Automobile aluminum ornament product quality inspection method based on big data

Abstract

The invention discloses an automobile aluminum ornament product quality inspection method based on big data, particularly relates to the field of quality management and quality inspection decision, and is used for solving the problem that existing aluminum ornament quality inspection depends on fixed rules and experience judgment and is difficult to dynamically optimize quality inspection performance and resource allocation in combination with after-sales quality information. According to the method, structured processing and damage mode division are performed on vehicle after-sales damage data, key process features are extracted in combination with production process time sequence data, and a defect early propagation model is constructed, so that prediction of occurrence probabilities of different defect types is realized; on the basis, a multi-layer hierarchical quality inspection strategy is dynamically generated according to the defect risk level, quality inspection execution cost and historical quality loss cost are integrated, and quality inspection performance indexes of a next production cycle are output through multi-objective optimization, so that the scientificity and decision value of quality risk management are improved while the cost is controlled, and the quality risk management efficiency is improved. Therefore, the management requirements of refined operation and cost constraint are met.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of quality management and quality inspection decision-making, more specifically, the present application relates to a quality inspection method for automobile aluminum trim parts based on big data. BACKGROUND

[0002] The existing quality management of automobile aluminum trim parts mainly focuses on production execution and result inspection, and the core goal is to control the outflow of unqualified products and reduce the quality risk after sale, but in actual application, there are still problems such as relying on experience for management decision, rigid resource allocation and insufficient use of cross-link information.

[0003] With the popularization of large-scale automobile manufacturing and multi-batch parallel production mode, aluminum trim parts show different quality performance under different production cycles and different process conditions. It is difficult to balance the quality risk and quality inspection investment under the condition of limited cost by simply relying on fixed quality inspection rules or static sampling inspection scheme. At the same time, enterprises accumulate a large amount of damage and loss data in the process of after-sales maintenance, quality claim and internal quality evaluation, but these data are mostly used for post-event statistics and responsibility analysis, and have not formed a systematic data processing mechanism for production and quality inspection decision-making, resulting in lack of foresight and overall coordination in quality management activities.

[0004] In the quality management practice, how to implement differentiated detection strategies for products of different risk levels based on quality loss and production state, reasonably allocate detection resources, and reduce potential quality risks while controlling execution cost has become a key problem in enterprise operation management. Therefore, it is necessary to build a data processing method for quality management and quality inspection decision-making, to realize the dynamic generation and performance optimization of quality inspection strategy through comprehensive analysis of quality events, production process and cost information, to provide enterprises with more economical and decision-making value quality management means, to adapt to the quality inspection management and control needs of fine operation and cost constraints. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a quality inspection method for automobile aluminum trim parts based on big data to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: A quality inspection method for automobile aluminum trim parts based on big data, comprising the following steps: S1, extracting damage records containing aluminum trim part batch identification from vehicle after-sales maintenance data source, performing data cleaning on the damage records, and generating structured batch damage data; S2, performing unsupervised learning based on spectral clustering on the batch damage data, and dividing the damage mode groups of aluminum trim parts; S3, trace the time sequence production process data of the corresponding production cycle according to the aluminum ornament part batch identification, and cut into production segment feature sequences by using a sliding window method; S4, for each damage mode group, input the production segment feature sequences of all batches contained into an integrated classification framework, and identify key process feature combinations; S5, based on the key process feature combinations of production, construct an aluminum ornament process defect early propagation network, and predict measurable defect types and occurrence probabilities; S6, according to the predicted defect types and occurrence probabilities, dynamically generate a multi-layer hierarchical quality inspection strategy; S7, integrate the estimated execution cost of the multi-layer hierarchical quality inspection strategy and the historical quality loss cost data corresponding to each damage mode group, and output a set of quality inspection performance indicators in the next production cycle through a dynamic double-objective optimization algorithm.

[0007] As a further scheme of the present application, in S1, the generation of structured batch damage data specifically includes: Access vehicle after-sales maintenance data sources and extract damage records related to aluminum ornament parts, which contain aluminum ornament part batch identification and damage morphology description; Perform data cleaning on the extracted damage records, unify the format of the part batch identification, and map the damage morphology description to a standardized damage morphology code; Integrate the cleaned damage records according to the aluminum ornament part batch identification, and generate a structured damage data set indexed by the part batch.

[0008] As a further scheme of the present application, in S2, the division of aluminum ornament part damage mode groups specifically includes: According to the frequency of each type of damage morphology code in the structured damage data set of each aluminum ornament part batch, calculate the cosine similarity between any two aluminum ornament part batches; Construct a symmetric part batch similarity matrix and calculate the corresponding Laplacian matrix, perform eigenvalue decomposition on the Laplacian matrix, select the eigenvectors corresponding to the first several largest eigenvalues, and form a new feature space; In the new feature space, all aluminum ornament part batches are divided into several mutually exclusive sets by using a clustering algorithm, each set is defined as a damage mode group, and the group identification is labeled.

