A cable on-line defect detection method and system based on spectral vision fusion

Abstract

The application relates to the technical field of cable defect detection, in particular to a cable online defect detection method and system based on spectrum vision fusion. The method comprises the following steps: collecting multi-modal data of a cable surface on a cable extrusion production line; extracting a fusion feature vector from the multi-modal data; inputting the fusion feature vector into a pre-trained defect detection model to obtain a defect type; determining a candidate process parameter list according to the defect type; sequentially calling process parameters from the candidate process parameter list to perform an adjustment re-inspection step until no defect is identified or the candidate process parameter list is traversed; wherein the adjustment re-inspection step comprises: adjusting the current process parameters and performing defect detection on the newly generated cable after the adjustment. The application can realize a full-link intelligent quality control closed loop of cable defect online detection, intelligent diagnosis and process self-adaptive adjustment, and greatly improve the defect management efficiency of the production line and the stability of the finished product quality.

Description

Technical Field

[0001] This application relates to the field of cable defect detection technology, and in particular to an online cable defect detection method and system based on spectral visual fusion. Background Technology

[0002] In the field of cable extrusion production and processing, the quality of finished cables directly determines the safety and stability of power transmission. As the core link in cable forming, the extrusion process is prone to various defects such as surface damage, internal bubbles, and uneven material composition. Therefore, high-precision, multi-dimensional, and real-time online defect detection is the key to cable production quality control.

[0003] Currently, the industry generally adopts traditional technical solutions for cable production quality inspection, such as single machine vision surface imaging inspection, manual offline sampling re-inspection, or simple laser ranging screening. Most of the related inspection methods only focus on the identification of apparent defects on the cable surface, and the existing related technical solutions and patents only optimize and improve single vision algorithms and single types of laser detection methods, without realizing the integrated application of multiple detection modalities. The core deficiency of existing technologies lies in their limited detection dimensions. They can only identify visible defects on the cable surface, such as bulges, breaks, and eccentricities, and cannot accurately detect hidden defects such as air bubbles, interface delamination, internal impurities, and uneven cross-linking distribution within the insulation layer. These hidden defects can lead to safety accidents such as breakdown, leakage, and short circuits during long-term cable service. At the same time, traditional manual sampling inspection suffers from problems such as lag, low coverage, and large human error. It cannot achieve real-time online full inspection of the entire production line, resulting in a high rate of missed and false detections. More importantly, existing detection technologies can only detect defects but cannot accurately trace the causes of defects or make targeted reverse adjustments to extrusion production process parameters. It is difficult to form an intelligent quality control closed loop of detection, diagnosis, and control, and thus cannot meet the high-precision quality control requirements of the entire chain in modern intelligent cable manufacturing. Summary of the Invention

[0004] To address the problem that existing technologies have limited detection dimensions and cannot form a closed loop of detection, diagnosis, and control, this application provides an online cable defect detection method and system based on spectral visual fusion.

[0005] Firstly, this application provides an online cable defect detection method based on spectral visual fusion, employing the following technical solution: An online cable defect detection method based on spectral visual fusion includes: On the cable extrusion production line, multimodal data of the cable surface is collected. The multimodal data includes machine vision array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data. Extract fused feature vectors from the multimodal data, and input the fused feature vectors into a pre-trained defect detection model to obtain the defect type; A list of candidate process parameters is determined based on the defect type; The process parameters are retrieved sequentially from the candidate process parameter list and the adjustment and re-inspection steps are performed until no defects are identified or the candidate process parameter list has been traversed. The adjustment and re-inspection step includes: adjusting the current process parameters and performing defect detection on the newly generated cable after adjustment.

[0006] By adopting the above technical solution, three types of multimodal data—visual, Raman spectroscopy, and laser ultrasound—are simultaneously collected on the cable extrusion production line. This enables comprehensive and simultaneous detection of cable surface morphology, material composition, and internal structural defects. Based on multimodal fusion features and pre-trained models, defect types are accurately identified. Candidate process parameter schemes are then generated based on defect type matching. Defect rectification is completed through iterative adjustment and re-inspection closed loop. This effectively solves the problems of incomplete dimensions and missed or false detection of hidden defects in traditional single detection methods. It realizes a full-link intelligent quality control closed loop of online cable defect detection, intelligent diagnosis, and adaptive process adjustment, significantly improving the efficiency of production line defect management and the stability of finished product quality.

[0007] In a preferred embodiment, this application can be further configured such that the extraction of the fused feature vector from the multimodal data includes: Visual feature vectors, spectral feature vectors, and ultrasonic feature vectors are extracted from the multimodal data; Collect current production environment data, which includes: ambient temperature, ambient humidity, extrusion temperature, and equipment vibration parameters; Real-time acquisition of quality assessment data for each modality; determination of modality fusion weights based on the current production environment data and the quality assessment data. The visual feature vector, the spectral feature vector, and the ultrasonic feature vector are weighted and fused according to the modal fusion weights to obtain the fused feature vector.

[0008] By adopting the above technical solution, during the multimodal feature fusion process, the fusion weights are dynamically allocated by simultaneously combining real-time production environment data and real-time quality assessment data of each modality. This replaces the traditional fixed-weight fusion mode and can adaptively adjust the fusion ratio of visual, spectral, and ultrasonic features according to the fluctuation of working conditions and the quality of data. This effectively reduces the feature distortion problem caused by environmental interference and data anomalies, strengthens the representation role of highly reliable modal features, and significantly improves the accuracy and robustness of fused features under complex production conditions, thereby ensuring the accuracy of subsequent defect identification.

