Machine vision-based intelligent detection system and method for flatness of brake spring plate

Abstract

The application relates to the technical field of brake component detection, and discloses a machine vision-based intelligent detection system and method for flatness of brake spring sheets. The method comprises the following steps: acquiring historical brake spring sheet image data, and generating a standard flatness curve according to the historical brake spring sheet image data; collecting real-time brake spring sheet image data through a machine vision system, and generating a real-time flatness curve; then calculating a flatness deviation curve according to the real-time flatness curve and the standard flatness curve; then analyzing the matching degree of the deviation curve and a reference defect model, and determining the flatness state of the brake spring sheet; finally, dynamically adjusting detection parameters based on the flatness state, and generating an optimized detection scheme. The method takes historical data as a benchmark, realizes non-contact detection by relying on machine vision, can accurately present the flatness deviation and clearly show the defect properties corresponding to the deviation, can dynamically optimize the detection scheme, improves the detection consistency, efficiency and adaptability, reduces the dependence on manual operation and the wear of components, and is suitable for batch detection of brake spring sheets.

Description

Technical Field

[0001] This invention relates to the field of brake component inspection technology, specifically to a machine vision-based intelligent inspection system and method for brake spring flatness. Background Technology

[0002] As a core component for the safe operation of various vehicles, the performance of the internal parts of the braking system directly affects the overall braking effect. Brake springs, as a key elastic element in the braking system, must maintain stable flatness under long-term stress, vibration, and temperature changes. Abnormal flatness deviations can lead to poor contact between the brake spring and surrounding components, causing problems such as delayed braking response and uneven brake pad wear. In severe cases, it can even affect the overall safety of the braking system. Therefore, accurate flatness testing of brake springs is a crucial step in the production and maintenance of brake components.

[0003] Traditional brake spring pad flatness inspection relies heavily on manual operation. Inspectors use tools such as dial indicators and straightedges to measure and record data point by point to determine whether the flatness is up to standard. This method not only requires inspectors to have extensive experience, but the inspection process is also time-consuming. The inspection of each spring pad often requires multiple adjustments to the measurement position to obtain comprehensive data, making it difficult to meet the efficient inspection needs of large-scale production scenarios. At the same time, manual inspection is susceptible to subjective factors. Different inspectors may have different standards for judging measurement errors, making it difficult to guarantee the consistency and reliability of the inspection results. Some minor flatness deviations may be missed, thus creating potential safety hazards for the subsequent use of the braking system.

[0004] As the manufacturing industry moves towards intelligence and automation, traditional manual inspection methods are gradually becoming inadequate for the high efficiency and accuracy requirements of modern production. Although some automated inspection equipment based on mechanical contact has emerged on the market, it may cause some wear to the surface of brake spring pads during the inspection process. This wear is especially problematic for spring pads with high surface precision requirements, as it can directly affect their performance. Furthermore, the inspection range of this type of mechanical contact inspection equipment is relatively fixed. For brake spring pads of different specifications and structures, it is often necessary to readjust the mechanical structure or replace the inspection tooling, resulting in poor flexibility and high maintenance costs. It is also difficult to achieve rapid switching between different types of brake spring pads for inspection.

[0005] With the gradual application of machine vision technology in industrial inspection, some companies have attempted to use it for component dimensional inspection. However, existing machine vision-based inspection methods mostly focus on surface defects or simple dimensional measurements. For flatness inspection of components with specific elastic structures, such as brake springs, a mature and systematic solution has yet to be developed. Existing methods often struggle to accurately extract the contour features of brake springs and generate reliable flatness curves. Furthermore, in the flatness deviation analysis process, there is a lack of reference models corresponding to the actual defect types, making it impossible to accurately determine the nature and severity of flatness deviations. This results in low practicality of the inspection results, making it difficult to effectively guide subsequent production adjustments or quality control. Summary of the Invention

[0006] The purpose of this invention is to provide a machine vision-based intelligent detection system and method for brake spring flatness, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a machine vision-based intelligent detection method for the flatness of brake spring pads, the method comprising:

[0008] Acquire historical brake spring pad image data, and generate a standard flatness curve based on the historical brake spring pad image data;

[0009] Real-time brake spring pad image data is acquired through a machine vision system, and a real-time flatness curve is generated based on the real-time brake spring pad image data.

[0010] Calculate the flatness deviation curve based on the real-time flatness curve and the standard flatness curve;

[0011] The flatness deviation curve is analyzed to determine the matching degree with the reference defect model, and the flatness status of the brake spring plate is determined.

[0012] The detection parameters are dynamically adjusted based on the flatness status to generate an optimized detection scheme.

[0013] Preferably, the step of generating a standard flatness curve based on the historical brake spring image data includes:

[0014] The historical brake spring pad image data is used to train a flatness model, which is used to extract features and generate a standard flatness curve.

[0015] Repeat the steps of training the flatness model under different detection conditions to obtain a standard flatness curve sequence.

[0016] Preferably, the step of calculating the flatness deviation curve based on the real-time flatness curve and the standard flatness curve includes:

[0017] Subtracting the flatness value of the standard flatness curve from the flatness value on the real-time flatness curve yields the flatness deviation curve;

[0018] Multiply the deviation value on the flatness deviation curve with the flatness value on the real-time flatness curve to obtain the flatness change curve.