[0009] As a further scheme of the present application, in S3, according to the aluminum ornament part batch identification, trace the time sequence production process data of the corresponding production cycle, and cut into production segment feature sequences by using a sliding window method, specifically including: According to the aluminum ornament part batch identification, retrieve the multi-dimensional time series data recorded on the complete production link of each batch from the historical production line; aligning and splicing the retrieved multi-dimensional time series data according to manufacturing process nodes to form a continuous global production process data stream of the batch; defining the width and step interval of the sliding window based on the theoretical production cycle length of a single product, and continuously intercepting the global production process data stream along the time axis; calculating the statistical characteristics of each type of production parameter within the time window for each data segment intercepted by the sliding window, arranging the statistical characteristics of all windows of the batch in chronological order to construct a production segment feature sequence corresponding to the batch.

[0010] As a further scheme of the present application, in S4, for each damage mode group, the production segment feature sequences of all batches contained therein are input into an integrated classification framework to identify the key process feature combination, which specifically includes: For each damage mode group, the production segment feature sequences corresponding to all aluminum trim part batches in the group are identified as a positive sample set, and the production segment feature sequences corresponding to normal part batches not classified into any damage mode group are extracted from historical production data as a negative sample set; An integrated classification model is trained using the positive sample set and the negative sample set, which takes the production segment feature sequence as input and predicts the batch classification as output; The cumulative contribution of each process feature to the sample classification result during the classification decision node division in the model training process is counted, the importance ranking of the process features relied on by the integrated classification model in the classification decision is calculated, and the highest importance feature subset is selected as the key process feature combination.

[0011] As a further scheme of the present application, in S5, the aluminum trim part process defect early propagation network is constructed to predict the measurable defect types and occurrence probabilities, which specifically includes: The key process feature combination is mapped to the corresponding manufacturing node, a directed graph network model is constructed based on the relationship between the manufacturing nodes, the nodes represent the manufacturing processes carrying the key process features, and the edges represent the defect propagation relationship along the process path; The conditional probability of defect state transition between adjacent nodes in the directed graph network model is calculated using the co-occurrence relationship between the same key process features and the final detected defects extracted from historical production data; In the current aluminum trim part batch, the production segment feature sequence containing the key process features is input, and the forward propagation reasoning is performed along the process path direction in the directed graph network model to calculate the occurrence probability of each type of defect detected at each downstream process in real time, and the defect types and their corresponding probability values whose probability exceeds the set confidence are output as the prediction.

[0012] As a further scheme of the present application, in the S6, the dynamically generated multi-layered hierarchical quality inspection strategy according to the predicted defect types and occurrence probabilities specifically comprises: dividing the occurrence probability of each predicted defect type into different risk level intervals; matching a preset detection intensity level to each risk level interval, the detection intensity level defining the detection type and sampling reference proportion corresponding to the risk level defect, and determining the quality inspection link for executing the corresponding detection intensity level; integrating the risk level, detection intensity level and quality inspection link information corresponding to all predicted defect types of the current aluminum trim part batch, to generate a dynamic quality inspection list with a multi-layered hierarchical quality inspection strategy.

[0013] As a further scheme of the present application, in the S7, the output quality inspection performance indicator set in the next production cycle specifically comprises: based on the detection intensity level and execution node defined in the dynamic quality inspection list, and in combination with the predicted output in the production plan, calculating the estimated total execution cost required for implementing the corresponding quality inspection strategy; for each damage mode group, statistically analyzing historical quality loss cost data, calculating the average loss cost of aluminum trim part defects in the group, and combining the predicted occurrence probability of each defect type to calculate the total expected quality risk cost; taking the minimization of the estimated total execution cost and the total expected quality risk cost as two optimization objectives, taking the detection node, sampling frequency and tolerance standard as optimization decision variables, constructing a multi-objective optimization problem model and solving it, and converting the solution result into a quality inspection performance indicator set for the next production cycle.

[0014] The present application has the following technical effects and advantages: The present application realizes quality management and quality inspection decision optimization for the whole process of automotive aluminum trim parts by uniformly modeling and correlatively analyzing vehicle post-sale damage data, production process time series data and quality cost information. Through automatic identification of damage modes and extraction of key process features, the present application can perceive potential quality risks of different production batches from the management decision level in advance, avoiding the lag caused by relying on post-detection or experience rules for quality inspection arrangement. Meanwhile, the present application introduces defect propagation modeling and probability prediction mechanism, so that the quality risk assessment is changed from static judgment to quantifiable and comparable risk measurement, which helps the management layer to make a rational trade-off between cost and quality. Further, by incorporating the quality inspection execution cost and historical quality loss cost into a unified multi-objective optimization framework, the present application can output quality inspection performance indicators for the next production cycle, realize fine allocation of detection resources, and reduce waste and risks caused by excessive detection or insufficient detection.