[0009] In a preferred embodiment, this application can be further configured such that: determining the modal fusion weights based on the current production environment data and the quality assessment data includes: Based on the pre-calibrated environmental interference mapping relationship, determine the interference coefficient of the current production environment data on each mode; Based on the quality assessment data, determine the overall quality score for each modality of data; The correction coefficient for the mode corresponding to the highest overall quality score is set to a first value less than one, and the correction coefficient for the mode corresponding to the lowest overall quality score is set to a second value greater than one. Calculate the product of the interference coefficient and the corresponding correction coefficient for each mode, and use it as the dynamic interference level value for the corresponding mode; The initial weights of each mode are adjusted according to the dynamic interference level value to obtain the mode fusion weight; wherein, the higher the dynamic interference level value, the smaller the weight adjustment value.

[0010] By adopting the above technical solution, the modal interference coefficients of each environmental parameter are quantified based on the pre-calibrated environmental interference mapping relationship. Then, positive and negative correction coefficients are set differently based on the comprehensive quality scores of each modality. The dynamic interference level value is calculated by coupling the interference coefficient and the correction coefficient. The initial weights of each modality are then adaptively fine-tuned according to the interference level, so that the modality with high data quality and low interference level receives higher weights. This further aligns with the real reliability of real-time detection data, accurately avoids detection deviations caused by working condition disturbances and data quality differences, and greatly improves the adaptive capability of multimodal feature fusion and the working condition adaptability of defect identification.

[0011] In a preferred example, this application can be further configured as: a pre-calibration process for environmental interference mapping relationships, including: Acquire historical operational data, including historical production environment data and historical quality assessment data for each modality; Analyze the historical production environment data and the historical quality assessment data to calculate the negative correlation coefficient between each environmental parameter and each quality assessment indicator; Based on the negative correlation coefficient, the interference coefficient of each mode corresponding to each environmental parameter interval is determined and stored as an environmental interference mapping relationship; wherein, the more serious the deviation of the environmental parameter from the standard operating condition and the larger the absolute value of the negative correlation coefficient, the larger the corresponding interference coefficient.

[0012] By adopting the above technical solution, the negative correlation between historical environmental data and modal quality data is explored. Based on the degree of deviation of environmental parameters from standard operating conditions and the magnitude of the negative correlation coefficient, the modal interference coefficient corresponding to each environmental parameter interval is calibrated. This ensures that the interference coefficient is larger in intervals with more severe abnormal operating conditions, accurately quantifies the negative impact intensity of the environment on each detection mode, and constructs a scientific and reliable prior interference mapping model. This provides an accurate and traceable calibration basis for online real-time dynamic weight adjustment, ensuring the rationality and accuracy of dynamic weight adjustment.

[0013] In a preferred embodiment, this application can be further configured such that determining the candidate process parameter list based on the defect type includes: Obtain a historical adjustment record library, which includes historical defect types, historical adjustment process parameter combinations, and historical adjustment effect tags; Retrieve historical adjustment records that match the defect type from the historical adjustment record database; Analyze the historical adjustment records to determine the historical elimination confidence level of each historical adjustment process parameter combination for the defect type; The historical adjustment process parameter combinations are sorted from high to low according to the historical elimination confidence level to generate a candidate process parameter list, where each element in the candidate process parameter list represents a process parameter combination.

[0014] By adopting the above technical solution, based on the historical defect rectification record database, the historical process adjustment scheme corresponding to the current defect type is accurately matched. By quantitatively calculating the defect elimination confidence of each process parameter combination and sorting them by priority to generate a candidate parameter list, the mode of blindly adjusting parameters based on manual experience is abandoned. The optimal rectification process scheme is selected based on big data patterns, which effectively improves the pertinence and effectiveness of process parameter adjustment, shortens the defect iteration rectification time, and improves the efficiency of closed-loop management of production line defects.

[0015] In a preferred embodiment, this application can be further configured such that: analyzing the historical adjustment records to determine the historical elimination confidence level of each historical adjustment process parameter combination for the defect type includes: The total number of adjustments and the number of successful eliminations for each historical adjustment process parameter combination for the defect type are recorded in the historical adjustment record. The ratio of the number of successful eliminations to the total number of adjustments for the historical combination of adjusted process parameters is calculated as the historical elimination confidence level.

[0016] By adopting the above technical solution, the total number of historical adjustments and successful eliminations of each process parameter combination for the corresponding defect are statistically analyzed. The success rate is directly used as the historical elimination confidence level, which quantitatively and intuitively reflects the actual defect repair capability of each process solution. This eliminates the interference of accidental rectification factors, ensures the authenticity and reliability of the candidate process parameter list ranking results, and allows the process solutions with higher priority to have higher defect repair stability, further ensuring the accuracy and reliability of process adaptive adjustment.

[0017] In a preferred embodiment, this application can be further configured such that the defect detection of the newly generated cable after adjustment includes: Obtain the physical transmission distance along the cable travel path between the defect detection point and the current process parameter adjustment execution point, and calculate the ratio of the physical transmission distance to the cable travel speed as the theoretical delay time; The system response delay is obtained, which includes command transmission delay and process actuator action delay; The sum of the theoretical delay and the system response delay is calculated and used as the delay detection time window; Using the moment when the process parameter adjustment instruction is issued as the starting point for timing, when the delayed detection time window arrives, the newly generated cable after adjustment is subjected to defect detection.

[0018] By adopting the above technical solution, taking into account the material transmission delay and system equipment response delay in continuous cable production, a precise delay detection time window is obtained through superposition calculation. Taking the time when the parameter adjustment command is issued as the unified timing starting point, the cable section produced under steady-state conditions of the new process parameters is accurately locked for re-inspection. This effectively avoids the problems of invalid detection and misjudgment of results caused by the transition of materials and equipment under the old process, ensuring that the re-inspection data after each process adjustment can truly reflect the process rectification effect, and improving the detection accuracy and operating condition adaptability of the entire closed-loop control system.