[0019] Preferably, the step of determining the flatness state of the brake spring sheet includes:

[0020] Obtain a reference defect model from the pre-stored defect library, and calculate the matching degree between the flatness deviation curve and the reference defect model within the time window;

[0021] Identify the reference defect model corresponding to the minimum matching degree, and superimpose the reference defect model with the standard flatness curve to generate a flatness prediction result.

[0022] Preferably, the method further includes the step of configuring the detection target:

[0023] Based on the quality standards of brake spring pads, set the main inspection target, auxiliary inspection target, and maintenance target;

[0024] The main detection objective includes minimizing flatness error;

[0025] The auxiliary detection objectives include optimizing detection efficiency;

[0026] The stated objectives include maintaining light stability.

[0027] Preferably, the step of configuring the detection target includes:

[0028] Based on the aforementioned quality standards, the primary detection target, auxiliary detection target, and maintenance target are classified and identified.

[0029] Key detection parameters are selected using the classification identifiers.

[0030] Preferably, the method further includes the step of performing an impact analysis:

[0031] Analyze the degree of influence of the key detection parameters on the main detection target and the auxiliary detection target;

[0032] Parameter optimization constraints are generated based on the degree of influence.

[0033] Preferably, the step of performing the impact analysis includes:

[0034] Calculate the influence factors of the key detection parameters;

[0035] Based on the parameter optimization constraints and the influence factors, the optimized detection parameters are obtained through parameter optimization.

[0036] Preferably, the method further includes a step of sparse reconstruction processing of the real-time brake spring image data:

[0037] The real-time brake spring pad image data is converted into a multidimensional dataset;

[0038] Feature extraction processing is performed on the multidimensional dataset to obtain target data for detection and anomaly indication data for indicating anomalies;

[0039] The optimized detection scheme is dynamically adjusted based on the anomaly indication data.

[0040] Preferably, the present invention also includes a machine vision-based intelligent detection system for brake spring flatness, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the machine vision-based intelligent detection method for brake spring flatness as described above.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] By generating a standard flatness curve using historical brake spring pad image data, an objective and practically applicable reference standard is provided for brake spring pad flatness testing. Compared to traditional testing methods that rely on manual experience to set reference standards, the standard curve generated based on historical data can fully reflect the flatness characteristics of qualified brake spring pads, avoiding inconsistencies in reference standards caused by differences in subjective human judgment. This provides a unified and reliable basis for comparison in subsequent flatness testing, helping to improve the consistency and impartiality of test results.

[0043] By acquiring real-time image data of brake spring pads and generating real-time flatness curves through a machine vision system, non-contact operation for brake spring pad flatness inspection has been achieved. This non-contact inspection method effectively avoids the wear that traditional mechanical contact inspection may cause to the spring pad surface, making it particularly suitable for brake spring pad inspection scenarios with high surface precision requirements, ensuring that the original performance of the spring pad is not affected after inspection. Simultaneously, the machine vision system has rapid image acquisition and processing capabilities, enabling it to complete image acquisition and flatness curve generation for a single brake spring pad in a short time. Compared to manual point-by-point measurement or mechanical contact inspection, this significantly reduces the inspection time per pad, substantially improves overall inspection efficiency, and is more suitable for the inspection needs of large-scale batch production scenarios. It can effectively reduce inspection waiting time on the production line and ensure the smoothness of the production process.

[0044] The flatness deviation curve is calculated by comparing the real-time flatness curve with the standard flatness curve, providing a clear and accurate representation of the flatness difference between the brake spring pads being tested in real-time and meeting the acceptable standard. This method of presenting the deviation curve allows inspectors to clearly observe the flatness deviation of various parts of the brake spring pad, including the location, direction, and magnitude of the deviation. Compared to traditional testing methods that rely solely on numerical comparison, this method facilitates the rapid location of areas with abnormal flatness. Furthermore, by analyzing the matching degree between the flatness deviation curve and the reference defect model to determine the flatness status, the abstract deviation data can be correlated with the actual types of defects that may exist. This not only determines whether the brake spring pad has a flatness problem but also further clarifies the nature of the defect corresponding to the deviation, such as local bulges, depressions, or overall bending. This provides richer and more practical information for subsequent quality analysis, helping technicians to specifically identify the causes of flatness deviations, such as mold precision issues during the production process or improper heat treatment process parameters.

[0045] The system dynamically adjusts detection parameters based on flatness status and generates optimized detection schemes, giving the entire detection process excellent flexibility and adaptability. When dealing with brake spring pads of different specifications and structures, unlike traditional detection equipment, it eliminates the need for frequent adjustments to mechanical structures or tooling changes. Instead, it dynamically optimizes the image acquisition parameters (such as light intensity, shooting angle, and resolution) or curve generation algorithm parameters of the machine vision system based on the current flatness status. This enables accurate detection of different types of brake spring pads, significantly reducing equipment adjustment and time costs and improving the versatility of the detection system. Furthermore, as detection work continues, the detection parameters are continuously optimized based on the accumulated flatness status data, gradually improving the rationality and accuracy of the detection scheme, creating a virtuous cycle. This allows the detection system to maintain high detection performance over the long term and better adapt to changes in product specifications or adjustments in detection requirements that may occur during production. Attached Figure Description

[0046] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent detection method for brake spring flatness based on machine vision described in this invention.

