A high-precision size detection method and system for a motorcycle engine cylinder

Abstract

This application relates to the field of intelligent inspection technology, specifically a high-precision dimensional inspection method and system for motorcycle engine cylinder blocks. By constructing a comprehensive risk index, it achieves precise quantification of production line status. High-precision full inspection is triggered only when the risk index exceeds the limit, while a low proportion of routine sampling is maintained during steady-state production. This concentrates expensive coordinate measuring machine / cylindricity measuring instrument resources on high-risk periods. By reverse-engineering production data feature templates for various defect types based on high-precision inspection results, the system can not only identify anomalies but also accurately identify the specific defect type associated with the anomaly through template matching. This application can establish a reference baseline and activate a general sentinel using normal batches without relying on large amounts of historical data. The template construction mechanism allows the system to automatically rebuild new templates as new batch data accumulates after process changes, without manual intervention or the need to re-accumulate months of data, demonstrating excellent adaptability to process changes.

Description

Technical Field

[0001] This application relates to the field of intelligent detection technology, specifically a high-precision detection method and system for the dimensions of a motorcycle engine cylinder block. Background Technology

[0002] Currently, the dimensional inspection of motorcycle engine blocks after precision machining mainly faces the following technical challenges:

[0003] (1) Fixed-ratio sampling leads to resource misallocation: Regardless of whether the production line is stable or not, sampling is carried out at a constant ratio. When the production line is stable, expensive cylindricity measuring instruments / coordinate measuring machines are wasted; when the production line fluctuates, insufficient sampling leads to batch dimensional deviations (roundness, taper) escaping to the assembly line, causing batch rework or scrapping of the entire machine.

[0004] (2) Traditional SPC threshold warning lacks physical interpretability: The statistical process control (SPC) method based on the global mean and 3σ standard deviation can only issue an "over-limit" signal and cannot distinguish the specific physical cause of the defect (such as being unable to distinguish between "honing strip wear" and "excessive hardness of blank"). This method has an extremely high false alarm rate and cannot provide accurate fault diagnosis directions for on-site personnel after the alarm is triggered.

[0005] (3) Machine learning solutions based on historical defect clustering face a cold start dilemma: Existing solutions rely on a large amount of historical high-defect batch data for cluster modeling, but such data is severely lacking when new production lines or new models are put into production. At the same time, when the process undergoes permanent changes (such as changing the material of the cutting tools), the old template library becomes invalid, and several months of data need to be accumulated again, which is not very practical. Summary of the Invention

[0006] In view of this, the purpose of this application is to provide a high-precision detection method and system for the dimensions of a motorcycle engine block, so as to solve the problems in the background art.

[0007] To achieve the above objectives, this application adopts the following technical solution:

[0008] This application discloses a high-precision method for measuring the dimensions of a motorcycle engine block, comprising the following steps:

[0009] S1, acquire historical production process data and test results of multiple historical batches of the motorcycle engine cylinder block production line within the most recent time window, and acquire real-time production data of the motorcycle engine cylinder block production line.

[0010] S2, based on the detection results, extract defective batches and normal batches from multiple historical batches, extract reference baselines for various production parameters from the historical production process data of the normal batches; extract correlation parameters and weights of correlation parameters that are associated with the increase in defect rate from the historical production process data of the defective batches.

[0011] S3, Calculate the comprehensive risk index based on the production data, the reference baseline of the related parameters, and the weights of the related parameters;

[0012] S4. When the comprehensive risk index is greater than or equal to the preset risk threshold, high-precision sampling inspection of the cylinder size of the real-time production batch is carried out based on the preset sampling ratio to obtain high-precision inspection results.

[0013] S5, when the high-precision detection result indicates the presence of defects, construct production data feature templates for various defect types based on the high-precision detection results and production data of multiple real-time production batches; when the high-precision detection result indicates the absence of defects, adjust the preset sampling ratio and / or risk threshold accordingly.

[0014] S6. When constructing production data feature templates for multiple defect types, the real-time collected production data is matched with the production data feature templates for multiple defect types. When the production data matches the production data feature template for any defect type, the cylinder is sampled and the corresponding defect type's dimensions are inspected with high precision. Otherwise, steps S3-S4 are executed.

[0015] In one embodiment of this application, reference baselines for various production parameters are extracted from the historical production process data of the normal batch, including:

[0016] The original value sequences of various production parameters are extracted from the historical production process data of the normal batch, and the original value sequences of the various production parameters are cleaned to obtain the value sequences. The data cleaning includes steady-state time point data extraction and physical boundary filtering.

[0017] Calculate the mean and standard deviation of the value series of various production parameters to obtain the reference baseline for various production parameters.

[0018] In one embodiment of this application, the extraction of correlation parameters and weights of these correlation parameters from historical production process data of defective batches includes:

[0019] Extract the target data segment corresponding to the production and processing time period of the defective cylinder block from the historical production process data of the defective batch;

[0020] Extract the time features and ambient temperature features of the target data segment, and extract a reference data segment that is consistent with the time features and ambient temperature features from the historical production process data of the normal batch.

[0021] The Mann-Whitney rank-sum test is performed on the value sequence of each parameter in the target data segment and the value sequence of the corresponding parameter in the reference data segment to obtain the significance level; and the parameters with significance levels lower than the preset significance threshold are used as the initial association parameters.

[0022] Extract the target fragment sequence of initial association parameters from the target data fragment, and extract the reference fragment sequence of initial association parameters from the reference data fragment;

[0023] The mutual information between the initial association parameters and the defect-generating event is calculated based on the target fragment sequence and the reference fragment sequence using the initial association parameters.

[0024] Collinearity analysis was performed on multiple initial correlation parameters to obtain multiple sets of initial correlation parameters that are collinear; and the initial correlation parameter with the largest mutual information in each initial correlation parameter set was taken as the correlation parameter.

[0025] The mutual information of multiple correlation parameters is normalized to obtain the weight of each correlation parameter.

[0026] In one embodiment of this application, collinearity analysis is performed on multiple initial correlation parameters to obtain multiple sets of initial correlation parameters that have collinear relationships, including:

[0027] Calculate the Pearson correlation coefficient for any two initial correlation parameters based on the target fragment sequence and the reference fragment sequence;

[0028] Two initial correlation parameters with a Pearson correlation coefficient greater than or equal to a preset correlation threshold are combined into an initial correlation parameter group that has a collinear relationship.

[0029] In one embodiment of this application, a comprehensive risk index is calculated based on the production data, the reference baseline of the correlation parameters, and the weights of the correlation parameters, including:

[0030] The single-parameter deviation score of the correlation parameter is calculated based on the production data and the reference baseline of the correlation parameter. The single-parameter deviation score The formula for calculation is:

[0031]

[0032] In the formula, For parameters associated with production data The current value, The mean is used as a reference baseline. The standard deviation is the reference baseline.

[0033] Single-parameter bias score of the correlation parameter based on the weight of the correlation parameter. We perform a weighted summation to obtain the overall risk.

[0034] In one embodiment of this application, production data feature templates for various defect types are constructed based on high-precision detection results and production data from multiple real-time production batches, including:

[0035] The defect type is extracted from the high-precision detection results, and the target production data segment corresponding to the defect type cylinder is extracted from the production data, as well as multiple reference production data segments of the normal cylinder that is closest to the defect type cylinder from the production data.