[0019] Secondly, this application provides an online cable defect detection system based on spectral visual fusion, employing the following technical solution: The data acquisition module is used to collect multimodal data of the cable surface on the cable extrusion production line. The multimodal data includes machine vision array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data. The defect identification module is used to extract a fusion feature vector from the multimodal data, input the fusion feature vector into a pre-trained defect detection model, and obtain the defect type. The parameter determination module is used to determine a list of candidate process parameters based on the defect type. The parameter adjustment module is used to sequentially retrieve process parameters from the candidate process parameter list and perform adjustment and re-inspection steps until no defects are identified or the candidate process parameter list has been traversed; wherein, the adjustment and re-inspection steps include: adjusting the current process parameters and performing defect detection on the newly generated cable after adjustment.

[0020] Thirdly, this application provides an electronic device that adopts the following technical solution: At least one processor; Memory; At least one application, wherein the at least one application is stored in memory and configured to be executed by at least one processor, the at least one application being configured to: perform the online cable defect detection method based on spectral visual fusion as described in any of the first aspects.

[0021] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the online cable defect detection method based on spectral visual fusion as described in any of the first aspects.

[0022] In summary, this application includes the following beneficial technical effects: This application simultaneously collects three types of multimodal data—visual, Raman spectroscopy, and laser ultrasound—on a cable extrusion production line, enabling comprehensive and simultaneous detection of cable surface morphology, material composition, and internal structural defects. It accurately identifies defect types based on multimodal fusion features and pre-trained models, then generates candidate process parameter schemes based on defect type matching. Through iterative parameter adjustment and re-inspection closed-loop, defect rectification is completed. This effectively solves the problems of incomplete dimensions and missed or false detections of latent defects in traditional single-detection methods. It achieves a full-chain intelligent quality control closed loop encompassing online cable defect detection, intelligent diagnosis, and adaptive process adjustment, significantly improving the efficiency of production line defect management and the stability of finished product quality. Attached Figure Description

[0023] Figure 1 This is a schematic flowchart of an online cable defect detection method based on spectral visual fusion provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of an online cable defect detection system based on spectral visual fusion provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] The following is in conjunction with the appendix Figure 1 To be continued Figure 3 This application will be described in further detail.

[0025] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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.

[0027] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0028] It should be noted that all data interaction processes involved in this application have corresponding transmission protocols, including authorized data collection and use, and both parties involved in the data interaction have completed data authorization through the execution of the protocol.

[0029] This application provides an online cable defect detection method based on spectral visual fusion, such as... Figure 1 As shown, the method provided in this application embodiment is executed by an electronic device, which can be a server or a terminal device. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal device and the server can be directly or indirectly connected via wired or wireless communication. This application embodiment does not impose any limitations on this connection. The method includes steps S101-S104, wherein: S101. On the cable extrusion production line, collect multimodal data of the cable surface. The multimodal data includes machine vision array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data.

[0030] The cable extrusion production line refers to the continuous production equipment chain for extruding cable insulation and sheath layers. Its core components include an extruder, die, traction machine, and cooling device. This is the core process leading to cable defects and also the online detection scenario for this method. Multimodal data refers to the synchronously collected cable production and inspection data across three differentiated dimensions: machine vision area array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data. These correspond to the three major inspection dimensions of cable appearance, material composition, and internal structure, respectively. These three types of data complement each other, covering all dimensions of defect information related to the cable surface, interior, and material.

[0031] Traditional single-mode detection can only cover defects in a single dimension. This step uses three types of detection equipment—vision, spectroscopy, and ultrasound—to simultaneously collect data. By leveraging the detection advantages of different sensing technologies, it identifies defects in cable surface morphology, material composition and performance, and hidden defects in internal structure, respectively. This achieves full-dimensional data collection of cable forming quality, providing a complete data foundation for accurate defect identification in the future.

[0032] At the fixed inspection station after shaping and cooling and before finished product winding on the cable extrusion production line, three types of inspection equipment are integrated: an area array industrial camera, a Raman spectrometer, and a laser ultrasonic flaw detector. These devices are synchronized with the production line control system and operate in real-time. During normal continuous operation of the production line, data acquisition from these three types of equipment is initiated simultaneously: the area array industrial camera continuously captures high-definition two-dimensional imaging data of the cable's outer surface, recording information on surface flatness, contour, and appearance defects; the Raman spectrometer performs fixed-point spectral acquisition of the cable sheath and insulation materials, obtaining compositional data such as molecular structure, component ratio, and degree of cross-linking; and the laser ultrasonic flaw detector emits laser ultrasonic signals that penetrate the cable surface, receives reflected echo signals, and collects internal flaw detection data such as bubbles, impurities, delamination, and structural inhomogeneity. Finally, these data are aggregated to form a real-time synchronized multimodal raw dataset of the cable.

[0033] S102. Extract the fusion feature vector from the multimodal data, input the fusion feature vector into the pre-trained defect detection model, and obtain the defect type.

[0034] The fused feature vector is a high-dimensional comprehensive feature set obtained by extracting specific features from the original data of visual, spectral, and ultrasonic modalities and then fusing them. It integrates all defect feature information of cable appearance, material, and internal structure. The pre-trained defect detection model is an artificial intelligence recognition model pre-trained based on massive amounts of multimodal cable defect sample data. It has the ability to match, classify, and recognize various cable defect features and can automatically determine the defect type. Defect types include, but are not limited to: porosity, impurities, eccentricity, damage, and abnormal cross-linking.

[0035] S103. Determine the list of candidate process parameters based on the defect type.