[0015] Compared with the prior art, the application is more in line with the actual needs of enterprise quality management and operation decision, improves the dynamic adaptability and economy of the quality inspection strategy, and provides a quality decision support scheme with management value and implementation feasibility for an automobile aluminum trim part manufacturing enterprise. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A schematic diagram of a quality inspection method for an automobile aluminum trim part based on big data is given. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the application.

[0018] Embodiment 1 Figure 1 A quality inspection method for an automobile aluminum trim part based on big data is given, which comprises the following steps: S1, extracting damage records containing aluminum trim part batch identification from a vehicle after-sales maintenance data source, performing data cleaning on the damage records, and generating structured batch damage data; S2, performing unsupervised learning based on spectral clustering on the batch damage data, and dividing damage mode groups of the aluminum trim part; S3, according to the aluminum trim part batch identification, tracing the time sequence production process data of the corresponding production cycle, and cutting the production process data into production segment feature sequences by using a sliding window method; S4, for each damage mode group, inputting the production segment feature sequences of all batches contained in the damage mode group into an integrated classification framework, and identifying key process feature combinations; S5, based on the key process feature combinations, constructing an early propagation network of aluminum trim part process defects, and predicting measurable defect types and occurrence probabilities; S6, dynamically generating a multi-layered recursive quality inspection strategy according to the predicted defect types and occurrence probabilities; S7, integrating the estimated execution cost of the multi-layered recursive quality inspection strategy and the historical quality loss cost data corresponding to each damage mode group, and outputting a set of quality inspection performance indicators in the next production cycle by using a dynamic double-objective optimization algorithm.

[0019] In S1, the structured batch damage data is generated.

[0020] The historical data formed in the vehicle after-sales maintenance link is uniformly accessed and processed. The accessed data source is the maintenance business data actually formed and long-term retained in the vehicle after-sales maintenance process. The data source contains maintenance records formed when the vehicle enters the maintenance process due to appearance, assembly or functional abnormalities during the use cycle. Each maintenance record corresponds to a specific vehicle, maintenance time and related parts information. For the specific part type of aluminum trim, the record content in the after-sales maintenance data source is scanned and screened piece by piece, and only the maintenance records related to actual damage of aluminum trim parts are extracted. To ensure the traceability and consistency of subsequent data processing, it is clearly required in the extraction process that each damage record at least contains batch identification information of the aluminum trim part and textual description information of the damage state, wherein the batch identification is used to uniquely refer to the manufacturing batch to which the aluminum trim belongs in the production stage, and the damage form description is used to reflect the actual damage form presented by the aluminum trim in the after-sales stage. After extraction, all maintenance records meeting the conditions are sorted and stored in the original business time sequence, keeping the original semantic integrity and time correlation of the damage records.

[0021] After the extraction of the damage record, systematic data cleaning and normalization processing is performed on the obtained damage record. In this embodiment, the format difference problem of the aluminum trim part batch identification is uniformly processed. Since the after-sales maintenance data comes from different maintenance sites and different input personnel, the batch identification may have inconsistent cases, separator differences or additional description characters in the actual record, so the batch identification is format normalized in the cleaning process, so that the aluminum trim of the same production batch corresponds to a unique and stable identification form at the data level. Subsequently, the damage form description content is normalized and mapped. Specifically, the damage phenomenon described in natural language form in the maintenance record is compared with the pre-arranged and fixed damage form classification table, and the descriptions with consistent or highly similar semantics are uniformly mapped to the corresponding standardized damage form code, such as the different expressions of surface scratch, paint peeling and deformation are uniformly classified into a specific damage form category. After the above cleaning and mapping processing is completed, all cleaned damage records are integrated and merged according to the aluminum trim part batch identification, and multiple damage records appearing under the same batch are associated and integrated to form a data structure indexed by the part batch. In the structured damage data set, each batch index is associated with its corresponding damage form code set and occurrence, thereby forming the basic data form required for subsequent damage pattern analysis and production process correlation analysis.

[0022] In S2, the damage mode groups of the aluminum trim parts are divided.

[0023] Based on the formed structured damage data set, the damage pattern codes corresponding to each aluminum trim part batch are statistically arranged. The occurrence of each part batch in the after-sales stage is counted by type, and the frequency description result reflecting the damage distribution characteristics of the batch is formed. The frequency description result takes the damage pattern category as the dimension and the occurrence frequency as the value, and completely retains the relative distribution relationship of different damage patterns in the same batch. On this basis, the damage frequency description results of any two different aluminum trim part batches are compared and calculated, and the closeness of the two batches in the damage pattern distribution structure is quantified by calculating the cosine similarity. The similarity calculation process takes the multi-dimensional vector composed of damage pattern frequencies as input, and the similarity result directly comes from the actual damage data itself. After completing the similarity calculation between batches, the similarity results obtained between all aluminum trim part batches are arranged in order according to the batch sequence to construct a symmetric part batch similarity matrix. The rows and columns of the similarity matrix correspond to the specific aluminum trim part batches, and the matrix elements in the same position represent the similarity between the corresponding two batches in the damage pattern distribution. According to the similarity matrix, a corresponding Laplacian matrix is constructed to reflect the overall association structure between batches. The eigenvalue decomposition operation is performed on the Laplacian matrix, and the feature vectors corresponding to several eigenvalues that meet the preset number requirement are selected from the decomposition result, and these feature vectors are combined according to the batch dimension to form a new low-dimensional feature representation space. The new feature space can not only retain the original similarity relationship structure, but also reduce the data dimension and weaken the noise interference, so that the position relationship of different aluminum trim part batches in the space is more clear and separable.