[0036] The target production data segment is standardized by performing a series of parameter values ​​to obtain the target vector. The reference production data segment is then standardized to obtain a reference vector. ;

[0037] For multiple reference vectors The average reference vector is obtained by averaging. ; and calculate the target vector With the average reference vector The difference is used to obtain the template feature vector. ;

[0038] In the template feature vector When there are no abnormal events during the extraction time, the template feature vector will be... Stored in a pre-constructed set of template feature vectors for the defect type;

[0039] When the number of template feature vectors in the template feature vector set of any defect type is greater than or equal to the preset update trigger threshold, unsupervised clustering is performed on the template feature vectors in the template feature vector set of the defect type to obtain multiple feature vector clusters;

[0040] The feature vector clusters with a number of samples within the cluster greater than or equal to a preset number threshold are considered as valid clusters, and the Euclidean standard deviation of all template feature vectors in the valid clusters is calculated.

[0041] When the Euclidean standard deviation is greater than the preset standard deviation threshold, the unsupervised clustering parameters are adjusted under the adjustment range constraint, and unsupervised clustering is performed on the template feature vectors in the template feature vector set of the defect type until there is an effective cluster with an Euclidean standard deviation less than or equal to the preset standard deviation threshold or the number of iterations exceeds the set value; when the Euclidean standard deviation is less than or equal to the preset standard deviation threshold, the center vector of the effective cluster is extracted to obtain the production data feature template of the defect type;

[0042] When at least one production data feature template exists for each defect type, the construction of production data feature templates for multiple defect types is completed.

[0043] In one embodiment of this application, feedback adjustment of the preset sampling ratio and / or risk threshold includes:

[0044] Calculate the false alarm rate, false negative rate, and template matching hit rate within the most recent time window; and obtain the high-precision detection load rate at the current moment.

[0045] When the false negative rate exceeds a preset percentage threshold, the risk threshold and the preset sampling ratio are updated to obtain the updated sampling ratio. and the updated risk threshold Among them, the updated sampling ratio and the updated risk threshold The formula for calculation is:

[0046]

[0047]

[0048] In the formula, The preset sampling ratio, To update the unit value proportionally, For the maximum sampling proportion, The risk threshold is... The first adjustment coefficient is a value less than 1;

[0049] When the false negative rate is less than or equal to a preset percentage threshold, the preset sampling ratio is adjusted based on the false positive rate and the high-precision detection load rate.

[0050] After constructing production data feature templates for multiple defect types, and when the template matching hit rate is higher than a preset hit rate threshold for a continuous time period, the preset sampling ratio is updated to obtain the updated sampling ratio. Among them, the updated sampling ratio The formula for calculation is:

[0051]

[0052] In the formula, The second adjustment coefficient is a value less than 1. This is the set safety lower limit.

[0053] In one embodiment of this application, when the false negative rate is less than or equal to a preset percentage threshold, the preset sampling ratio is adjusted based on the false positive rate and the high-precision detection load rate, including:

[0054] When the false alarm rate is greater than a first percentage and the high-precision detection load rate is less than a second percentage, the preset sampling ratio is reduced based on a preset first proportional coefficient.

[0055] When the false alarm rate is less than the third percentage and the high-precision detection load rate is greater than the fourth percentage, the preset sampling ratio is reduced based on the preset second proportional coefficient.

[0056] When the false alarm rate is greater than a first percentage and the high-precision detection load rate is greater than a fourth percentage, the preset sampling ratio is updated, and the updated sampling ratio is... The formula for calculation is:

[0057]

[0058] In the formula, For proportional update coefficients, This is the minimum sampling proportion;

[0059] The percentages are arranged in descending order as follows: fourth percentage, second percentage, first percentage, and third percentage.

[0060] In one embodiment of this application, real-time collected production data is matched with production data feature templates for multiple defect types, including:

[0061] Extract a data segment of a single cylinder block from the production data, and standardize the value sequence of each associated parameter in the data segment to obtain the current feature vector;

[0062] Calculate the difference between the current feature vector and the pre-maintained baseline feature vector to obtain the feature to be matched;

[0063] Calculate the cosine similarity and normalized distance between the feature to be matched and production data feature templates of various defect types. If the cosine similarity between any production data feature template and the feature to be matched is greater than or equal to a preset similarity threshold and the normalized distance is less than or equal to a preset distance threshold, it is determined that the production data matches the production data feature template and there is a corresponding type of defect risk. Otherwise, it is determined that there is no defect risk.

[0064] This application also provides a high-precision dimensional detection system for motorcycle engine blocks, including:

[0065] The acquisition module is used to acquire historical production process data and test results of multiple historical batches of the motorcycle engine cylinder block production line within the most recent time window, and to acquire real-time production data of the motorcycle engine cylinder block production line.

[0066] The feature extraction module is used to extract defective batches and normal batches from multiple historical batches based on the detection results, extract reference baselines for various production parameters from the historical production process data of the normal batches, and extract correlation parameters and weights of the correlation parameters that are associated with the increase in defect rate from the historical production process data of the defective batches.

[0067] The risk calculation module is used to calculate a comprehensive risk index based on the production data, the reference baseline of the related parameters, and the weights of the related parameters.

[0068] The sampling inspection module is used to perform high-precision sampling inspection of the cylinder block size of the real-time production batch based on a preset sampling ratio when the comprehensive risk index is greater than or equal to a preset risk threshold, so as to obtain high-precision inspection results.

[0069] The template construction module is used to construct production data feature templates for various defect types based on high-precision detection results and production data from multiple real-time production batches when the high-precision detection results indicate the presence of defects; and to adjust the preset sampling ratio based on feedback when the high-precision detection results indicate the absence of defects.

[0070] The dual-path detection module is used to match real-time collected production data with production data feature templates for multiple defect types when constructing production data feature templates for multiple defect types. When the production data matches the production data feature template for any defect type, the cylinder body is sampled and high-precision dimensional detection of the corresponding defect type is performed; otherwise, the process returns to the risk calculation module.

[0071] The beneficial effects of this application are as follows: This application provides a high-precision dimensional inspection method and system for motorcycle engine cylinder blocks. By constructing a comprehensive risk index, it achieves precise quantification of production line status. High-precision full inspection is triggered only when the risk index exceeds the limit, while a low proportion of routine sampling is maintained during steady-state production. This concentrates expensive coordinate measuring machine / cylindricity measuring instrument inspection resources on high-risk periods, significantly reducing the ineffective use of inspection resources while ensuring the detection rate. By constructing production data feature templates for various defect types based on high-precision inspection results, the system can not only determine the existence of anomalies but also accurately identify the specific defect type (such as out-of-tolerance ellipticity, out-of-tolerance taper, etc.) through template matching, providing clear troubleshooting directions for on-site maintenance personnel and significantly shortening downtime for troubleshooting. This application can establish a reference baseline and activate a general sentinel through normal batches without relying on a large amount of historical scrap data. At the same time, the template construction mechanism allows the system to automatically rebuild new templates as new batch data accumulates after process changes, without manual intervention or re-accumulating months of data, demonstrating good adaptability to process changes. Attached Figure Description

[0072] The present application will be further described below with reference to the accompanying drawings and embodiments:

[0073] Figure 1 This is a flowchart illustrating a high-precision method for detecting the dimensions of a motorcycle engine cylinder block in one embodiment of this application;

[0074] Figure 2 This is a schematic diagram illustrating the extraction process of reference baselines, associated parameters, and weights of associated parameters for various production parameters in one embodiment of this application.