[0036] The candidate process parameter list consists of multiple combinations of optimizable process parameters selected for the currently identified defect type. The list is sorted by the effectiveness of defect repair.

[0037] After obtaining the current cable defect type, the system automatically searches the database to match all historical process adjustment schemes that highly correspond to the current defect. Based on the historical adjustment success rate and defect elimination effect, the system calculates and rates the repair confidence of each process parameter combination. According to the repair effect from best to worst, all matching process parameter combinations are sorted and organized to generate an ordered list of candidate process parameters. Each item in the list is a complete set of process parameter combinations that can be directly called.

[0038] S104. Retrieve process parameters sequentially from the candidate process parameter list and perform adjustment and re-inspection steps until a defect-free result is identified or the candidate process parameter list has been traversed. The adjustment and re-inspection steps include: adjusting the current process parameters and performing defect detection on the newly generated cable after adjustment.

[0039] A single process adjustment scheme cannot guarantee the elimination of defects 100%. By prioritizing iterative parameter adjustment, the optimal repair scheme is adopted first for production optimization. After each parameter adjustment, the quality of newly produced cables is monitored in real time. Production defects are eliminated to the maximum extent through iterative process. When the defects cannot be repaired, the adjustment is terminated in time, taking into account both production quality and production efficiency, and realizing a full-chain intelligent closed loop of detection, control and re-inspection.

[0040] From the generated list of candidate process parameters, the process parameter combination with the highest repair confidence and ranked first is prioritized. A parameter adjustment command is then sent to the extrusion production line equipment to automatically update parameters such as extruder temperature, production speed, and raw material ratio. After parameter adjustment, the production line is allowed to stabilize and the cable is output under the new parameters. The newly produced cable segment is then subjected to the aforementioned multimodal defect detection process to re-identify any defects. If the detection result is no defects, the production quality is deemed to meet standards, the process adjustment and re-inspection process is stopped, and the production line resumes normal steady-state operation. If the corresponding defect is still detected, the current parameter combination is abandoned, and the next process parameter combination in the list is retrieved sequentially (at this time, the previous process parameter combination can be restored to its value before adjustment). The parameter adjustment, production, and defect re-inspection process is repeated. This iterative process continues until no defects are detected or all parameter combinations in the candidate process parameter list are completely traversed, ending the current defect control loop.

[0041] This embodiment simultaneously collects three types of multimodal data—visual, Raman spectroscopy, and laser ultrasound—on the cable extrusion production line to achieve comprehensive and simultaneous detection of cable surface morphology, material composition, and internal structural defects. Relying on multimodal fusion features and pre-trained models, it accurately identifies defect types, generates candidate process parameter schemes based on defect type matching, and completes defect rectification through iterative parameter adjustment and re-inspection closed loop. This effectively solves the problems of incomplete dimensions and missed or false detection of hidden defects in traditional single detection methods, realizing a full-link intelligent quality control closed loop of online cable defect detection, intelligent diagnosis, and adaptive process adjustment, significantly improving the efficiency of production line defect management and the stability of finished product quality.

[0042] One possible implementation of this application embodiment involves extracting a fused feature vector from multimodal data, including: Extract visual feature vectors, spectral feature vectors, and ultrasonic feature vectors from multimodal data; Collect current production environment data, including: ambient temperature, ambient humidity, extrusion temperature, and equipment vibration parameters; Real-time acquisition of quality assessment data for each modality, and determination of modality fusion weights based on current production environment data and quality assessment data; The visual feature vector, spectral feature vector, and ultrasonic feature vector are weighted and fused according to the modality fusion weights to obtain the fused feature vector.

[0043] In this embodiment, the acquired machine vision array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data are first subjected to unified denoising, filtering, data alignment, and normalization preprocessing to eliminate equipment acquisition noise and data scale differences. For the machine vision array imaging data, an image feature extraction algorithm is used to extract key information such as cable surface texture, contour, defect location, and deformation deviation to construct a fixed-dimensional visual feature vector. For the Raman spectroscopy material composition data, core information such as spectral characteristic peak position, peak intensity, peak width, composition matching degree, and crosslinking degree parameters are extracted to construct a spectral feature vector. For the laser ultrasonic internal flaw detection data, core information such as ultrasonic echo delay, signal attenuation, number of abnormal reflection points, and structural layering characteristic parameters are extracted to construct an ultrasonic feature vector. Finally, three sets of independent, dimensionally unified single-mode feature vectors are obtained.

[0044] Production environment data represents the on-site operating parameters that directly affect the accuracy of the three modal detection methods. These parameters can include ambient temperature, ambient humidity, extrusion temperature, and equipment vibration amplitude. Different production environment parameters have varying degrees of interference with visual, spectral, and ultrasonic detection. For example, temperature and humidity affect the accuracy of spectral detection, extrusion temperature interferes with the spectral characteristics of cable materials, and equipment vibration causes blurred visual imaging and ultrasonic signal deviation.

[0045] Modality fusion weights represent the dynamic weight coefficients assigned to the three types of feature vectors: visual, spectral, and ultrasonic. The magnitude of the weight corresponds to the reliability of the current detection data for the corresponding modality. The less environmental interference and the higher the data quality, the greater the corresponding weight and the higher its proportion in the fused features.

[0046] The extracted visual, spectral, and ultrasonic feature vectors are then weighted and multiplied with the corresponding modal fusion weights obtained from the solution. The weighted feature vectors are then subjected to dimensional concatenation, feature fusion, and redundant dimension compression to unify the feature data scale and distribution, generating a final fused feature vector containing all valid information about surface defects, material defects, and internal structural defects. This fused feature vector is then input into a pre-trained defect detection model to complete the subsequent defect type identification process.