[0047] Figure 2 A flowchart for generating a standard flatness curve;

[0048] Figure 3 A flowchart for determining the flatness state;

[0049] Figure 4 A flowchart for configuring the detection target;

[0050] Figure 5 The flowchart for sparse reconstruction processing. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1 This invention provides a machine vision-based intelligent detection method for the flatness of brake spring pads, the method comprising:

[0053] Historical brake spring pad image data is acquired, and a standard flatness curve is generated based on this historical data. The historical brake spring pad image data originates from images of qualified products accumulated during previous inspections, and flatness features are extracted using image processing techniques. The generation of the standard flatness curve involves statistical analysis of the historical data to establish a benchmark reference. Real-time brake spring pad image data is acquired using a machine vision system, which includes a high-resolution camera, lighting equipment, and an image acquisition card to ensure image clarity and consistency. A real-time flatness curve is generated based on the real-time image data, reflecting the actual flatness of the current brake spring pad. A flatness deviation curve is calculated based on the real-time flatness curve and the standard flatness curve. The flatness deviation curve is derived through mathematical operations and is used to quantify the difference between the real-time data and the standard. The matching degree between the flatness deviation curve and a reference defect model is analyzed. The reference defect model is pre-existing in the system and contains feature patterns of common defect types. The matching degree analysis determines the flatness status of the brake spring pad, such as qualified, defective, or requiring further inspection. Inspection parameters, including image acquisition frequency, lighting intensity, and analysis algorithm settings, are dynamically adjusted based on the flatness status to generate an optimized inspection plan aimed at improving inspection accuracy and efficiency.

[0054] Example 1: See Figure 2 In generating the standard flatness curve, the implementation method involves the systematic processing and analysis of historical brake spring pad image data. This historical data originates from a database of qualified product images accumulated over a long production process. Each image undergoes rigorous quality inspection and certification and is accompanied by complete production environment metadata. Image acquisition utilizes fixed-mount high-resolution industrial cameras, coupled with a uniformly diffused LED lighting system to ensure image consistency and comparability. All historical images are stored according to unified specifications and formats, establishing a comprehensive image database management system.

[0055] In the data preprocessing stage, grayscale conversion and noise filtering are performed on the original historical images. Median filtering is used to eliminate random noise while preserving edge information. The image enhancement stage adjusts contrast through histogram equalization, making the surface features of the spring sheet more clearly visible. The preprocessed images are cropped to a uniform size, focusing on the effective detection area of ​​the spring sheet and eliminating background interference. The flatness model is trained based on a convolutional neural network in a deep learning architecture. The network structure employs a multi-level feature extraction design; the lower convolutional layers capture local edge and texture features, while higher convolutional layers integrate global morphological information. Pooling layers progressively reduce the feature map dimension, and fully connected layers ultimately output flatness values. During training, the input data is the preprocessed historical images, and the label data is the actual flatness value of the corresponding spring sheet obtained through precision measurement instruments. The training and validation sets are divided according to time series, with earlier data used for training and more recent data used for validation to simulate the temporal characteristics of actual detection.

[0056] Model training is performed on a GPU cluster, employing backpropagation to optimize network parameters. The loss function calculates the mean squared error between the predicted flatness values ​​and the true values, and the optimizer minimizes the loss function through an adaptive learning rate adjustment strategy. The number of training iterations is determined based on the error convergence on the validation set to avoid overfitting. The trained flatness model has the ability to directly regress the flatness curve from the input image. This curve consists of a series of discrete points, each corresponding to a flatness measurement at a specific location on the spring sheet.

[0057] Repeated model training under different detection conditions is a crucial step in generating a standard flatness curve sequence. Variations in detection conditions mainly arise from three aspects: changes in illumination intensity, camera shooting angle shifts, and fluctuations in the spring plate's placement position. For illumination conditions, a gradient illumination database ranging from 500 lux to 1500 lux was established, and independent flatness models were trained for each illumination level. For camera angle variations, a dataset of tilted images within a range of ±5 degrees was collected, and dedicated models for the corresponding angles were trained. Spring plate position fluctuations were simulated using image data from different clamping states for model training.

[0058] Each training iteration generates a flatness model under specific conditions, and each model produces a corresponding standard flatness curve. These curves together form a standard flatness curve sequence, with each curve in the sequence annotated with detailed conditional parameter information. The curve sequence is stored using a hierarchical database structure, indexed and arranged according to conditional parameters, supporting fast retrieval and matching queries. The generation process of the standard flatness curves also includes post-processing optimization steps, smoothing the original curves output by the model, using a moving average algorithm to eliminate random fluctuations and preserve true trend characteristics. Standardization of the curve data converts the values ​​to a unified dimensional range, facilitating subsequent comparative analysis. The final generated standard flatness curve sequence not only contains the curve numerical data but also stores the corresponding confidence index, reflecting the reliability of the curve under specific conditions.