[0075] Figure 3 This is a schematic diagram of the production data feature template construction process in one embodiment of this application;

[0076] Figure 4 This is a schematic diagram of the feedback adjustment process in one embodiment of this application;

[0077] Figure 5 This is a schematic diagram of the template matching process in one embodiment of this application. Detailed Implementation

[0078] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0079] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the layers related to this application and are not drawn according to the actual number, shape and size of the layers in the actual implementation. In the actual implementation, the form, number and proportion of each layer can be arbitrarily changed, and the layer layout may also be more complex.

[0080] Numerous details are explored in the following description to provide a more thorough explanation of embodiments of this application; however, it will be apparent to those skilled in the art that embodiments of this application may be practiced without these specific details.

[0081] Figure 1 This is a flowchart illustrating a high-precision dimensional detection method for a motorcycle engine block, as shown in one embodiment of this application. Figure 1 As shown, the high-precision dimension detection method for a motorcycle engine block according to this application mainly includes the following steps:

[0082] S1, acquire historical production process data and test results of multiple historical batches of the motorcycle engine cylinder block production line within the most recent time window, and acquire real-time production data of the motorcycle engine cylinder block production line.

[0083] (1) Motorcycle engine block production line:

[0084] In motorcycle cylinder block machining, the core processes that determine the final dimensional accuracy are usually honing or precision boring. This process is the focus of data collection and includes:

[0085] Honing machine: Uses an oilstone (honing bar) to reciprocate and grind the inner wall of the cylinder, ultimately determining the final cylinder diameter, roundness, cylindricity and surface roughness.

[0086] Related auxiliary equipment: cutting fluid circulation system (temperature control), clamping system (clamping force), blank loading machine.

[0087] (2) Historical production process data;

[0088] Historical production process data refers to the time-series data of physical quantities collected and stored in real time by the equipment's PLC and sensors during the honing process, including but not limited to:

[0089] (2-1) Force / current data: Load current (A) or torque of the honing spindle drive motor (reflecting cutting resistance).

[0090] (2-2) Thermal data: Real-time temperature of cutting fluid in the honing zone (°C) (Thermal expansion and contraction directly affect the final cold-state dimensions).

[0091] (2-3) Position / feed data: Radial feed or reciprocating speed of the honing head.

[0092] (2-4) Workpiece data: the hardness (HB) of the blanks in this batch.

[0093] (3) Test results;

[0094] The test results are the physical quality results measured by pneumatic plug gauges or cylindricity testers after the cylinder block is sent to the metrology room or online testing station after the honing process. Exemplarily, these include:

[0095] Quantitative results: measured cylinder diameter (e.g., 76.002 mm), roundness error (0.003 mm), and cylindricity error (0.005 mm).

[0096] Qualitative label (Y): Based on process tolerances (such as diameter tolerance ±0.01mm), quantitative results are converted into binary labels of "0 (acceptable batch)" or "1 (defective batch / out of tolerance)".

[0097] S2, based on the detection results, extract defective batches and normal batches from multiple historical batches, extract reference baselines for various production parameters from the historical production process data of the normal batches; extract correlation parameters and weights of correlation parameters that are associated with the increase in defect rate from the historical production process data of the defective batches.

[0098] Specifically, the overall defect rate of each batch is calculated from the test results. Batches with an overall defect rate less than or equal to the classification threshold are considered normal batches, while batches with an overall defect rate greater than the classification threshold are considered defective batches.

[0099] Due to the lack of defective samples and their uneven distribution in the early stages, this application does not directly calculate the defect rate for each defect type. Instead, it first calculates the overall defect rate, extracts defective batches and normal batches from the overall defect rate, and then extracts a reference baseline. This baseline serves as a general sentinel for overall risk monitoring. Specifically, since there are far more qualified batches than defective batches in historical data (the positive-to-negative sample ratio is usually greater than 100:1), directly calculating the defect rate for specific defect types such as out-of-tolerance roundness and out-of-tolerance taper would face statistical failure due to sample sparsity. Most defect types occur 0 times within the window period, making it impossible to calculate meaningful mutual information or mean baseline.

[0100] Therefore, the overall approach of this application is as follows: First, batch coarse screening is performed based on the overall defect rate. Then, a reference baseline, correlation parameters, and the weights of these parameters are extracted based on the coarse screening results to calculate the overall risk. When the overall risk is high, high-precision sampling inspection is performed. Next, a refined posterior classification is performed based on the high-precision inspection results. After classification, high-precision matching inspection based on accurate templates and fallback inspection based on overall risk are performed. This achieves accurate detection from cold start to increased data volume, effectively balancing detection accuracy and efficiency.

[0101] Figure 2 This is a schematic diagram illustrating the extraction process of reference baselines, associated parameters, and weights of various production parameters in one embodiment of this application. Figure 2 As shown, based on the above overall approach, the extraction process of reference baselines, related parameters, and weights of related parameters for various production parameters includes:

[0102] S2-1, Extract the original value sequence of various production parameters from the historical production process data of the normal batch, and perform data cleaning on the original value sequence of the various production parameters to obtain the value sequence. The data cleaning includes steady-state time point data extraction and physical boundary filtering.

[0103] Data in the non-steady-state range reflects the inertial impact of acceleration and deceleration rather than cutting force. Data outside the physical limits comes from sensor failure or emergency stop events, both of which contaminate the statistical baseline.

[0104] Therefore, in this application, for each historical batch marked as a "normal batch", complete raw sensor timing data (such as spindle current and cutting fluid temperature) of the honing process are read from the Manufacturing Execution System (MES) or machine tool PLC log. Then, a two-step cleaning process is performed: ① Steady-state extraction: The first 5 seconds (tool feed impact segment) and the last 3 seconds (tool retraction idle segment) of each machining cycle are removed, retaining only the intermediate steady-state cutting segment; ② Physical boundary filtering: Significantly distorted data exceeding physical limits (such as current < idle value or > short-circuit value) are removed.

[0105] S2-2, calculate the mean and standard deviation of the value series of various production parameters to obtain the reference baseline of various production parameters.

[0106] Calculate the arithmetic mean of the parameter value sequence after cleaning. and sample standard deviation Store the two as pairs of parameters. The reference baseline is used, and the current product model is associated with the time period index.

[0107] S2-3, Extract the target data segment corresponding to the production and processing time period of the defective cylinder body from the historical production process data of the defective batch;

[0108] The purpose of steps S2-3 is to accurately pinpoint the final inspection error to the physical machining period that produced the result, eliminating interference data from other qualified parts in the same batch. Therefore, based on the specific serial number of the "defective cylinder body" marked in the high-precision final inspection report, the machining timestamp on the honing machine is traced back, and the complete sensor waveform segment from clamping and positioning to machining completion of the cylinder body is accurately extracted from the machine tool log.

[0109] S2-4, Extract the time features and ambient temperature features of the target data segment, and extract a reference data segment that is consistent with the time features and ambient temperature features from the historical production process data of the normal batch;

[0110] Specifically, the time characteristics of the batch to which the target data segment belongs and the workshop ambient temperature are extracted as index conditions. Multiple normal cylinder block data segments from the same shift with ambient temperature fluctuations within ±1℃ are selected from the normal batch library as the control group for the defective sample.