[0047] In this embodiment, during the multimodal feature fusion process, real-time production environment data and real-time quality assessment data of each modality are combined simultaneously to dynamically allocate fusion weights, replacing the traditional fixed-weight fusion mode. It can adaptively adjust the fusion ratio of visual, spectral, and ultrasonic features according to the fluctuation of working conditions and the quality of data, effectively reducing feature distortion caused by environmental interference and data anomalies, strengthening the representation role of highly reliable modal features, significantly improving the accuracy and robustness of fused features under complex production conditions, and thus ensuring the accuracy of subsequent defect identification.

[0048] One possible implementation of this application embodiment determines the modal fusion weights based on current production environment data and quality assessment data, including: Based on the pre-calibrated environmental interference mapping relationship, determine the interference coefficient of the current production environment data on each mode; Based on the quality assessment data, determine the overall quality score for each modality of data; The correction coefficient for the mode corresponding to the highest overall quality score is set to a first value less than one, and the correction coefficient for the mode corresponding to the lowest overall quality score is set to a second value greater than one. Calculate the product of the interference coefficient and the corresponding correction coefficient for each mode, and use it as the dynamic interference level value for the corresponding mode; The initial weights of each mode are adjusted according to the dynamic interference level to obtain the mode fusion weights; among them, the higher the dynamic interference level, the smaller the weight adjustment value.

[0049] The environmental interference mapping relationship refers to the pre-established correspondence based on historical data calibration. This relationship quantifies the association rules between different production environment parameters and the interference levels of the three detection modalities: vision, spectroscopy, and ultrasound. It enables a standardized mapping from environmental parameters to modal interference levels. Each environmental parameter in the environmental interference mapping relationship contains multiple possible value ranges, with each range corresponding to an interference coefficient for each modality. The pre-calibrated and stored environmental interference mapping relationship database is called, and the four types of real-time collected production environment data—temperature, humidity, extrusion temperature, and vibration amplitude—are matched to their respective parameter ranges. Based on the interference coefficients corresponding to each parameter range, the single-modal interference coefficient of each environmental parameter for the machine vision modality, Raman spectroscopy modality, and laser ultrasound modality is determined. For each modality, the average single-modal interference coefficient of each environmental parameter for that modality can be calculated as the interference coefficient of the current production environment data for that modality.

[0050] Interference coefficients obtained solely through fixed environment mapping can only reflect the patterns of interference in normal environments and cannot adapt to sudden interference situations such as instantaneous equipment jitter, data acquisition anomalies, and abrupt changes in operating conditions. By introducing real-time data quality assessment indicators for each modality and dynamically correcting the initial interference coefficients, the limitations of fixed mapping models can be overcome, accurately reflecting the real-time detection state and improving the realism of interference assessment.

[0051] While acquiring multimodal data, the system simultaneously calculates quality assessment data for visual, spectral, and ultrasonic modalities in real time. Quality assessment indicators include, but are not limited to, quantitative metrics such as image sharpness, signal-to-noise ratio, feature completeness, and signal stability. The mapping relationship between quality assessment indicator values ​​and scores can be pre-defined. For each modal's quality assessment data, the average score of each quality assessment indicator or a weighted average based on preset weights is calculated as the comprehensive quality score for that modality. The correction coefficient corresponding to the highest comprehensive quality score is set as the lowest (e.g., 0.9); the correction coefficient corresponding to the second highest comprehensive quality score is set as 1; and the correction coefficient corresponding to the lowest comprehensive quality score is set as the highest (e.g., 1.1). The product of the interference coefficient and the correction coefficient for each modality is calculated as the dynamic interference level of the current production environment data to that modality.

[0052] A higher dynamic interference level indicates that the corresponding mode is more affected by the current production environment data. The dynamic interference levels of the three modes are sorted to obtain the highest, middle, and lowest dynamic interference levels. The ratio of the highest to the middle dynamic interference level is used as the first adjustment coefficient, and the ratio of the lowest to the middle dynamic interference level is used as the second adjustment coefficient.

[0053] The initial weights of each modality can be the same or set differently according to business needs; this embodiment does not impose any limitations, and the sum of the initial weights of all modalities is 1. For the modality corresponding to the highest dynamic interference level, the product of the first adjustment coefficient and the initial weight of that modality is calculated as the adjusted weight; for the modality corresponding to the intermediate dynamic interference level, its initial weight is not adjusted; for the modality corresponding to the lowest dynamic interference level, the product of the second adjustment coefficient and the initial weight of that modality is calculated as the adjusted weight. The combination of the adjusted weights of all modalities is the modality fusion weight.

[0054] This embodiment quantifies the modal interference coefficients of each environmental parameter based on the pre-calibrated environmental interference mapping relationship. Then, it sets positive and negative correction coefficients based on the differences in the comprehensive quality scores of each modality. The dynamic interference level value is calculated by coupling the interference coefficient and the correction coefficient. Based on the interference level, the initial weights of each modality are adaptively fine-tuned in reverse, so that the modality with high data quality and low interference level receives higher weights. This further aligns with the real reliability of real-time detection data, accurately avoids detection deviations caused by operating condition disturbances and data quality differences, and greatly improves the adaptive capability of multimodal feature fusion and the operating condition adaptability of defect identification.

[0055] One possible implementation of this application embodiment, the pre-calibration process of environmental interference mapping relationship, includes: Acquire historical operational data, which includes historical production environment data and historical quality assessment data for each modality; Analyze historical production environment data and historical quality assessment data to calculate the negative correlation coefficient between each environmental parameter and each quality assessment indicator. Based on the negative correlation coefficient, the interference coefficient of each mode corresponding to each environmental parameter range is determined and stored as an environmental interference mapping relationship; among them, the more serious the deviation of the environmental parameter from the standard operating condition and the larger the absolute value of the negative correlation coefficient, the larger the corresponding interference coefficient.