[0059] In practical applications, the standard flatness curve sequence needs to be updated and maintained regularly. Image data of newly qualified products are continuously added to the historical database, triggering the incremental training process of the model. Incremental training adopts transfer learning technology to fine-tune the original model parameters, maintaining model stability while adapting to the slow changes in production conditions. The model performance monitoring system continuously evaluates the accuracy of the standard flatness curve, and initiates a complete retraining process when detection conditions change significantly. The entire implementation relies on the support of a distributed computing architecture. Image storage uses a distributed file system, model training tasks are distributed to multiple computing nodes for parallel execution, and the database system adopts a highly available cluster configuration. At the software implementation level, a dedicated model training management platform is developed, integrating functional modules such as data preprocessing, model training, curve generation, and sequence management, providing a visual operation interface and automated process control.

[0060] The practical application of the standard flatness curve sequence demonstrates its integration with the real-time detection process. At the start of real-time detection, the system first identifies the current detection environment parameters, including measured illumination intensity, camera attitude data, and spring plate positioning information. These parameters serve as query conditions, retrieving the most matching standard curve from the standard flatness curve sequence. The matching algorithm calculates the similarity between the current parameters and the condition parameters of each curve in the sequence, selecting the standard curve with the highest similarity as the benchmark reference for this detection.

[0061] Example 2: See Figure 3In the process of calculating the flatness deviation curve, the system first acquires the real-time flatness curve and the standard flatness curve. The real-time flatness curve is generated from the image processing of the currently inspected brake spring pad, through feature extraction of high-resolution images acquired by the machine vision system. The standard flatness curve is generated from a flatness model trained on historical data, representing the flatness characteristics under ideal conditions. Both curves use the same data structure and coordinate system, storing a series of flatness value points in array form, with each value point corresponding to the measurement result at a specific location on the spring pad surface.

[0062] When calculating the flatness deviation curve, the system performs point-by-point subtraction. The algorithm iterates through each data point on the real-time flatness curve, subtracting its value from the corresponding data point value on the standard flatness curve. This calculation process ensures an accurate correspondence of deviation values, with each deviation value reflecting the flatness difference at a specific location. The calculated deviation values ​​form a new data sequence, namely the flatness deviation curve. This curve is stored in the same coordinate system as the original curve for easy subsequent analysis and visualization. The flatness change curve is generated based on multiplication. The system multiplies each deviation value on the flatness deviation curve with the flatness value of the corresponding point on the real-time flatness curve. This calculation enhances sensitivity to areas of significant deviation, as deviations in areas with larger flatness values ​​are amplified, while deviations in areas with smaller flatness values ​​are relatively weakened. The calculation results form the flatness change curve, which provides another perspective on deviation representation and helps identify flatness changes with practical impact.

[0063] When determining the flatness state of brake spring pads, the system accesses a pre-stored defect library containing mathematical models of various typical defects, such as warping, denting, and twisting. Each reference defect model is stored as a data curve, representing the characteristic behavior of that type of defect in flatness inspection. The system establishes a time window mechanism, setting an appropriate time range based on the inspection cycle, within which the flatness deviation curves are analyzed holistically. Matching degree calculation employs multiple mathematical metrics. The system calculates the similarity between the flatness deviation curve and each reference defect model, using algorithms such as correlation coefficient, Euclidean distance, or dynamic time warping for quantitative comparison. These calculations generate a set of matching degree values, each corresponding to the degree of matching between the flatness deviation curve and a specific reference defect model. The matching degree values ​​are arranged in ascending order, with the minimum value representing the highest similarity, meaning the flatness deviation curve is closest to that defect pattern.

[0064] After identifying the reference defect model corresponding to the minimum matching degree, the system performs a curve overlay operation, mathematically fusing the selected reference defect model with the standard flatness curve using methods such as weighted averaging or curve addition. The overlay process considers the strength coefficient and directional characteristics of the defect model, ensuring that the generated flatness prediction result retains both the baseline characteristics of the standard curve and incorporates the typical manifestations of the defect pattern. This prediction result forms an estimate of future flatness changes in the brake spring pad, providing a reference for quality assessment.

[0065] When a brake spring pad is detected, the system acquires an image in real time and generates a real-time flatness curve. It is assumed that the standard flatness curve shows a uniform distribution of flatness values ​​under ideal conditions with minimal fluctuations. The real-time curve may exhibit significant fluctuations in certain areas; these fluctuations are converted into peak values ​​on the deviation curve through subtraction. Areas with larger deviation values ​​will produce more significant changes after multiplication, highlighting the importance of these critical areas. The system retrieves multiple reference models from the defect library, such as a slight warping model and a local dent model. Calculations show that the current deviation curve has the smallest matching degree with the local dent model, indicating that the deviation pattern most closely resembles the characteristics of a dent defect. The system overlays this dent model with the standard flatness curve to generate a predicted result curve. This result curve shows a potential trend of decreasing flatness in specific areas, corroborating the measured deviation.