[0111] S2-5, Perform the Mann-Whitney rank-sum test on the value sequence of each parameter in the target data segment and the value sequence of the corresponding parameter in the reference data segment to obtain the significance level; and use the parameters with significance levels lower than the preset significance threshold as the initial association parameters;

[0112] For each production parameter, the value sequences of the target group (defective samples) and the control group (paired normal samples) are input into the Mann-Whitney U test model to calculate the significance level. Value. The parameters are marked as initial associated parameters.

[0113] This allows for the rapid selection of candidate features that are statistically correlated with defect events from dozens of sensor features, significantly reducing the feature dimension of subsequent mutual information calculation, lowering computational overhead, and avoiding the introduction of noise from irrelevant features.

[0114] S2-6, extract the target segment sequence of initial association parameters from the target data segment, and extract the reference segment sequence of initial association parameters from the reference data segment;

[0115] Based on the screening results of S2-5, the complete sequences of all initial correlation parameters in the target defective fragment and the paired normal fragment are extracted from the original data matrix and organized into a two-dimensional array (number of samples × number of features) according to the sample dimension.

[0116] S2-7, Calculate the mutual information between the initial association parameters and the defect-generating event based on the target fragment sequence and the reference fragment sequence of the initial association parameters;

[0117] The formula for calculating mutual information is:

[0118]

[0119] In the formula, For mutual information, Indicates the input random variable. This indicates that a random variable will be output. For the value space of the feature, For the value space of the label, Indicates features in The interval label is The probability of the label Indicates the presence of defects, label This indicates that there are no defects. For the feature in The probability of an interval, For the tag The probability of.

[0120] S2-8, collinearity analysis is performed on multiple initial correlation parameters to obtain multiple initial correlation parameter groups with collinear relationships; and the initial correlation parameter with the largest mutual information in each initial correlation parameter group is taken as the correlation parameter;

[0121] Physically coupled variables (such as spindle power and spindle current) carry highly repetitive information. If all of this information is retained, the contribution of truly independent factors will be diluted in the weighted summation of S3 due to the dispersion of weights. Therefore, the following method is used to extract collinear relationships and remove redundant parameters:

[0122] S2-8-1, Calculate the Pearson correlation coefficient of any two initial correlation parameters based on the target fragment sequence and the reference fragment sequence;

[0123] Calculate the Pearson correlation coefficients between all pairs of initial correlation parameters. Pearson correlation coefficient The formula for calculation is:

[0124]

[0125] In the formula, For samples (fragment sequences) of The value of the parameter, for The sample mean of the parameter. For samples (fragment sequences) of The value of the parameter, for The sample mean of the parameter. This represents the number of samples.

[0126] S2-8-2 combines two initial correlation parameters whose Pearson correlation coefficient is greater than or equal to a preset correlation threshold into an initial correlation parameter group that has a collinear relationship.

[0127] Will The parameters are grouped into the same collinear group. For each collinear group, only the parameter with the highest mutual information value within the group is retained as the final association parameter, and the rest are discarded.

[0128] S2-9, normalize the mutual information of multiple correlation parameters to obtain the weight of each correlation parameter.

[0129] The mutual information values ​​of the m correlation parameters retained after collinearity filtering are normalized: This makes the sum of all weights equal to 1.

[0130] S3, Calculate a comprehensive risk index based on the production data, the reference baseline of the related parameters, and the weights of the related parameters, including:

[0131] S3-1, Calculate the single-parameter deviation score of the associated parameter based on the production data and the reference baseline of the associated parameter. The single-parameter deviation score The formula for calculation is:

[0132]

[0133] In the formula, For parameters associated with production data The current value, The mean is used as a reference baseline. The standard deviation is the reference baseline.

[0134] Because the physical dimensions of the related parameters are different (e.g., current is measured in amperes (A) and temperature in degrees Celsius (°C), they cannot be directly compared or added. This is achieved by comparing the measured values ​​with a historical steady-state baseline. The offset, divided by the inherent natural fluctuation range of the parameter. This maps all parameters to a dimensionless scale that deviates from a certain intrinsic standard deviation. The denominator is taken as... Because in statistics, under the normal distribution Fluctuations within this range are considered normal process fluctuations; only deviations exceeding this range are significantly amplified, thus assigning... Clear statistical significance.

[0135] S3-2, Single-parameter deviation score of the associated parameters based on the weights of the associated parameters. We perform a weighted summation to obtain the overall risk. .

[0136] Comprehensive Risks The formula for calculation is:

[0137]

[0138] The influence of each related parameter on the cylinder block dimensional accuracy varies (e.g., the causal weight of spindle load fluctuation is much higher than that of cutting fluid temperature), making it impossible to simply calculate the deviation score using an arithmetic average. By assigning different weights and then summing them, the parameters that contribute more to the defect rate (determined by the mutual information quantification in S2-7) occupy a higher share in the total risk index, thereby achieving weighted aggregation of risks.

[0139] S4. When the comprehensive risk index is greater than or equal to the preset risk threshold, high-precision sampling inspection of the cylinder size of the real-time production batch is carried out based on the preset sampling ratio to obtain high-precision inspection results.

[0140] The system automatically picks up all cylinder blocks from the current production batch and the subsequent Q-piece batches (Q is dynamically calculated based on the production line cycle time and full inspection time, typically 15-20 pieces) and places them into an independent physical isolation buffer area using an automated robotic arm or conveyor belt diversion device. At this point, the preset sampling ratio is effectively 100% (this batch and the immediately following Q-piece batches all enter the buffer), rather than a statistically significant small-scale inspection. The production line continues uninterrupted, and subsequent batches are produced normally, awaiting the buffer inspection results before a release or isolation decision is made.

[0141] For each cylinder block within the buffer area, the following combined precision measurement process is executed sequentially:

[0142] (1) Cylindricity gauge scanning: The cylinder is placed vertically on the air-float rotating worktable, the probe contacts the cylinder wall, the workpiece rotates one revolution, and 360° continuous contour is collected in three axial sections (upper, middle and lower), and the roundness error (μm) and cylindricity error (μm) are output.

[0143] (2) Verification by 12-point method using digital micrometer: In order to eliminate the systematic deviation of a single instrument, the micrometer after the ring gauge has been calibrated and zeroed is used manually or automatically to measure the diameter in four directions (0°, 45°, 90°, 135°) in a cross-section of the cylinder at the top, middle and bottom sections, and a total of 12 diameter values ​​are collected. The maximum and minimum values ​​and ellipticity of each cross-section are calculated.

[0144] (3) Ultrasonic wall thickness gauge: Select 6 measurement points evenly on the outer wall of the cylinder (corresponding to different circumferential positions of the cylinder) to detect the minimum wall thickness and wall thickness uniformity, and ensure that the wall thickness is not lower than the design safety limit.

[0145] (4) Profile roughness tester for honing texture: The probe is dragged along the generatrix of the cylinder wall to measure the core roughness of the honing texture. and the depth of the anti-friction groove To assess whether the oil storage capacity meets the standards.

[0146] (5) Comprehensive judgment: The above four data are summarized into the MES system. If any single item reaches the scrap limit (such as roundness ≥ 0.05mm, wall thickness ≤ design lower limit, roughness exceeds the standard), the whole batch is judged as unqualified and transferred to the rework / scrapping process; only when all items are qualified can the buffer area be released in batches.