[0056] In this embodiment, for each environmental parameter (ambient temperature, ambient humidity, extrusion temperature, vibration amplitude) in the dataset, a paired correlation calculation is performed with each quality assessment indicator of the visual, spectral, and ultrasonic modalities. The Pearson correlation coefficient algorithm is used to solve for the correlation coefficient between each parameter and indicator; significantly negatively correlated associations are selected. The larger the absolute value of the correlation coefficient, the stronger the negative impact and the more significant the interference effect of the environmental parameter on the data quality of the corresponding modality. All valid correlation coefficients are retained to establish a parameter modality negative impact correlation data table.

[0057] Each environmental parameter is divided into uniform intervals, covering the normal production interval, fluctuation deviation interval, and extreme anomaly interval. Based on the obtained parameter modal negative impact correlation data table, a negative interference coefficient is assigned to each environmental parameter interval: the more severe the deviation of the environmental parameter from the standard operating condition and the larger the absolute value of the correlation coefficient, the larger the corresponding negative interference coefficient, indicating a stronger negative impact of that interval on the modal data; conversely, when the parameter is in the standard optimal operating condition interval, the negative interference coefficient is the smallest, indicating almost no negative impact. After completing the calibration of the negative interference coefficients for all parameter intervals and all modes, the environmental parameter intervals and their corresponding modal negative interference coefficients are associated and bound, and stored as a standardized environmental interference mapping relationship data table.

[0058] This embodiment explores the negative correlation between historical environmental data and modal quality data. Based on the degree to which environmental parameters deviate from standard operating conditions and the magnitude of the negative correlation coefficient, it calibrates the modal interference coefficient corresponding to each environmental parameter interval. This ensures that the more severe the abnormal operating conditions, the larger the interference coefficient. It accurately quantifies the negative impact of the environment on each detection mode and constructs a scientific and reliable prior interference mapping model. This provides an accurate and traceable calibration basis for online real-time dynamic weight adjustment, ensuring the rationality and accuracy of dynamic weight adjustment.

[0059] One possible implementation of this application embodiment involves determining a list of candidate process parameters based on defect type, including: Obtain the historical adjustment record library, which includes historical defect types, historical adjustment process parameter combinations, and historical adjustment effect tags; Retrieve historical adjustment records that match the defect type from the historical adjustment record database; Analyze historical adjustment records to determine the historical elimination confidence level of each historical adjustment process parameter combination for the defect type; Sort each historical adjustment process parameter combination from high to low according to the historical elimination confidence level, and generate a candidate process parameter list. Each element in the candidate process parameter list represents a process parameter combination.

[0060] The historical adjustment record database is a pre-built and continuously updated production line process rectification database used to store historical defect rectification data for the entire cable extrusion production process. Each record forms a complete closed-loop data sample, and its core includes three types of fields: historical defect type, historical adjustment process parameter combination, and historical adjustment effect label. The real-time defect type output by the current defect detection model is used as a search keyword and input into the historical adjustment record database for precise matching retrieval; all archived records in the database are traversed to filter out all historical rectification records whose defect type is completely consistent with the current defect type; and all successfully matched historical samples are collected.

[0061] Historical elimination confidence is used to quantify the reliability of a set of historical process parameter combinations in eliminating the current defect type and the probability of successful rectification. The higher the confidence, the stronger the stability and effectiveness of the parameter combination in repairing the corresponding defect. Based on the historical elimination confidence of each set of process parameter combinations, all valid process parameter combinations are sorted in descending order, i.e., the higher the confidence value, the higher the sorting priority. The sorted process parameter combinations are then sequentially aggregated and arranged to form a structured candidate process parameter list. Each element in the list is a complete process parameter combination that can be directly deployed to the production line for execution.

[0062] This embodiment is based on a historical defect rectification record database, accurately matching the historical process adjustment schemes corresponding to the current defect type. By quantitatively calculating the defect elimination confidence of each process parameter combination and sorting them by priority, a candidate parameter list is generated. This abandons the mode of blindly adjusting parameters based on manual experience, and relies on big data patterns to screen the optimal rectification process scheme, effectively improving the pertinence and effectiveness of process parameter adjustment, shortening the defect iteration rectification time, and improving the efficiency of closed-loop management of production line defects.

[0063] One possible implementation of this application embodiment involves analyzing historical adjustment records to determine the historical elimination confidence level of each historical adjustment process parameter combination for the defect type, including: Statistics are compiled on the total number of adjustments and the number of successful eliminations for each historical combination of adjusted process parameters for the defect type in the historical adjustment record. The ratio of the number of successful eliminations to the total number of adjustments for historical combinations of adjusted process parameters is calculated and used as the historical elimination confidence level.

[0064] Different combinations of process parameters result in varying stability in repairing the same defect, and the quality of a solution cannot be judged solely based on the outcome of a single rectification attempt. By statistically analyzing the total number of adjustments and the number of successful rectifications for each parameter combination, the historical rectification performance of the solution can be objectively reflected.

[0065] Based on the matched historical rectification sample set specific to the current defect type, the samples are grouped and categorized according to different historical adjustments of process parameters. For each independent combination of process parameters, all corresponding historical rectification records are traversed, and the adjustment effect tags in each record are verified. The total number of times this parameter combination is executed for the current defect type is recorded as the total number of adjustments. Valid records with the adjustment effect tag indicating success are separately collected, and the cumulative number of successful defect eliminations for this parameter combination is calculated, completing the statistical analysis of sample data for all process parameter combinations.