[0066] The entire processing is implemented on an embedded computing platform, employing a multi-threaded parallel computing architecture. The deviation calculation thread handles numerical computation, the matching degree calculation thread processes model comparisons, and the overlay generation thread performs curve fusion operations. Data exchange is conducted through a shared memory area to ensure processing efficiency and real-time performance. The system includes an exception handling mechanism; when an outlier occurs during calculation, a review process is initiated to prevent incorrect judgments. Regarding database management, a pre-stored defect database is regularly updated and maintained. New defect types are converted into mathematical models and added to the database after expert verification, while outdated defect models are adjusted and optimized based on actual detection results. The storage of reference defect models uses a hierarchical structure, categorized and indexed according to defect severity and frequency of occurrence to improve matching and retrieval efficiency.

[0067] The visualization subsystem provides a graphical interface to display the entire processing. Operators can observe the generation process of the flatness deviation curve in real time, view the matching degree values ​​with each reference model, and the final superimposed prediction results. The interface design highlights key information and uses color coding to distinguish deviation areas of different severity levels, assisting in manual judgment and decision-making. The quality control process includes a verification mechanism for the calculation results. The system records the entire process data of each deviation calculation and matching degree analysis, including intermediate calculation results and timestamp information. This log data is used for subsequent traceability analysis and can provide complete calculation basis when disputes arise. The frequency and distribution characteristics of different defect types are regularly statistically analyzed to provide data reference for production process improvement.

[0068] In terms of system integration, the flatness status judgment results are transmitted to the production line control system via an industrial bus interface. Based on the judgment results, the system can trigger the sorting device to separate defective products or adjust processing parameters to compensate for subsequent products. The entire detection and judgment process is synchronized with the production line rhythm, achieving online real-time quality monitoring. Flatness deviation calculation and status judgment form a complete analytical closed loop. From raw data acquisition to final result output, each link has been carefully designed and optimized to ensure the accuracy and reliability of the judgment results.

[0069] Example 3: See Figure 4 In the process of configuring and implementing inspection targets, the system establishes a multi-level target system based on the quality standards of brake spring pads. These quality standards are derived from industry technical specifications, product design requirements, and customer acceptance standards, forming quantified technical indicator documents. The system parses these documents, extracts key quality characteristic parameters, and maps them to the inspection target settings. The primary inspection target focuses on minimizing flatness error, requiring the system to prioritize measurement accuracy and control flatness deviation within allowable limits. Auxiliary inspection targets focus on optimizing inspection efficiency, involving a balance between processing speed, resource consumption, and response time. The maintenance target emphasizes maintaining illumination stability, including elements such as consistent illumination intensity, color temperature stability, and uniformity. Target classification and identification adopt a structured coding system, assigning a unique identifier to each inspection target in the format OBJ-Type-Priority, where OBJ represents the target object, Type distinguishes the target type, and Priority defines the priority value. The primary inspection target identifier includes the MAX accuracy level, auxiliary targets include the OPT optimization type, and maintenance targets include the STB stability category. Priority values ​​range from 0 to 100, with the primary detection target set to the highest priority of 100. Auxiliary targets are set to values ​​between 50 and 80 based on specific needs, while the primary target is kept at a fixed priority of 70. This classification and labeling method facilitates rapid system identification and target management.

[0070] The selection process for key detection parameters based on classification identifiers employs a weighted allocation algorithm. The system calculates the contribution of each detection parameter to each type of target, establishing a parameter-target correlation matrix. Rows in the correlation matrix represent detection parameters, columns represent detection targets, and matrix element values ​​represent the degree of influence of that parameter on the target. The degree of influence is obtained through historical data analysis, using normalized values ​​in the range of 0-1. The selection of key detection parameters follows a weighted scoring rule.

[0071]

[0072] in: Indicates the first The weighted score of each detection parameter, Indicates the first The weight coefficient of each detection target, Indicates the first The parameter for the first The impact index of each objective This represents the total number of detected targets. Weighting coefficient. The influence index is derived from the conversion of target priorities. The system uses a parameter-target correlation matrix. The top k parameters with the highest weighted scores are selected as key detection parameters, with the value of k determined based on system configuration. The detection parameter system comprises three main categories: image acquisition parameters, processing parameters, and analysis parameters. Image acquisition parameters include camera exposure time, gain value, resolution settings, and frame rate control; processing parameters cover filter intensity, edge detection threshold, and feature extraction accuracy; and analysis parameters include flatness calculation algorithm selection, tolerance range settings, and judgment rule definitions. Each parameter has an adjustable range and a default value, and the system dynamically adjusts these parameter values ​​based on the target configuration.

[0073] The target configuration interface provides a visual operating environment, allowing users to view current target settings, adjust priority values, and modify parameter relationships through a graphical interface. The system displays the impact of configuration changes on the selection of key parameters in real time and provides a simulated preview of the configuration effect. All configuration changes are recorded in the version management system, supporting configuration rollback and historical traceability. During real-time detection, the system dynamically adjusts detection parameters based on the currently configured target system. When a flatness error is detected approaching the allowable limit, the system automatically increases the priority of the primary detection target, increases image acquisition resolution, and reduces processing speed to achieve higher accuracy. When the production line pace accelerates, the system appropriately increases the weight of auxiliary targets and optimizes algorithm parameters to improve processing efficiency. When lighting conditions change, the system activates a target protection mechanism, adjusting camera parameters to compensate for changes in lighting.