[0024] After obtaining the new feature space, the representation of all aluminum trim part batches in the feature space is taken as input to perform clustering analysis operation. The batch features are grouped by using unsupervised clustering method, and the batches close to each other are automatically grouped into the same set by calculating the distance relationship between the batches in the feature space. The clustering process does not depend on the pre-defined damage category label, but is completely based on the distribution structure of the batches in the feature space to complete the division, so as to ensure that the clustering result can truly reflect the internal mode of damage pattern distribution. After clustering, a plurality of mutually exclusive batch sets are obtained, and the aluminum trim part batches in each batch set have significant consistency in damage pattern distribution. Then, each batch set is defined as a damage mode group, and each group is assigned a unique group identifier for distinguishing and referencing different damage modes in subsequent production process analysis, key feature identification and quality inspection strategy generation process.

[0025] In S3, according to the aluminum trim part batch identifier, the time sequence production process data of the corresponding production cycle is traced back, and the sliding window method is used to cut into production segment feature sequences.

[0026] According to the batch identification bound by the aluminum ornament part in the production stage, the production process data recorded in the historical production line is systematically searched. The batch identification is taken as the only index condition to retrieve the multi-dimensional time series data formed by the batch in the complete production link from the production execution record. The multi-dimensional time series data is derived from the process parameter data continuously collected and recorded in the actual production process of each manufacturing process, covering the continuous running state information under multiple process nodes. To ensure the consistency and splicing of different process data in the time dimension, after the retrieval is completed, the obtained multi-dimensional time series data is sequentially arranged according to the manufacturing process nodes, and the data segments corresponding to each process are aligned according to the established process sequence in the production process. In the alignment process, the actual time records of process entry and exit are taken as the reference, the data belonging to the same process node is completely retained, and the time dislocation problem caused by the difference in data collection frequency or record delay is eliminated. After completing the process alignment, the time series data corresponding to each process node is spliced according to the actual sequence of the production link to form a continuous data stream covering the batch from the initial process to the final process. The continuous data stream maintains monotonic increase in the time dimension and fully reflects the whole production process of the batch in the process dimension, thereby constituting the global production process data stream that can represent the overall production behavior of the batch.

[0027] After obtaining the global production process data stream corresponding to the batch, time-axis-based segmentation processing is performed on the data stream to form a production segment feature sequence. The theoretical production cycle length corresponding to the completion of all manufacturing processes of a single aluminum ornament product on the production line is taken as the reference benchmark to determine the time width and step interval of the sliding window. The window width and step interval are determined according to the production rhythm and process continuity, for example, the window width is set to cover the time length of several continuous process running periods, and the step interval is set to a certain proportion of the window width to ensure the continuity between adjacent windows rather than complete fragmentation. Along the time axis direction of the global production process data stream, the data stream is continuously intercepted, and each interception generates a production data segment within the corresponding time period. For each production data segment, statistical feature calculation operations are performed on various production process parameters contained therein, including descriptive indicators that can reflect the overall level and fluctuation state of the parameters, thereby converting the original high-frequency time series data into structured feature representation. After completing the feature calculation of a single window, the statistical feature results corresponding to all sliding windows in the same batch are arranged in time sequence to form a production segment feature sequence consistent with the time evolution sequence. The production segment feature sequence maintains time continuity while achieving segmented characterization of the dynamic changes in the production process.

[0028] In the S4, for each damage mode group, the production segment feature sequence of all batches contained is input into the integrated classification framework to identify the key process feature combinations.

[0029] For the results of the damage mode group division, the historical production data is further organized and screened to form a sample set for classification modeling. Taking the damage mode group as the basic unit of sample division, each damage mode group is processed separately. According to the aluminum trim part batch identification, all parts batches belonging to the damage mode group are retrieved one by one, and the production segment feature sequences formed by these batches in the production stage are called. The production segment feature sequence is generated by the sliding window method, which fully reflects the process feature changes of the corresponding batch at each time segment in the production cycle. On this basis, the production segment feature sequences corresponding to all batches in the group are uniformly included in the positive sample set, and each sample in the positive sample set clearly corresponds to a specific batch and its time-ordered production segment features. Subsequently, from the historical production data, aluminum trim part batches that are not classified into any damage mode group are screened, which have no clear damage record in the after-sales stage and are not associated with any damage pattern code in the structured damage data set. The production segment feature sequences corresponding to these batches are extracted and integrated as negative samples. Through the above method, the positive sample set and the negative sample set are consistent in data source, feature composition and time expression form, only differ in damage mode attribution, thereby providing a clear and distinguishable sample basis for the subsequent construction of classification model.