[0147] Furthermore, when the comprehensive risk index is less than the preset risk threshold, high-precision full inspection is not triggered, and the process reverts to the conventional steady-state sampling strategy. That is, a low-proportion (initially set at 3%~5%) of the batch is randomly sampled. The sampling method is as follows: after the batch is processed, an automatic sorting device picks up samples at fixed intervals (e.g., 1 piece for every 20 pieces processed). Only a pneumatic electronic plunger gauge (non-contact, with extremely low measuring force) is used to read the diameter of a single section 10mm from the upper end face (the area of ​​maximum piston ring wear), and the surface roughness is visually inspected or compared with a sample block. If the diameter of the sampled part exceeds the upper or lower tolerance limit, the batch is immediately upgraded to a "high-risk state" and enters the above-mentioned situation A to perform 100% full inspection in the isolation buffer; if the sampled part is qualified, the entire batch is released normally.

[0148] S5, when the high-precision detection result indicates the presence of defects, construct production data feature templates for various defect types based on the high-precision detection results and production data of multiple real-time production batches; when the high-precision detection result indicates the absence of defects, adjust the preset sampling ratio and / or risk threshold accordingly.

[0149] Specifically, if the test data of any cylinder block within the buffer area reaches or exceeds any of the preset physical conditions in any of the following four core indicators, the batch is judged to have defects. The judgment criteria are detailed in Table 1:

[0150] Table 1. Defect Triggering Conditions

[0151]

[0152] If the high-precision detection results indicate defects, the production data corresponding to the defective cylinder body will be saved to the set of the corresponding defect type. After accumulating enough samples, the corresponding production data feature template can be constructed.

[0153] Figure 3 This is a schematic diagram of the production data feature template construction process in one embodiment of this application, such as... Figure 3 As shown, it specifically includes:

[0154] S5-1, extract the defect type from the high-precision detection result, extract the target production data segment corresponding to the defect type cylinder from the production data, and extract multiple reference production data segments of the normal cylinder that is closest to the defect type cylinder from the production data.

[0155] When the S4 high-precision detection determines that a cylinder block has a defect (such as out-of-tolerance roundness), the system traces back its processing timestamp from the MES system based on the cylinder block's unique serial number, and extracts complete sensor waveform data for that time period (including clamping and positioning, steady-state cutting to tool retraction) from the machine tool PLC log as the target production data segment. Simultaneously, the system automatically retrieves continuous production records from the two hours prior to the processing time of the defective cylinder block, selects three normal cylinder blocks of the same model and process that passed final inspection, and extracts their corresponding production data segments as reference production data segments.

[0156] The purpose of selecting the most recent normal sample as a control is to eliminate common-mode factors such as gradual wear and tear of equipment and slow drift of ambient temperature, so that the incremental features extracted later can be attributed only to the sudden or specific anomaly that caused the defect.

[0157] S5-2, Standardize the value sequences of various parameters in the target production data segment to obtain the target vector. The reference production data segment is then standardized to obtain a reference vector. ;

[0158] For the parameter value sequences (such as spindle load current, cutting fluid temperature, vibration amplitude, etc.) extracted in S5-1 from the target segment and reference segment, the global reference baseline established in step S2-2 is used. (mean) Standardize (standard deviation) using the formula: ,in, The standardized vector is denoted as follows, depending on the sequence type: and , This is the vector before standardization.

[0159] S5-3, for multiple reference vectors The average reference vector is obtained by averaging. ; and calculate the target vector With the average reference vector The difference is used to obtain the template feature vector. ;

[0160] First, calculate the arithmetic mean of the three reference vectors along their dimensions, and obtain... Then, the difference between the target vector and the average reference vector is calculated: This vector is the template feature increment vector of the defective sample. Through interpolation, the overlapping steady-state drift components in the target and reference vectors are canceled out, retaining only the specific deviation increment of the defective cylinder relative to a normal cylinder from the same period. This increment represents the fingerprint direction of the physical cause of the defect in the production parameter space.

[0161] S5-4, in the template feature vector When there are no abnormal events during the extraction time, the template feature vector will be... Stored in a pre-constructed set of template feature vectors for the defect type;

[0162] In Before storing the template feature vector set corresponding to the defect type, the system first retrieves the equipment alarm logs within 5 minutes before and after the machining time of the defective cylinder block to check for non-continuous abnormal events such as "power grid voltage interruption," "emergency stop triggering," and "sudden drop in cutting fluid pressure." If such events exist, the system... Not involved in template building (discard directly, only record the reason for discarding); if no abnormal events are confirmed, then... Store the template feature vector set for this defect type (e.g., "ellipticity out of tolerance"). middle.

[0163] S5-5, when the number of template feature vectors in the template feature vector set of any defect type is greater than or equal to the preset update trigger threshold, unsupervised clustering is performed on the template feature vectors in the template feature vector set of the defect type to obtain multiple feature vector clusters;

[0164] The system continuously monitors the template feature vector set for each defect type. Number of samples .when When the time is right, the automatic clustering process is triggered. DBSCAN (a density-based spatial clustering algorithm) is used to cluster all elements in the set. The vectors are clustered, with the initial parameters set as neighborhood radius ε=0.5 and minimum number of samples MinPts=2.

[0165] The same type of defect (e.g., "out of ovality tolerance") can be caused by a variety of different physical factors (e.g., "honing strip wear" and "excessive clamping force"). When vectors have different directions, directly averaging them will generate a mixing center with no physical meaning. Clustering can naturally group samples according to the direction of the inducing factor, forming multiple sub-patterns.

[0166] S5-6, the feature vector clusters with a number of samples within the cluster greater than or equal to a preset number threshold are taken as valid clusters, and the Euclidean standard deviation of all template feature vectors in the valid clusters is calculated.

[0167] For all feature vector clusters generated by clustering, first remove the number of samples within each cluster. Clusters (considered noise points and not included in the template library). For each remaining valid cluster... First, calculate its cluster center vector. Then calculate the Euclidean standard deviation of the cluster:

[0168]

[0169] in, Let L2 be the L2 norm of the vector.

[0170] Euclidean standard deviation quantifies the average distance of each sample within a cluster from the cluster center in space, directly reflecting the clustering degree of that sub-pattern. The lower the clustering degree, the more consistent the physical causes of the template, the more stable the features, and the higher the confidence of subsequent matching.

[0171] S5-7, when the Euclidean standard deviation is greater than the preset standard deviation threshold, adjust the unsupervised clustering parameters under the adjustment range constraint and return to the template feature vectors in the template feature vector set of the defect type for unsupervised clustering until there is an effective cluster with an Euclidean standard deviation less than or equal to the preset standard deviation threshold or the number of iterations exceeds the set value; when the Euclidean standard deviation is less than or equal to the preset standard deviation threshold, extract the center vector of the effective cluster to obtain the production data feature template of the defect type;

[0172] Determine the validity of each cluster Is it ≤ preset standard deviation threshold? (The recommended value for engineering is 0.6~0.8).

[0173] like Extract the cluster center vector As a formal production data feature template, it is stored in the template library and associated with the corresponding defect type label.

[0174] like And if the number of iterations is less than 3: under the preset adjustment range constraint (e.g., ε decreases by 0.05, MinPts remains unchanged), re-execute the clustering of S5-5 to try to further split the clusters with excessive dispersion.

[0175] If the target cannot be met even after 3 iterations: the system does not discard the cluster, but instead downgrades it to a broad matching template, which is the standard deviation of the cluster. Multiply by a safety factor of 1.5 as the boundary radius, and mark it as a low-confidence template in the system. Its matching priority is lower than that of the formal template. At the same time, push an alarm to the quality engineer's console: "Defect type X has a complex mixed pattern. It is recommended to manually review the subclass division."