[0074] The parameter adjustment mechanism employs a closed-loop control principle. The system monitors detection results and target achievement in real time, compares them with preset target values, and calculates parameter adjustments based on deviations. The adjustment calculation uses an incremental PID algorithm to avoid system fluctuations caused by sudden parameter changes. After each parameter adjustment, the system evaluates the improvement in target achievement as a reference for subsequent adjustments. The target conflict resolution mechanism handles the interrelationships between multiple targets. When a primary detection target conflicts with an auxiliary target, the system arbitrates based on priority settings, prioritizing the basic requirements of the higher-priority target. For conflicts between targets of equal priority, the system uses a Pareto optimization strategy to find the optimal parameter settings that simultaneously accommodate multiple targets. The conflict resolution process records a detailed decision log, including the conflict type, arbitration rules, and final selection.

[0075] The system periodically evaluates the effectiveness of target configurations, collects testing data and quality reports over a period of time, and analyzes the actual achievement of each target. It calculates indicators such as target achievement rate, parameter adjustment frequency, and number of conflicts to assess the rationality of the current configuration. Based on the evaluation results, the system automatically adjusts target weights or suggests configuration modifications to the user. Configuration export and import functions support standardized applications. Validated target configuration schemes can be exported as standard format files for rapid deployment in similar testing systems. During import, the system automatically verifies configuration compatibility, adjusts parameter ranges to adapt to hardware differences, and ensures reliable configuration migration. Through clear target classification and identification, a systematic parameter selection mechanism is established to optimize testing performance. A dynamic adjustment mechanism ensures the system can adapt to changes in the production environment and maintain testing effectiveness under various constraints.

[0076] Example 4: The system first identifies a set of key detection parameters, which are derived from important parameters selected during the target configuration phase, including adjustable variables in image acquisition, processing, and analysis. The system establishes a parameter impact assessment model, based on historical operational data and theoretical analysis, to quantify the degree of influence of each parameter on the primary and secondary detection targets. The parameter impact analysis employs a multi-dimensional assessment method. For minimizing flatness error in the primary detection target, the system analyzes the trend of each parameter change on measurement accuracy. For optimizing detection efficiency in the secondary detection target, the system assesses the impact characteristics of parameter adjustments on processing speed and resource consumption. The impact level is divided into three levels: high, medium, and low, each corresponding to different numerical weights. High impact indicates that a small change in the parameter will cause a significant change in the target indicator; medium impact indicates that the change in the parameter is roughly proportional to the change in the target indicator; and low impact indicates that a large adjustment of the parameter is required to cause a visible change in the target indicator.

[0077] The system generates parameter optimization constraints based on quality standards and equipment performance limitations. Constraint types include numerical range constraints, interrelationship constraints, and dynamic adjustment constraints. Numerical range constraints define the upper and lower limits of allowable adjustment for each parameter, preventing exceedances of equipment capabilities. Interrelationship constraints handle the coupling relationships between parameters, avoiding conflicting adjustment strategies. Dynamic adjustment constraints flexibly change based on real-time monitoring status, adapting to different production conditions.

[0078] The impact factor calculation employs a weighted statistical method. The system collects historical parameter adjustment records and target indicator change data to calculate the average impact strength of each parameter on each target. Impact factor values ​​are normalized to between 0 and 1; the closer the value is to 1, the greater the parameter's impact on the target. The calculation process considers time decay, assigning higher weights to recent data to ensure the impact factor reflects the actual characteristics of the current system. The parameter optimization process uses a multi-objective optimization algorithm. The system simultaneously considers the primary detection target and auxiliary detection targets, seeking the optimal parameter combination that balances multiple targets. The optimization algorithm is based on the non-dominated sorting principle, generating a Pareto front solution set for parameter optimization. From the solution set, the parameter configuration that best meets the current requirements is selected, prioritizing the basic requirements of the primary detection target, while optimizing the performance of auxiliary targets under this premise.

[0079] The real-time adjustment mechanism plays a crucial role in the optimization process. The system monitors changes in the external environment and fluctuations in production rhythm, promptly triggering parameter re-optimization. When a change in lighting conditions or a product type switch is detected, the system automatically initiates a rapid optimization process, finding parameter settings adapted to the new operating conditions in a short time. See Table 1 for an analysis of the impact of key detection parameters.

[0080] Table 1: Impact analysis of key detection parameters.

[0081] Parameter name Impact index on the main target Impact index on auxiliary targets Impact Factor Constraint range Exposure time 0.85 0.45 0.72 [100,500]μs Image resolution 0.92 0.78 0.88 [1024,2048] pixels Filter strength 0.63 0.35 0.54 [1,5] Level Feature extraction accuracy 0.88 0.82 0.86 [0.1, 0.9] Threshold Analyze algorithm complexity 0.75 0.91 0.82 [1,3] Level

[0082] The impact index indicates the direct influence of a parameter on a single objective, while the impact factor comprehensively reflects the overall importance of the parameter's influence on multiple objectives. The constraint range specifies the technical boundaries by which each parameter can be adjusted.

