial parameter items and the actual number of parameter items for each component in the first spare parts list, determine the parameter completeness score for each component in the first spare parts list. Based on the information source type of each component in the first spare parts list, determine the source accuracy score of each component in the first spare parts list; Based on the matching accuracy score, parameter completeness score, and source accuracy score of each component in the first spare parts list, the first confidence level of each component in the first spare parts list is determined.

3. The method as described in claim 1, characterized in that, The step of determining the target spare parts required by the target vehicle based on the first confidence level of each spare part in the first spare parts list and the second confidence level of each spare parts in the second spare parts list includes: Obtain the intersection of the first spare parts list and the second spare parts list; Determine the first weighting coefficient for the intersecting parts; Based on the first weighting coefficient, the first confidence level and the second confidence level of the intersecting parts are weighted and calculated to obtain the first fusion confidence level of the intersecting parts; The target component is determined based on the first fusion confidence level and the first confidence threshold of the intersection components.

4. The method as described in claim 1, characterized in that, The step of determining the target spare parts required by the target vehicle based on the first confidence level of each spare part in the first spare parts list and the second confidence level of each spare parts in the second spare parts list includes: Obtain the union of the first spare parts list and the second spare parts list; Determine the second weighting coefficient of the union of the components; Based on the second weighting coefficient, the first confidence level and the second confidence level of the union parts are weighted and calculated to obtain the second fusion confidence level of the union parts; The target component is determined based on the second fusion confidence level and the second confidence threshold of the union components.

5. The method as described in claim 1, characterized in that, The step of determining the target spare parts required by the target vehicle based on the first confidence level of each spare part in the first spare parts list and the second confidence level of each spare parts in the second spare parts list includes: Obtain the associated parts combinations from the first parts list and the second parts list; Determine the third weighting coefficient for each component in the associated component combination; Based on the third weighting coefficient of each component in the associated component combination, the first confidence level and the second confidence level of each component in the associated component combination are weighted and calculated to obtain the third fusion confidence level of each component in the associated component combination; The target component is determined based on the third fusion confidence level and the third confidence threshold of each component in the associated component combination.

6. The method according to any one of claims 3-5, characterized in that, The step of determining the target spare parts required by the target vehicle based on the first confidence level of each spare part in the first spare parts list and the second confidence level of each spare parts in the second spare parts list includes: If the first confidence level of each part in the first part list is less than the fourth confidence level threshold, and the second confidence level of each part in the second part list is less than the fifth confidence level threshold, then obtain the candidate part database for the target vehicle. Based on the candidate parts database and the vehicle diagnostic data, the target parts required for the target vehicle are matched.

7. The method as described in claim 1, characterized in that, The method further includes: Obtain the vehicle parts atlas, repair manual, and vehicle model-parts association rule library for the target vehicle; The first candidate matching model is trained based on the vehicle parts map, the repair manual, and the vehicle model-parts association rule base to obtain the first matching model; Obtain vehicle repair failure cases and actual parts transaction records for the target vehicle; Based on the vehicle repair failure cases and the actual transaction records of the spare parts, determine the parts matching labels; The second candidate matching model is trained based on the vehicle repair failure cases, the actual transaction records of the spare parts, and the matching tags of the parts to obtain the second matching model.

8. A vehicle parts matching device, characterized in that, The device includes an acquisition unit and a processing unit; The acquisition unit is used to acquire vehicle diagnostic data of the target vehicle; the vehicle diagnostic data includes at least one of the following: vehicle identification number, fault code, and problem description text; The processing unit is used to input the vehicle diagnostic data into a first matching model and a second matching model respectively, to obtain a first spare parts list output by the first matching model and a second spare parts list output by the second matching model; the model parameters of the first matching model and the model parameters of the second matching model are different. Based on the first component information of each component in the first component list, determine the first confidence level of each component in the first component list; Based on the second component information of each component in the second component list, determine the second confidence level of each component in the second component list; The target parts required for the target vehicle are determined based on the first confidence level of each part in the first parts list and the second confidence level of each part in the second parts list.

9. An electronic device, characterized in that, include: A processor and a memory, the processor being connected to the memory, the memory being used to store a computer program, the processor being used to execute the computer program stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-7.

Images