[0030] After the construction of the positive sample set and the negative sample set, the two types of samples are uniformly organized to form a data set for model training. The integrated classification model used is composed of multiple classification units based on decision structure. Each classification unit learns the division of sample features during the training process, and the results are fused at the model level. During model training, the production segment feature sequence is used as the input data form, and each input sample contains multiple production segment features arranged in chronological order, which are used to describe the process state evolution of the corresponding part batch during the entire production cycle. The output form of the model is the batch classification prediction result, which is used to determine whether the input sample belongs to the damage mode group or not. For example, for a certain damage mode group, 120 batches of aluminum decorative part batches in the group are selected as positive samples, and 180 batches from normal batches that are not classified into any damage mode group are selected as negative samples. The production segment feature sequence corresponding to each batch is composed of 10 continuous sliding windows, each window containing statistical features of production parameters such as temperature, pressure, and speed calculated within the window. During model training, an integrated classification model composed of random forest classifier and gradient boosting decision tree classifier is used, where the random forest part contains 100 decision trees, and the maximum depth of each decision tree is set to 8. The gradient boosting decision tree part is set to 80 iterations, and the single learning step is set to 0.1. During training, the positive and negative samples are divided into training data and validation data in a ratio of 8:2, and the model is trained iteratively with batch classification results as output targets. When the classification accuracy of the validation set does not improve for 5 consecutive rounds, the training is terminated, and the final integrated classification model for batch classification prediction is obtained.

[0031] During training, samples in the positive sample set and the negative sample set are input into the model simultaneously, and each classification unit continuously adjusts the internal decision structure according to the differences in sample features during the training stage, so that the model gradually forms the ability to distinguish the characteristics of different damage mode groups. During training, the consistent mapping relationship between input features and output classification is strictly maintained to ensure that the model can give stable and reproducible batch classification prediction results for any input production segment feature sequence after training.

[0032] After the integrated classification model training is completed, the decision structure formed in the model is analyzed to identify the process features that play a major role in the classification result. For all process features used in the model training process, the calling and participation of the process features in the sample division at each classification decision node are counted. In specific implementation, the contribution of each process feature to the sample division result when participating in the decision node division in the model training stage is recorded, and the contribution reflects the actual discrimination strength of the feature in distinguishing positive samples from negative samples. The specific way is to record the number of times each process feature is used to split nodes in random forest and gradient boosting decision tree, and simultaneously accumulate the class purity influence level of the corresponding split on the positive and negative sample division result. The cumulative contribution of the same process feature in all decision trees and iteration rounds is summarized to form the comprehensive contribution value of the feature, and the importance ranking result is obtained. The class purity refers to the concentration degree of the distribution of positive samples and negative samples in the sample set, and the influence degree is measured by comparing the reduction of the class distribution uncertainty in the sample set before and after the split. The greater the reduction of uncertainty after the split, the greater the contribution of the process feature to the sample division. Subsequently, according to the importance ranking, a feature subset with a high contribution degree is selected from all process features, and the selection process is based on the principle of covering the main discrimination information to avoid introducing redundant or noise features. Finally, the selected feature subset is defined as the key process feature combination of the corresponding damage mode group. The key process feature combination is directly related to specific production processes and production parameters in structure, and reflects the production process abnormal features highly related to the damage mode in semantics.

[0033] In S5, an early propagation network of process defects of the aluminum trim part is constructed, and the measurable defect types and occurrence probabilities are predicted.

[0034] The key process feature combination identified is subjected to process-level mapping processing. Each feature in the key process feature combination is derived from the statistical performance of a specific production parameter in a particular manufacturing process, so in the implementation process, according to the correspondence between the parameters and the processes in the production process record, the parameters involved in the calculation of each key process feature are associated with the manufacturing process node where the parameters are generated. After mapping, each manufacturing process involved in the complete production link of the aluminum trim part is defined as a node in the graph network, and each node carries the key process feature information associated with the process. After the nodes are constructed, the directed connection relationship between the nodes is established according to the sequence relationship of the processes in the actual production process of the aluminum trim part. The directed connection is used to represent the possible propagation direction of the defects in the process path. The directed graph network model constructed in this way reflects the processing sequence of the aluminum trim part from the front-end process to the back-end process in structure, and clearly depicts the path constraint relationship of the defect transmission along the process flow in semantics.

[0035] After the structure of the directed graph network model is completed, the historical production data and detection data are jointly analyzed to quantify the transition relationship of defect states between adjacent manufacturing nodes. In specific implementation, for any pair of adjacent manufacturing nodes with a direct connection relationship in the directed graph network, the occurrence of specific defects detected in the downstream node is counted under the condition that the upstream node corresponding key process feature shows abnormal performance in the historical production batch. The statistical process is carried out in units of historical batches, and for each batch, whether it shows the corresponding key process feature in the upstream process stage and whether it records a specific defect type in the subsequent detection link are checked at the same time. By accumulating the number of historical batches that meet the above conditions and comparing it with the total number of batches in which the upstream key process feature appears, the conditional probability value of the occurrence of a specific defect state in the downstream node under the condition of a given upstream key process feature state is calculated. The conditional probability value is recorded as the attribute of the corresponding edge in the directed graph network model, which is used to describe the strength and possibility of defect propagation between adjacent process nodes.