[0083] The system implements dynamic management of parameter optimization constraints. These constraints are not fixed but updated in real-time based on equipment status and environmental factors. When camera performance degradation is detected, the system automatically adjusts the exposure time constraint range to avoid using unreliable parameter intervals. When processing resources are strained, the constraint range for the analysis algorithm's complexity is correspondingly reduced to ensure stable system operation. A periodic update mechanism for influencing factors maintains the accuracy of the analysis. The system recalculates the influencing factors of all parameters weekly, incorporating the latest operational data. The update process uses a sliding time window approach, prioritizing recent data and gradually discarding outdated data. The updated influencing factors are used to optimize the weight settings for parameter optimization, making the optimization results more aligned with actual needs.

[0084] The parameter optimization process employs a hierarchical optimization strategy. The first layer quickly determines the approximate range of parameters and prioritizes adjusting parameters with high importance based on the ranking of influence factors. The second layer performs a fine-grained search within the defined range to find the optimal parameter combination. This hierarchical strategy improves optimization efficiency and reduces computational resource consumption. An anomaly handling mechanism plays a crucial role in the impact analysis process. When parameter adjustments are detected to cause system performance anomalies, the optimization process is immediately halted, and the system reverts to safe parameter settings. The system records the anomaly, analyzes its causes, updates the influence factor calculation model, and prevents similar situations from recurring. A visual monitoring interface provides a comprehensive visualization of the entire impact analysis process. Operators can observe the real-time impact trends of parameter adjustments on target indicators, view influence factor calculation results, and monitor the progress of the optimization process. The interface offers graphical editing functions for parameter constraints, supporting manual intervention and adjustments.

[0085] In terms of system integration, the impact analysis module works closely with the target configuration module. When the target configuration changes, the impact analysis update process is automatically triggered, recalculating the degree of parameter impact and impact factors. Data exchange with the real-time detection module ensures that the impact analysis is based on the latest detection results, maintaining the timeliness and accuracy of the analysis. An effectiveness evaluation mechanism regularly checks the implementation effect of the impact analysis. The system compares the improvement of target indicators before and after parameter optimization, analyzes the prediction accuracy of impact factors, and evaluates the rationality of constraints. The evaluation results are used to improve the impact analysis model and optimization algorithm, forming a closed loop of continuous improvement. Systematic impact degree assessment provides a scientific basis for parameter adjustment, reasonable optimization constraints ensure the safety and reliability of the adjustment process, accurate impact factor calculation guides the optimization direction, and multi-objective optimization algorithms achieve balance and coordination between different objectives. This implementation method reflects the self-optimization and adaptive capabilities of the intelligent detection system, enabling the system to maintain excellent detection performance in complex and ever-changing industrial environments.

[0086] Example 5: See Figure 5During the sparse reconstruction process, the system first receives real-time brake spring pad image data from the image acquisition module. This data is input in the form of a two-dimensional pixel matrix, containing grayscale values ​​or RGB color information. The system initiates a data conversion process to reconstruct the two-dimensional image data into a multidimensional dataset. This conversion process is achieved through a data reshaping algorithm, which combines the coordinate information, color channel values, and acquisition timestamp of each pixel into a multidimensional vector. These vectors are organized into a three-dimensional data cube according to their spatial relationships, where two dimensions represent the image plane coordinates, and the third dimension represents the feature depth. The feature depth dimension includes derived data such as pixel intensity, gradient magnitude, and neighborhood features. The data cube is stored using a tensor structure, supporting efficient multidimensional data operations and access.

[0087] When performing feature extraction on a multidimensional dataset, the system employs a hierarchical feature learning strategy. The initial feature extraction stage uses edge detection operators and texture analysis algorithms to extract basic features from the data cube. These features include geometric attributes such as contour orientation, curvature variation, and surface roughness. The intermediate feature extraction stage combines the initial features into a higher-level feature representation through feature fusion techniques. This stage uses a sparse coding algorithm to learn latent feature patterns in the data, generating compact feature representations. The advanced feature extraction stage applies a feature selection mechanism to filter out the most discriminative subset of features from a large set. The entire feature extraction process produces two types of output data: target data for detection and anomaly indication data for indicating anomalies.

[0088] The target data contains feature information directly related to flatness detection. These features are optimized and selected to accurately characterize the flatness properties of the brake spring pads. The target data is organized as feature vectors, with each vector corresponding to a comprehensive feature description of a detection area. These vectors are passed to the detection algorithm module for flatness calculation and quality assessment. Anomaly indication data captures feature information that deviates from normal patterns, potentially indicating potential defects or anomalies. Anomaly indication data includes feature anomaly scores, deviation degree indices, and anomaly pattern identifiers. The system sets dynamic thresholds for the anomaly indication data, adjusting the anomaly judgment criteria based on historical data statistical characteristics.

[0089] The process of dynamically adjusting and optimizing the detection scheme based on anomaly indication data employs an adaptive control mechanism. The system monitors the changing trends of the anomaly indication data in real time, and initiates the scheme adjustment process when an anomaly pattern is detected. The adjustment strategy is differentiated according to the anomaly type and severity. For minor local anomalies, the system increases the sampling density and detection frequency in that area to improve monitoring accuracy. For globally significant anomalies, the system adjusts the feature extraction parameters to enhance sensitivity to anomaly features. Scheme adjustment involves multiple aspects, including modification of image acquisition parameters, optimization of feature extraction algorithm parameters, and adjustment of detection judgment thresholds.