[0066] For each combination of process parameters, the success rate of rectification is calculated as the number of successful eliminations divided by the total number of adjustments. This success rate is used as the basic weighting coefficient, which is then directly mapped to the initial weighted confidence level of the corresponding process parameter combination. Invalid samples with a total of 0 adjustments are directly removed to avoid data anomalies caused by a zero denominator. Finally, the initial weighted confidence levels of all valid process parameter combinations are obtained.

[0067] This embodiment statistically analyzes the total number of historical adjustments and successful eliminations of each process parameter combination for the corresponding defect. The success rate is directly used as the historical elimination confidence level, which quantitatively and intuitively reflects the actual defect repair capability of each process scheme. It eliminates the interference of accidental rectification factors, ensures the authenticity and reliability of the candidate process parameter list ranking results, and allows the process schemes with higher priority to have higher defect repair stability, further ensuring the accuracy and reliability of process adaptive adjustment.

[0068] One possible implementation of this application embodiment involves defect detection of the newly generated cable after adjustment, including: Obtain the physical transmission distance along the cable travel path between the defect detection point and the current process parameter adjustment execution point, and calculate the ratio of the physical transmission distance to the cable travel speed as the theoretical delay time; The system response latency is obtained, which includes command transmission latency and process actuator action latency. The sum of the theoretical delay and the system response delay is calculated and used as the delay detection time window; The timing starts from the moment the process parameter adjustment instruction is issued, and defect detection is performed on the newly generated cable after adjustment when the delayed detection time window arrives.

[0069] Cable extrusion is a continuous, uninterrupted production process. After process parameter adjustments, cables formed with the old parameters will remain in the production line path. Effective re-inspection can only be carried out after the cables formed with the new parameters are transported to the inspection station. By calculating the theoretical transmission time by the ratio of physical distance to production line speed, the basic time node for the arrival of new material at the inspection point can be accurately determined, avoiding the distortion of inspection results caused by interference from old material. The physical transmission distance along the path from the process adjustment execution point to the defect detection point is pre-fixed and stored; the uniform traction speed of the cable on the production line is collected in real time; the physical transmission distance is divided by the cable travel speed to calculate the theoretical delay time for the newly formed cable to be transported from the adjustment point to the inspection point, thus completing the quantitative calculation of material transmission delay.

[0070] Process parameter adjustments are not instantaneous; there is an inherent response delay in communication transmission and equipment mechanical actions. Cables produced during this period are still transitional products. The communication logs and equipment execution feedback data of the production line control system are read in real time to obtain the command transmission delay corresponding to this process parameter adjustment. Simultaneously, the action response time and stabilization time of each process actuator are collected to obtain the process actuator action delay. These two types of delay data are then combined to obtain the total system response delay corresponding to this parameter adjustment.

[0071] Effective re-inspection requires the simultaneous fulfillment of two conditions: completion of equipment condition adjustment and transmission of new parameters via cable to the inspection point. The theoretical delay time calculated step-by-step is summed with the system response delay to obtain the final delay detection time window; this time window represents the standard waiting time for effective defect re-inspection after a single process adjustment.

[0072] The timing begins at the moment the system issues the process parameter adjustment command. Real-time timing starts when the system timing reaches the preset delayed detection time window. When the system timing reaches the preset delayed detection time window, the multimodal defect detection process is automatically triggered. For the cable segment that has reached the current detection point and has been stabilized by the new process parameters, visual, spectral, and ultrasonic multimodal data acquisition and defect identification are performed to complete the targeted re-inspection after the process adjustment and obtain the actual defect rectification results.

[0073] This embodiment comprehensively considers the material transmission delay and system equipment response delay in continuous cable production, and calculates an accurate delay detection time window by superimposing the data. Taking the moment the parameter adjustment command is issued as the unified timing start point, it accurately locks the cable section produced under steady-state conditions of the new process parameters for re-inspection. This effectively avoids the problems of invalid detection and misjudgment of results caused by the transition of materials and equipment under the old process, and ensures that the re-inspection data after each process adjustment can truly reflect the process rectification effect, thereby improving the detection accuracy and operating condition adaptability of the entire closed-loop control system.

[0074] This application provides an online cable defect detection system based on spectral visual fusion. See [link to relevant documentation]. Figure 2 The system 200 includes a data acquisition module 201, a defect identification module 202, a parameter determination module 203, and a parameter adjustment module.

[0075] The data acquisition module 201 is used to collect multimodal data of the cable surface on the cable extrusion production line. The multimodal data includes machine vision area array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data.

[0076] The defect identification module 202 is used to extract fused feature vectors from multimodal data, input the fused feature vectors into a pre-trained defect detection model, and obtain the defect type.

[0077] The parameter determination module 203 is used to determine a list of candidate process parameters based on the defect type.

[0078] The parameter adjustment module 204 is used to sequentially retrieve process parameters from the candidate process parameter list and perform adjustment and re-inspection steps until a defect-free result is identified or the candidate process parameter list has been traversed. The adjustment and re-inspection steps include: adjusting the current process parameters and performing defect detection on the newly generated cable after adjustment.

[0079] This application provides an electronic device, such as... Figure 3 As shown, Figure 3 The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of this electronic device 300 does not constitute a limitation on the embodiments of this application.

[0080] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0081] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0082] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0083] The memory 303 is used to store the application code that executes the solution of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the aforementioned embodiment of the online cable defect detection method based on spectral visual fusion.

[0084] Figure 3 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0085] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the content shown in the aforementioned embodiment of the online cable defect detection method based on spectral visual fusion.

[0086] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0087] This application provides a computer program product, including a computer program that, when executed by a processor, implements the content shown in the aforementioned embodiment of the online cable defect detection method based on spectral visual fusion.