[0176] The above process forcibly splits clusters with excessive dispersion, ensuring that each sub-pattern of the formal template has high cohesion; by setting retry limits and degradation mechanisms, it ensures that the system can still produce usable templates even without a perfect solution, avoiding getting stuck in an infinite loop due to the inability of data to be perfectly aggregated.

[0177] S5-8: When at least one production data feature template exists for each defect type, complete the construction of production data feature templates for multiple defect types.

[0178] The system continuously monitors all defined defect types (such as ellipticity out-of-tolerance, taper out-of-tolerance, roughness out-of-tolerance, insufficient wall thickness, etc.). When each defect type has at least one formal template or broad template in the template library, the system determines that the production data feature templates for multiple defect types have been constructed, and then activates the dual-template parallel matching process of S6. If a defect type has no template for a long period of time (e.g., more than 30 days) due to extremely few samples, the system still allows S6 to start, but only maintains the general sentinel detection of S3-S4 for that type.

[0179] Figure 4 This is a schematic diagram of the feedback adjustment process in one embodiment of this application, as shown below. Figure 4 As shown, when the high-precision detection result indicates the absence of defects, feedback adjustment is performed on the preset sampling ratio and / or risk threshold, specifically including:

[0180] S5-9: Calculate the false alarm rate, false negative rate, and template matching hit rate within the most recent time window; and obtain the high-precision detection load rate at the current moment.

[0181] Specifically, the system uses a 24-hour sliding statistical window (recommended configuration is every 50 batches or every shift) to summarize the following four core indicators from the MES database and testing equipment logs:

[0182] False alarm rate Sentinel trigger ( After that, the number of batches that passed the high-precision full inspection / the total number of sentinel triggers;

[0183] underreporting rate Sentinel not triggered ( However, the number of batches that failed the routine low-percentage sampling inspection / the total number of batches that failed the inspection on the production line;

[0184] Template matching hit rate : The number of batches that successfully matched the existing exclusive template in real-time production data / the total number of production batches after the template construction was completed;

[0185] High-precision detection load rate : Actual occupancy time / total available time of high-precision testing equipment (cylindricity tester, etc.) (usually 8 hours / shift).

[0186] False positive rate reflects the oversensitivity of the general sentinel, false negative rate reflects the risk of insufficient coverage, match hit rate reflects the maturity of the dedicated template library, and load rate reflects the scarcity of detection resources. These four indicators together constitute the input vector for feedback adjustment.

[0187] S5-10, when the false negative rate is greater than a preset percentage threshold, update the risk threshold and the preset sampling ratio to obtain the updated sampling ratio. and the updated risk threshold Among them, the updated sampling ratio and the updated risk threshold The formula for calculation is:

[0188]

[0189]

[0190] In the formula, The preset sampling ratio, To update the unit value proportionally, For the maximum sampling proportion, The risk threshold is... The first adjustment coefficient is a value less than 1;

[0191] Specifically, the percentage threshold can be 0.5%. A false negative rate exceeding 0.5% means that defective cylinder blocks have escaped to subsequent processes or the assembly line, indicating that the current general-purpose sentinel is too insensitive and the routine sampling density is insufficient, which is a quality incident-level alarm. At this point, the sampling ratio should be increased, while simultaneously making the sentinel more sensitive (reducing...). This ensures that the same type of defect can be detected in the next batch in a timely manner.

[0192] S5-11, when the false negative rate is less than or equal to a preset percentage threshold, the preset sampling ratio is adjusted based on the false positive rate and the high-precision detection load rate, specifically including:

[0193] S5-11-1, when the false alarm rate is greater than the first percentage and the high-precision detection load rate is less than the second percentage, the preset sampling ratio is reduced based on the preset first proportional coefficient;

[0194] A high false alarm rate indicates numerous false detections by the sentinel, while a low load rate on the detection equipment suggests excess capacity. This indicates that the current routine sampling ratio is set too high, consuming unnecessary testing manpower and resources without creating bottlenecks for the equipment, representing a waste of efficiency. Appropriately reducing the sampling ratio could free up this inefficient work.

[0195] S5-11-2, when the false alarm rate is less than the third percentage and the high-precision detection load rate is greater than the fourth percentage, the preset sampling ratio is reduced based on the preset second proportional coefficient;

[0196] The extremely low false positive rate indicates that the current sentinel model is highly accurate, but the load rate of the detection equipment has exceeded the 85% congestion threshold. High load likely means a backlog of high-risk, full-check tasks in the buffer, with routine spot checks crowding out equipment schedules that should be reserved for emergency full checks of the buffer, resulting in resource congestion. Appropriately reducing the proportion of routine spot checks will free up equipment time to ensure priority response for high-risk buffers.

[0197] S5-11-3, when the false alarm rate is greater than the first percentage and the high-precision detection load rate is greater than the fourth percentage, update the preset sampling ratio. The updated sampling ratio is... The formula for calculation is:

[0198]

[0199] In the formula, For proportional update coefficients, This is the minimum sampling proportion;

[0200] The percentages are arranged in descending order as follows: fourth percentage (85%), second percentage (70%), first percentage (20%), and third percentage (5%).

[0201] A high false alarm rate indicates a large number of invalid triggers by the sentinel, while a detection equipment load rate exceeding 85% indicates severe overload. The combination of these two factors suggests the system is trapped in false alarm-induced congestion, wasting capacity and preventing timely detection of truly high-risk batches. At this point, it is imperative to drastically reduce routine spot checks, freeing up equipment resources from ineffective full-inspection tasks, alleviating congestion first, and then optimizing the model.

[0202] S5-12, after constructing production data feature templates for multiple defect types, and when the template matching hit rate is higher than a preset hit rate threshold for a continuous time period, the preset sampling ratio is updated to obtain the updated sampling ratio. Among them, the updated sampling ratio The formula for calculation is:

[0203]

[0204] In the formula, The second adjustment coefficient is a value less than 1. This is the set safety lower limit.

[0205] The system continuously monitors the matching hit rate after the template library is built. Within 7 consecutive calendar days (or 100 consecutive batches) When the (preset hit rate threshold) is reached, the system implements a gradual reduction in rewards. A high template matching hit rate indicates that the system has successfully evolved from the general sentinel stage to the precision diagnosis stage, with most potential defects being intercepted in advance during the template matching stage of S6 and diverted to the corresponding high-precision detection. At this point, the marginal value of routine spot checks decreases, and their proportion can be gradually and steadily reduced while ensuring safety, further allocating detection resources to the template matching and buffer full inspection stages.

[0206] S6. When constructing production data feature templates for multiple defect types, the real-time collected production data is matched with the production data feature templates for multiple defect types. When the production data matches the production data feature template for any defect type, the cylinder is sampled and the corresponding defect type's dimensions are inspected with high precision. Otherwise, steps S3-S4 are executed.

[0207] Figure 5 This is a schematic diagram of the template matching process in one embodiment of this application, such as... Figure 5 As shown, real-time collected production data is matched with production data feature templates for various defect types, including:

[0208] S6-1, Extract a data segment of a single cylinder block from the production data, and standardize the value sequence of each associated parameter in the data segment to obtain the current feature vector;

[0209] After the real-time batch processing is completed, the system extracts a complete processing time segment (including clamping and positioning, steady-state cutting to tool retraction) from the machine tool PLC log based on the current cylinder block's serial number. It then extracts the value sequences of various related parameters (such as average spindle load, peak cutting fluid temperature, vibration amplitude, etc.) determined in step S2-8. Subsequently, the system standardizes the value sequences of each parameter using the global reference baseline established in step S2-2. After standardization, the mean value of the steady-state cutting segment is taken for each parameter, and all parameters are aggregated into a fixed-dimensional vector, denoted as the current feature vector. .