[0090] In terms of data storage and management, the system establishes a dedicated data warehouse to store various types of data generated during the sparse reconstruction process. Raw image data is stored in a compressed format to reduce storage space usage. Feature data is saved in a structured format for easy retrieval and analysis. Anomaly indication data is stored in association with detection results, forming a complete data traceability chain. The data management system provides efficient data query and access interfaces, supporting real-time processing and historical data analysis. Regarding process optimization, the system implements a parallel computing architecture to accelerate the sparse reconstruction process. Data transformation and feature extraction tasks are distributed to multiple computing units for parallel execution, improving processing efficiency. A dynamic load balancing mechanism automatically adjusts the allocation of computing resources based on the amount of data processed, ensuring timely system response. Memory management employs a caching optimization strategy, retaining frequently used data and intermediate results in a high-speed cache to reduce data access latency.

[0091] The quality control process includes a verification mechanism for the sparse reconstruction results. The system periodically checks the accuracy and completeness of feature extraction and compares the consistency of data processing results across different periods. The reliability of anomaly indication data is verified by comparing it with actual detection results, and the anomaly detection algorithm is continuously optimized. Processing parameters and algorithm settings are recorded in the system log, supporting problem tracing and performance analysis.

[0092] In terms of system integration, the sparse reconstruction processing module is deeply integrated with the entire detection platform, and the data interface adopts a standardized protocol to ensure smooth data exchange with other modules. Processing results are transmitted to the detection and judgment module in real time, providing feature data support for flatness state analysis. Simultaneously, feedback information from other modules is received to optimize processing parameters and adjust strategies. Multidimensional data transformation provides rich data representations, hierarchical feature learning captures feature information at different levels, and anomaly indication data provides a basis for adjusting the detection scheme. The entire processing demonstrates the adaptive capability and refined processing characteristics of the intelligent detection system, providing reliable technical support for brake spring flatness detection.

[0093] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine vision-based intelligent detection method for the flatness of brake spring sheets, characterized in that, Includes the following steps: Acquire historical brake spring pad image data, and generate a standard flatness curve based on the historical brake spring pad image data; Real-time brake spring pad image data is acquired through a machine vision system, and a real-time flatness curve is generated based on the real-time brake spring pad image data. Calculate the flatness deviation curve based on the real-time flatness curve and the standard flatness curve; The flatness deviation curve is analyzed to determine the matching degree with the reference defect model, and the flatness status of the brake spring plate is determined. Based on the flatness status, the detection parameters are dynamically adjusted to generate an optimized detection scheme; The step of determining the flatness state of the brake spring pads includes: Obtain a reference defect model from the pre-stored defect library, and calculate the matching degree between the flatness deviation curve and the reference defect model within the time window; Identify the reference defect model corresponding to the minimum matching degree, and superimpose the reference defect model with the standard flatness curve to generate a flatness prediction result; It also includes the step of configuring the detection target: Based on the quality standards of brake spring pads, set the main inspection target, auxiliary inspection target, and maintenance target; The main detection objective includes minimizing flatness error; The auxiliary detection objectives include optimizing detection efficiency; The stated objectives include maintaining light stability; The steps for configuring the detection target include: Based on the aforementioned quality standards, the primary detection target, auxiliary detection target, and maintenance target are classified and identified. Key detection parameters are selected using the classification identifiers; The selection of key detection parameters is based on the weighted scoring rules: in: Indicates the first The weighted score of each detection parameter, Indicates the first The weight coefficient of each detection target, Indicates the first The parameter for the first The impact index of each objective Indicates the total number of targets detected; It also includes the step of performing an impact analysis: Analyze the degree of influence of the key detection parameters on the main detection target and the auxiliary detection target; Parameter optimization constraints are generated based on the degree of influence. The steps for performing the impact analysis include: Calculate the influence factors of the key detection parameters; Based on the parameter optimization constraints and the influencing factors, parameter optimization is performed to obtain optimized detection parameters; It also includes the step of sparse reconstruction processing of the real-time brake spring image data: The real-time brake spring pad image data is converted into a multidimensional dataset; Feature extraction processing is performed on the multidimensional dataset to obtain target data for detection and anomaly indication data for indicating anomalies; The optimized detection scheme is dynamically adjusted based on the anomaly indication data.

2. The intelligent detection method for brake spring flatness based on machine vision according to claim 1, characterized in that, The step of generating a standard flatness curve based on the historical brake spring pad image data includes: The historical brake spring pad image data is used to train a flatness model, which is used to extract features and generate a standard flatness curve. Repeat the steps of training the flatness model under different detection conditions to obtain a standard flatness curve sequence.

3. The intelligent detection method for brake spring flatness based on machine vision according to claim 2, characterized in that, The step of calculating the flatness deviation curve based on the real-time flatness curve and the standard flatness curve includes: Subtracting the flatness value of the standard flatness curve from the flatness value on the real-time flatness curve yields the flatness deviation curve; Multiply the deviation value on the flatness deviation curve with the flatness value on the real-time flatness curve to obtain the flatness change curve.

4. A machine vision-based intelligent detection system for the flatness of brake spring pads, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent detection method for brake spring flatness based on machine vision as described in any one of claims 1 to 3.

Images