[0036] When the current aluminum decoration part batch enters the analysis stage, the production segment feature sequence corresponding to the batch is extracted, and the feature state information matching the key process feature combination is identified from the sequence. Then, the production segment feature sequence is loaded as input into the directed graph network model that has been constructed, and the forward propagation reasoning is performed from the upstream process node to the downstream process node in the actual order of the manufacturing process. In specific implementation, at each manufacturing node, the possibility of defect state occurrence is calculated and updated step by step according to the key process feature state carried by the node and the conditional probability information recorded on the connected edge, and the updated defect probability is passed to the subsequent node. Through continuous propagation and accumulation calculation along the direction of the process path, the occurrence probability results of various defects detected in each downstream manufacturing process and corresponding detection link are finally obtained. After completing the full-path reasoning, the defect occurrence probability is compared with the pre-set confidence standard, where the confidence standard is set according to the historical detection stability data, for example, the confidence is set to 0.7, which is used to screen defects with high occurrence credibility. Finally, the defect types with occurrence probability exceeding the confidence standard and their corresponding probability values are taken as the defect prediction output results of the batch, providing probability input for quality inspection strategy generation.

[0037] In S6, a multi-layer hierarchical quality inspection strategy is dynamically generated according to the predicted defect type and occurrence probability.

[0038] The occurrence probability corresponding to each defect type is uniformly processed, and is classified according to a pre-set risk level interval. The risk level interval is set based on historical quality data of the enterprise and quality inspection management experience, for example, the defect occurrence probability is classified into a low risk interval, a medium risk interval and a high risk interval from low to high, and the boundaries of each interval are fixed and stable before implementation. After completing the risk level classification, a corresponding detection intensity level is matched for each risk level interval, and the detection intensity level is jointly defined by the detection type and the sampling reference proportion, and the specific quality inspection link of implementing the detection intensity is also specified. The detection type is used to refer to the strictness of the detection method, and the sampling reference proportion is used to limit the proportion of products that need to be sampled in the corresponding quality inspection link. When the predicted occurrence probability of a certain surface scratch type defect is classified into the low risk interval, the corresponding matched detection intensity level is the conventional appearance sampling, and the sampling reference proportion is set to a low level, for example, sampling is performed at a fixed proportion in the end appearance inspection process. When the predicted occurrence probability of a certain structure deformation type defect is classified into the high risk interval, the corresponding matched detection intensity level is the strengthened detection level, the detection type includes the combined mode of size measurement and appearance review, the sampling reference proportion is set to a high level, and the detection is performed simultaneously in the finished product detection process and the previous key process. Through the above method, defects of different risk levels are clearly distinguished in the allocation of detection resources.

[0039] After matching the risk level and detection intensity level of each defect type, all predicted defect types involved in the current aluminum trim part batch are comprehensively sorted. The risk level information corresponding to each defect type in the batch is read one by one, and the detection intensity level associated with the risk level and the specific quality inspection link position are synchronously called. Subsequently, the above information is summarized according to the defect type dimension, and multiple defect risks that may be involved in the same batch are presented side by side to form a quality inspection decision result covering multiple risk levels. In the integration process, multiple detection requirements under the same quality inspection link are uniformly merged to ensure that the highest corresponding detection intensity level is implemented at the implementation level. Finally, the defect type, risk level, detection intensity level and quality inspection link information obtained by integration are structured and sorted in the form of a list to generate a dynamic quality inspection list with multi-level hierarchical characteristics. The dynamic quality inspection list clearly lists the detection type and corresponding sampling reference proportion to be performed in each quality inspection link in terms of content, and embodies the hierarchical relationship from high risk to low risk in terms of sequence, thereby forming a clear and executable hierarchical quality inspection arrangement within the same production batch, ensuring consistency and operability in the quality inspection execution process.

[0040] In the S7, a set of quality inspection performance indicators in the next production cycle is output.

[0041] According to the generated dynamic quality inspection list, the cost of all quality inspection activities to be performed in the current production cycle is calculated. In specific implementation, the detection intensity level information corresponding to each quality inspection link is read from the dynamic quality inspection list one by one, including the specific detection type and the matching sampling reference ratio. At the same time, combined with the expected yield data in the production plan, the product quantity covered by each quality inspection link in the production cycle is confirmed. Then, for each quality inspection link, according to the actual detection method corresponding to its detection type, the unit execution cost record of the detection method in the historical execution is called, which includes the actual cost elements such as manual detection working hours, equipment occupation time and necessary consumable consumption. After determining the unit execution cost, according to the sampling reference ratio and the expected yield of the quality inspection link, the detection times required for the link in the entire production cycle are calculated, and the total execution cost corresponding to the quality inspection link is obtained accordingly. After completing the above cost calculation for all quality inspection links listed in the dynamic quality inspection list, the execution costs of each quality inspection link are added up to obtain the estimated total execution cost required for implementing the current dynamic quality inspection strategy.