[0210] S6-2, Calculate the difference between the current feature vector and the pre-maintained benchmark feature vector to obtain the feature to be matched;

[0211] During real-time batch processing, the system simultaneously retrieves consecutive normal batches from the MES system that are closest in processing time to the current batch (e.g., within the previous 2 hours), and extracts their baseline feature vectors after undergoing the same standardization process as S6-1 and storing them. (This vector is either the mean of the feature vectors of the corresponding normal batch or directly taken from the mean of the normal batch stored during the S5-3 calculation process.) Then, the difference vector is calculated: This vector is the feature increment to be matched.

[0212] S6-3, calculate the cosine similarity and normalized distance between the feature to be matched and the production data feature templates of various defect types, and determine that the production data matches the production data feature template and there is a corresponding type of defect risk when the cosine similarity between any production data feature template and the feature to be matched is greater than or equal to a preset similarity threshold and the normalized distance is less than or equal to a preset distance threshold; otherwise, determine that there is no defect risk.

[0213] The formula for calculating cosine similarity is: .

[0214] The formula for calculating the normalized distance is: ,in, Boundary radius, boundary radius Use the Euclidean standard deviation of S5-6 multiplied by a fixed coefficient (e.g., Alternatively, the robust MAD (Median Absolute Deviation) method can be used to determine it.

[0215] Cosine similarity measures the directional consistency between two incremental vectors, is not sensitive to amplitude, and can identify whether the increase in load and temperature is consistent with the direction of the physical cause of the template.

[0216] Normalized distance measures the severity of the deviation, ensuring that the current deviation level does not exceed the empirical tolerance limit of the template. Matches with only consistent direction but extremely small amplitudes are considered normal fluctuations (and should not be matched), while matches with only large amplitudes but inconsistent directions belong to other unknown patterns (and should not be incorrectly matched). Dual-judgment, through cross-validation of two orthogonal dimensions—direction and amplitude—upgrades the matching logic from one-dimensional similarity to two-dimensional joint decision-making, thereby effectively improving matching accuracy.

[0217] In one specific embodiment, the system first loads 60 batches of historical production data through "Import Example Historical Data," and automatically completes the baseline extraction of normal batches and the calculation of the correlation parameter weights for defective batches. The log displays the reference baselines for each production parameter (such as load baseline μ=41.997, σ=2.884) and the correlation parameter weights obtained after normalization through mutual information and collinearity analysis (such as the vibration parameter weight being 0.668, indicating that vibration has the greatest impact on the defect rate).

[0218] Subsequently, the system acquires the production parameters of the current batch (load 43.5A, temperature 36.2℃, vibration 0.25mm / s), calculates a comprehensive risk index R=0.3136 based on the reference baseline and weights, and decomposes the contribution of each parameter (vibration contributes 0.255, which is dominant). Since the comprehensive risk index 0.3136 is less than the preset risk threshold of 1.0, the system does not trigger high-precision full inspection, and only performs conventional low-proportion sampling inspection, with the inspection conclusion being qualified. Meanwhile, the log shows that the template library is currently empty, so the dual-mode parallel matching of S6 cannot be executed, and the system automatically falls back to the general sentinel detection mode.

[0219] This log example fully demonstrates the core processes from S2 to S4, and verifies that the technical path of the present application, which performs overall risk screening first and then executes high-precision detection as needed, effectively avoids the waste of detection resources, and provides a data basis for the gradual accumulation of subsequent templates and the self-evolution of the system.

[0220] The high-precision dimension detection method for a motorcycle engine cylinder block of the present application realizes accurate quantification of production line status by constructing a comprehensive risk index. High-precision full inspection is triggered only when the risk index exceeds the limit, and low-proportion conventional sampling inspection is maintained when the production line is in a steady state, so that the expensive inspection resources of coordinate measuring machines / cylindricity meters are centrally allocated to high-risk periods, which significantly reduces the invalid occupation of inspection resources while ensuring the detection rate. By reversely constructing production data feature templates for multiple defect types based on high-precision detection results, the system can not only judge the presence of anomalies, but also accurately identify the specific defect type to which the anomaly belongs (such as out-of-tolerance ovality, out-of-tolerance taper, etc.) through template matching, providing clear fault troubleshooting guidance for on-site maintenance personnel and greatly shortening the downtime troubleshooting time. The present application does not need to rely on a large amount of historical scrap data to establish a reference baseline through normal batches and start the general sentinel; at the same time, the template construction mechanism enables the system to automatically reconstruct new templates with the accumulation of new batch data after the process is changed, without manual intervention or re-accumulation of data for several months, and has good self-adaptability to process changes.

[0221] The present application further provides a high-precision dimension detection system for a motorcycle engine cylinder block, comprising:

[0222] An acquisition module, configured to acquire historical production process data and detection results of multiple historical batches of the motorcycle engine cylinder block production line within the latest time window, and acquire production data of the real-time production batch of the motorcycle engine cylinder block production line;

[0223] The feature extraction module is used to extract defective batches and normal batches from multiple historical batches based on the detection results, extract reference baselines for various production parameters from the historical production process data of the normal batches, and extract correlation parameters and weights of the correlation parameters that are associated with the increase in defect rate from the historical production process data of the defective batches.

[0224] The risk calculation module is used to calculate a comprehensive risk index based on the production data, the reference baseline of the related parameters, and the weights of the related parameters.

[0225] The sampling inspection module is used to perform high-precision sampling inspection of the cylinder block size of the real-time production batch based on a preset sampling ratio when the comprehensive risk index is greater than or equal to a preset risk threshold, so as to obtain high-precision inspection results.

[0226] The template construction module is used to construct production data feature templates for various defect types based on high-precision detection results and production data from multiple real-time production batches when the high-precision detection results indicate the presence of defects; and to adjust the preset sampling ratio based on feedback when the high-precision detection results indicate the absence of defects.

[0227] The dual-path detection module is used to match real-time collected production data with production data feature templates for multiple defect types when constructing production data feature templates for multiple defect types. When the production data matches the production data feature template for any defect type, the cylinder body is sampled and high-precision dimensional detection of the corresponding defect type is performed; otherwise, the process returns to the risk calculation module.

[0228] This application discloses a high-precision dimensional inspection system for motorcycle engine cylinder blocks. By constructing a comprehensive risk index, it achieves precise quantification of production line status. High-precision full inspection is triggered only when the risk index exceeds limits, while a low proportion of routine sampling is maintained during steady-state production. This concentrates expensive coordinate measuring machine / cylindricity measuring instrument resources on high-risk periods, significantly reducing the ineffective use of inspection resources while ensuring detection rates. By constructing production data feature templates for various defect types based on high-precision inspection results, the system can not only identify anomalies but also accurately identify the specific defect type (such as out-of-tolerance ellipticity or taper) through template matching. This provides clear troubleshooting directions for on-site maintenance personnel, significantly reducing downtime for troubleshooting. This application can establish a reference baseline and activate a general sentinel from normal batches without relying on a large amount of historical scrap data. Furthermore, the template construction mechanism allows the system to automatically rebuild new templates as new batch data accumulates after process changes, without manual intervention or the need to re-accumulate months of data, demonstrating excellent adaptability to process changes.