[0088] The above are only some embodiments of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for online cable defect detection based on spectral visual fusion, characterized in that, include: On the cable extrusion production line, multimodal data of the cable surface is collected. The multimodal data includes machine vision array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data. Extract fused feature vectors from the multimodal data, and input the fused feature vectors into a pre-trained defect detection model to obtain the defect type; A list of candidate process parameters is determined based on the defect type; The process parameters are retrieved sequentially from the candidate process parameter list and the adjustment and re-inspection steps are performed until no defects are identified or the candidate process parameter list has been traversed. The adjustment and re-inspection step includes: adjusting the current process parameters and performing defect detection on the newly generated cable after adjustment.

2. The online cable defect detection method based on spectral visual fusion according to claim 1, characterized in that, The extraction of the fused feature vector from the multimodal data includes: Visual feature vectors, spectral feature vectors, and ultrasonic feature vectors are extracted from the multimodal data; Collect current production environment data, which includes: ambient temperature, ambient humidity, extrusion temperature, and equipment vibration parameters; Real-time acquisition of quality assessment data for each modality; determination of modality fusion weights based on the current production environment data and the quality assessment data. The visual feature vector, the spectral feature vector, and the ultrasonic feature vector are weighted and fused according to the modal fusion weights to obtain the fused feature vector.

3. The online cable defect detection method based on spectral visual fusion according to claim 2, characterized in that, The step of determining the modal fusion weights based on the current production environment data and the quality assessment data includes: Based on the pre-calibrated environmental interference mapping relationship, determine the interference coefficient of the current production environment data on each mode; Based on the quality assessment data, determine the overall quality score for each modality of data; The correction coefficient for the mode corresponding to the highest overall quality score is set to a first value less than one, and the correction coefficient for the mode corresponding to the lowest overall quality score is set to a second value greater than one. Calculate the product of the interference coefficient and the corresponding correction coefficient for each mode, and use it as the dynamic interference level value for the corresponding mode; The initial weights of each mode are adjusted according to the dynamic interference level value to obtain the mode fusion weight; wherein, the higher the dynamic interference level value, the smaller the weight adjustment value.

4. The online cable defect detection method based on spectral visual fusion according to claim 1, characterized in that, The pre-calibration process for environmental interference mapping relationships includes: Acquire historical operational data, including historical production environment data and historical quality assessment data for each modality; Analyze the historical production environment data and the historical quality assessment data to calculate the negative correlation coefficient between each environmental parameter and each quality assessment indicator; Based on the negative correlation coefficient, the interference coefficient of each mode corresponding to each environmental parameter interval is determined and stored as an environmental interference mapping relationship; wherein, the more serious the deviation of the environmental parameter from the standard operating condition and the larger the absolute value of the negative correlation coefficient, the larger the corresponding interference coefficient.

5. The online cable defect detection method based on spectral visual fusion according to claim 1, characterized in that, The step of determining the candidate process parameter list based on the defect type includes: Obtain a historical adjustment record library, which includes historical defect types, historical adjustment process parameter combinations, and historical adjustment effect tags; Retrieve historical adjustment records that match the defect type from the historical adjustment record database; Analyze the historical adjustment records to determine the historical elimination confidence level of each historical adjustment process parameter combination for the defect type; The historical adjustment process parameter combinations are sorted from high to low according to the historical elimination confidence level to generate a candidate process parameter list, where each element in the candidate process parameter list represents a process parameter combination.

6. The online cable defect detection method based on spectral visual fusion according to claim 5, characterized in that, The analysis of the historical adjustment records, determining the historical elimination confidence level of each historical adjustment process parameter combination for the defect type, includes: The total number of adjustments and the number of successful eliminations for each historical adjustment process parameter combination for the defect type are recorded in the historical adjustment record. The ratio of the number of successful eliminations to the total number of adjustments for the historical combination of adjusted process parameters is calculated as the historical elimination confidence level.

7. The online cable defect detection method based on spectral visual fusion according to claim 1, characterized in that, The defect detection of the newly generated cable after adjustment includes: Obtain the physical transmission distance along the cable travel path between the defect detection point and the current process parameter adjustment execution point, and calculate the ratio of the physical transmission distance to the cable travel speed as the theoretical delay time; The system response delay is obtained, which includes command transmission delay and process actuator action delay; The sum of the theoretical delay and the system response delay is calculated and used as the delay detection time window; Using the moment when the process parameter adjustment instruction is issued as the starting point for timing, when the delayed detection time window arrives, the newly generated cable after adjustment is subjected to defect detection.

8. An online cable defect detection system based on spectral visual fusion, characterized in that, include: The data acquisition module is used to collect multimodal data of the cable surface on the cable extrusion production line. The multimodal data includes machine vision array imaging data, Raman spectroscopy material composition data, and laser ultrasonic internal flaw detection data. The defect identification module is used to extract a fusion feature vector from the multimodal data, input the fusion feature vector into a pre-trained defect detection model, and obtain the defect type. The parameter determination module is used to determine a list of candidate process parameters based on the defect type. The parameter adjustment module is used to sequentially retrieve process parameters from the candidate process parameter list and perform adjustment and re-inspection steps until no defects are identified or the candidate process parameter list has been traversed; wherein, the adjustment and re-inspection steps include: adjusting the current process parameters and performing defect detection on the newly generated cable after adjustment.

9. An electronic device, characterized in that, include: At least one processor; Memory; At least one application, wherein the at least one application is stored in memory and configured to be executed by at least one processor, said at least one application being configured to: perform the online cable defect detection method based on spectral visual fusion as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed in the computer, the computer is instructed to perform the online cable defect detection method based on spectral visual fusion as described in any one of claims 1-7.

Images