[0042] The quality risk cost is systematically quantified, and for each damage mode group, the quality loss record related to the group is extracted from the historical quality management data of the enterprise, including the actual economic loss caused by defect rework, scrap, claim, etc. Through the arrangement and statistics of historical data, the average quality loss cost corresponding to a single aluminum decoration part defect under the damage mode group is calculated. After completing the calculation of the average loss cost, combined with the defect probability information obtained in the previous defect prediction step, the quality risk that may occur in the current production cycle is estimated. The occurrence probability of each type of defect under the same damage mode group is associated with the corresponding average quality loss cost, and the expected quality risk cost that the group may bring under the current production condition is calculated. After performing the above calculation for all damage mode groups, the expected quality risk costs corresponding to each group are summarized to obtain the total expected quality risk cost under the current dynamic quality inspection strategy.

[0043] After obtaining the estimated total execution cost and the total expected quality risk cost, an optimization decision is made on the quality inspection strategy. In this embodiment, minimizing the estimated total execution cost and minimizing the total expected quality risk cost are taken as two optimization objectives that exist simultaneously and restrict each other, and a multi-objective optimization problem model is constructed. In the optimization model, the detection nodes involved in the dynamic quality inspection list, the sampling frequency corresponding to each detection node, and the tolerance standard used in the detection process are taken as optimization decision variables. In specific implementation, the value range of the above decision variables is limited on the premise that the detection intensity level and the process constraint condition are not destroyed, and through a multi-objective optimization solving process, a parameter combination that achieves a reasonable balance between cost investment and quality risk is searched. After the solving is completed, one solution that meets the current production management objective is selected from the obtained Pareto optimal solution set, and the detection node configuration, sampling frequency adjustment result, and tolerance standard adjustment amount corresponding to the solution are sorted and output. Finally, the above parameter combination is converted into a structured form as the quality inspection performance indicators of the next production cycle, which are used to control the production standards of the aluminum trim part quality.

[0044] The above embodiments can be realized wholly or partially by software, hardware, firmware, or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center through a wired (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0045] Those of skill in the art would understand that the modules and algorithms described in connection with the examples described herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. The

[0046] Those of skill in the art would understand that, for the purposes of description and brevity, the above description has described the system, device and module in detail, and the corresponding process in the foregoing method embodiments can be referred to for the specific working process of the system, device and module, which will not be described here.

[0047] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative, for example, the division of the modules is merely a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed modules can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or other forms.

[0048] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical modules, which can be located in one place or distributed on a plurality of network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0049] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can exist physically, or two or more modules can be integrated in one module.

[0050] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0051] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0052] Finally: the above is only a preferred embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A quality inspection method for automotive aluminum trim parts based on big data, characterized in that, Includes the following steps: S1. Extract damage records containing batch identifiers of aluminum trim parts from vehicle after-sales maintenance data sources, perform data cleaning on the damage records, and generate structured batch damage data. S2. Perform unsupervised learning based on spectral clustering on batch damage data to classify the damage pattern groups of aluminum decorative parts. S3. Based on the batch identifier of the aluminum trim parts, trace the time-series production process data of the corresponding production cycle, and cut it into production segment feature sequences using the sliding window method; S4. For each damage pattern family, input the feature sequences of all batches of production fragments contained therein into the integrated classification framework to identify key process feature combinations. S5. Based on the combination of key production process characteristics, construct an early propagation network for aluminum decorative parts process defects to predict the types of measurable defects and their probability of occurrence. S6. Dynamically generate multi-level hierarchical quality inspection strategies based on the predicted defect types and occurrence probabilities. S7. Integrate the estimated execution cost of multi-level hierarchical quality inspection strategies with historical quality loss cost data corresponding to each damage mode group, and output a set of quality inspection performance indicators for the next production cycle through a dynamic bi-objective optimization algorithm.

2. The quality inspection method for automotive aluminum trim parts based on big data according to claim 1, characterized in that, In step S1, generating structured batch damage data specifically includes: Access vehicle after-sales maintenance data sources and extract damage records for aluminum trim parts. The records include batch identifiers of the aluminum trim parts and descriptions of the damage morphology. Data cleaning is performed on the extracted damage records to unify the format of part batch identifiers and map the damage morphology descriptions to standardized damage morphology codes. The damage records after cleaning are integrated according to the batch identifier of the aluminum trim parts to generate a structured damage dataset indexed by the part batch.

3. The quality inspection method for automotive aluminum trim parts based on big data according to claim 1, characterized in that, In step S2, the specific classification of damage mode groups for aluminum trim parts includes: Based on the frequency of occurrence of various damage morphology codes of each batch of aluminum trim parts in the structured damage dataset, calculate the cosine similarity between any two batches of aluminum trim parts. Construct a symmetric batch similarity matrix for parts and calculate the corresponding Laplacian matrix. Perform eigenvalue decomposition on the Laplacian matrix and select the eigenvectors corresponding to the first few largest eigenvalues ​​to form a new feature space. In the new feature space, a clustering algorithm is used to divide all aluminum trim parts into several mutually exclusive sets. Each set is defined as a damage pattern family and labeled with a family identifier.

4. The quality inspection method for automotive aluminum trim parts based on big data according to claim 1, characterized in that, In step S3, based on the batch identifier of the aluminum trim parts, the time-series production process data corresponding to the production cycle is traced, and the sliding window method is used to cut it into a production segment feature sequence, specifically including: Based on the batch identifier of aluminum trim parts, retrieve multi-dimensional time series data of each batch recorded along the complete production chain from historical production lines; The retrieved multidimensional time series data are aligned and stitched together according to the manufacturing process nodes to form a continuous global production process data stream for this batch; Based on the theoretical production cycle time of a single product, the width and step interval of the sliding window are defined, and the global production process data stream is continuously captured along the time axis. For each data segment captured by the sliding window, calculate the statistical characteristics of various production parameters contained therein within the time window. Arrange the statistical characteristics of all windows in the batch in chronological order to form the production segment feature sequence corresponding to this batch.

5. The quality inspection method for automotive aluminum trim parts based on big data according to claim 1, characterized in that, In step S4, for each damage pattern family, the feature sequences of all batches of production fragments contained therein are input into the integrated classification framework to identify key process feature combinations, specifically including: For each damage mode family, the production segment feature sequence corresponding to the batch identifier of all aluminum trim parts in the family is used as the positive sample set. Normal parts batches that are not classified into any damage mode family are extracted from historical production data, and the corresponding production segment feature sequence is used as the negative sample set. An ensemble classification model is trained using positive and negative sample sets, taking production fragment feature sequences as input and batch classification prediction as output. The cumulative contribution of each process feature to the sample distinction result during the classification decision node division in the model training process is statistically analyzed. The importance ranking of the process features on which the integrated classification model depends in the classification decision is calculated, and the subset of features with the highest importance is selected as the key process feature combination.

6. The quality inspection method for automotive aluminum trim parts based on big data according to claim 1, characterized in that, In step S5, constructing an early propagation network for process defects in aluminum decorative parts and predicting the types and probabilities of measurable defects specifically includes: Key process features are combined and mapped to corresponding manufacturing nodes. A directed graph network model is constructed based on the sequential relationship of manufacturing nodes. Nodes represent manufacturing operations that carry key process features, and edges represent the propagation relationship of defects along the process path. By utilizing the co-occurrence relationship between the same key process features extracted from historical production data and the final detected defects, the conditional probability of defect state transition between adjacent nodes in the directed graph network model is calculated. In the current batch of aluminum decorative parts, the feature sequence of production segments containing key process characteristics is used as input. Forward propagation reasoning is performed along the process path in the directed graph network model to calculate the probability of various defects that are finally detected in each downstream process in real time. The defect types with probabilities exceeding the set confidence level and their corresponding probability values ​​are used as prediction outputs.

7. The quality inspection method for automotive aluminum trim parts based on big data according to claim 1, characterized in that, In step S6, the dynamic generation of a multi-level hierarchical quality inspection strategy based on the predicted defect type and probability of occurrence specifically includes: The probability of occurrence for each predicted defect type is divided into different risk level ranges; For each risk level range, a preset detection intensity level is matched. The detection intensity level defines the detection type and sampling benchmark ratio corresponding to the risk level defect, and at the same time determines the quality inspection process to be performed for the corresponding detection intensity level. By integrating the risk level, inspection intensity level, and quality inspection information corresponding to all predicted defect types in the current batch of aluminum trim parts, a dynamic quality inspection checklist with multi-level hierarchical quality inspection strategies is generated.

8. The quality inspection method for automotive aluminum trim parts based on big data according to claim 1, characterized in that, In step S7, the set of quality inspection performance indicators for the next production cycle is output, specifically including: Based on the inspection intensity level and execution node defined in the dynamic quality inspection checklist, and combined with the expected output in the production plan, calculate the estimated total execution cost required to implement the corresponding quality inspection strategy. For each damage mode family, historical quality loss cost data is collected, the average loss cost of aluminum trim parts defects under that family is calculated, and the total expected quality risk cost is calculated by combining the predicted probability of occurrence of various defects. Minimizing the estimated total execution cost and the total expected quality risk cost are the two optimization objectives. The detection nodes, sampling frequency and tolerance standards are the optimization decision variables. A multi-objective optimization problem model is constructed and solved. The solution results are converted into a set of quality inspection performance indicators for the next production cycle.

Images