[0229] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0230] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to cause the terminal to perform any of the methods in this embodiment.

[0231] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0232] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0233] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0234] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0235] In the above embodiments, although the present application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. The embodiments of the present application are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0236] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for high-precision measurement of the dimensions of a motorcycle engine cylinder block, characterized in that, include: S1, acquire historical production process data and test results of multiple historical batches of the motorcycle engine cylinder block production line within the most recent time window, and acquire real-time production data of the motorcycle engine cylinder block production line. S2, based on the detection results, extract defective batches and normal batches from multiple historical batches, and extract reference baselines for various production parameters from the historical production process data of the normal batches; Extract correlation parameters and their weights that are associated with the increase in defect rate from historical production process data of defective batches; S3, Calculate the comprehensive risk index based on the production data, the reference baseline of the related parameters, and the weights of the related parameters; S4. When the comprehensive risk index is greater than or equal to the preset risk threshold, high-precision sampling inspection of the cylinder size of the real-time production batch is carried out based on the preset sampling ratio to obtain high-precision inspection results. S5, when the high-precision detection result indicates the presence of defects, construct production data feature templates for various defect types based on the high-precision detection results and production data of multiple real-time production batches; when the high-precision detection result indicates the absence of defects, adjust the preset sampling ratio and / or risk threshold accordingly. S6, when constructing production data feature templates for multiple defect types, the real-time collected production data is matched with the production data feature templates for multiple defect types. When the production data matches the production data feature template for any defect type, the cylinder is sampled and the corresponding defect type's dimensions are inspected with high precision. Otherwise, proceed to steps S3-S4.

2. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 1, characterized in that, Reference baselines for various production parameters are extracted from the historical production process data of the normal batch, including: The original value sequences of various production parameters are extracted from the historical production process data of the normal batch, and the original value sequences of the various production parameters are cleaned to obtain the value sequences. The data cleaning includes steady-state time point data extraction and physical boundary filtering. Calculate the mean and standard deviation of the value series of various production parameters to obtain the reference baseline for various production parameters.

3. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 1, characterized in that, Extract correlation parameters and their weights that are associated with an increase in the defect rate from historical production process data of defective batches, including: Extract the target data segment corresponding to the production and processing time period of the defective cylinder block from the historical production process data of the defective batch; Extract the time features and ambient temperature features of the target data segment, and extract a reference data segment that is consistent with the time features and ambient temperature features from the historical production process data of the normal batch. The Mann-Whitney rank-sum test is performed on the value sequence of each parameter in the target data segment and the value sequence of the corresponding parameter in the reference data segment to obtain the significance level; and the parameters with significance levels lower than the preset significance threshold are used as the initial association parameters. Extract the target fragment sequence of initial association parameters from the target data fragment, and extract the reference fragment sequence of initial association parameters from the reference data fragment; The mutual information between the initial association parameters and the defect-generating event is calculated based on the target fragment sequence and the reference fragment sequence using the initial association parameters. Collinearity analysis was performed on multiple initial correlation parameters to obtain multiple sets of initial correlation parameters that are collinear; and the initial correlation parameter with the largest mutual information in each initial correlation parameter set was taken as the correlation parameter. The mutual information of multiple correlation parameters is normalized to obtain the weight of each correlation parameter.

4. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 3, characterized in that, Collinearity analysis was performed on various initial correlation parameters, resulting in multiple sets of initial correlation parameters exhibiting collinear relationships, including: Calculate the Pearson correlation coefficient for any two initial correlation parameters based on the target fragment sequence and the reference fragment sequence; Two initial correlation parameters with a Pearson correlation coefficient greater than or equal to a preset correlation threshold are combined into an initial correlation parameter group that has a collinear relationship.

5. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 1, characterized in that, A comprehensive risk index is calculated based on the production data, the reference baseline of the associated parameters, and the weights of the associated parameters, including: The single-parameter deviation score of the correlation parameter is calculated based on the production data and the reference baseline of the correlation parameter. The single-parameter deviation score The formula for calculation is: In the formula, For parameters associated with production data The current value, The mean is used as a reference baseline. The standard deviation is the reference baseline. Single-parameter bias score of the correlation parameter based on the weight of the correlation parameter. We perform a weighted summation to obtain the overall risk.

6. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 1, characterized in that, Based on high-precision inspection results from multiple real-time production batches and production data, production data feature templates for various defect types are constructed, including: The defect type is extracted from the high-precision detection results, and the target production data segment corresponding to the defect type cylinder is extracted from the production data, as well as multiple reference production data segments of the normal cylinder that is closest to the defect type cylinder from the production data. The target production data segment is standardized by performing a series of parameter values ​​to obtain the target vector. The reference production data segment is then standardized to obtain a reference vector. ; For multiple reference vectors The average reference vector is obtained by averaging. ; and calculate the target vector With the average reference vector The difference is used to obtain the template feature vector. ; In the template feature vector When there are no abnormal events during the extraction time, the template feature vector will be... Stored in a pre-constructed set of template feature vectors for the defect type; When the number of template feature vectors in the template feature vector set of any defect type is greater than or equal to the preset update trigger threshold, unsupervised clustering is performed on the template feature vectors in the template feature vector set of the defect type to obtain multiple feature vector clusters; The feature vector clusters with a number of samples within the cluster greater than or equal to a preset number threshold are considered as valid clusters, and the Euclidean standard deviation of all template feature vectors in the valid clusters is calculated. When the Euclidean standard deviation is greater than the preset standard deviation threshold, the unsupervised clustering parameters are adjusted under the adjustment range constraint, and unsupervised clustering is performed on the template feature vectors in the template feature vector set of the defect type until there is an effective cluster with an Euclidean standard deviation less than or equal to the preset standard deviation threshold or the number of iterations exceeds the set value; when the Euclidean standard deviation is less than or equal to the preset standard deviation threshold, the center vector of the effective cluster is extracted to obtain the production data feature template of the defect type; When at least one production data feature template exists for each defect type, the construction of production data feature templates for multiple defect types is completed.

7. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 1, characterized in that, Feedback adjustment of the preset sampling ratio and / or risk threshold includes: Calculate the false alarm rate, false negative rate, and template matching hit rate within the most recent time window; and obtain the high-precision detection load rate at the current moment. When the false negative rate exceeds a preset percentage threshold, the risk threshold and the preset sampling ratio are updated to obtain the updated sampling ratio. and the updated risk threshold Among them, the updated sampling ratio and the updated risk threshold The formula for calculation is: In the formula, The preset sampling ratio, To update the unit value proportionally, For the maximum sampling proportion, The risk threshold is... The first adjustment coefficient is a value less than 1; When the false negative rate is less than or equal to a preset percentage threshold, the preset sampling ratio is adjusted based on the false positive rate and the high-precision detection load rate. After constructing production data feature templates for multiple defect types, and when the template matching hit rate is higher than a preset hit rate threshold for a continuous time period, the preset sampling ratio is updated to obtain the updated sampling ratio. Among them, the updated sampling ratio The formula for calculation is: In the formula, The second adjustment coefficient is a value less than 1. This is the set safety lower limit.

8. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 7, characterized in that, When the false negative rate is less than or equal to a preset percentage threshold, the preset sampling ratio is adjusted based on the false positive rate and the high-precision detection load rate, including: When the false alarm rate is greater than a first percentage and the high-precision detection load rate is less than a second percentage, the preset sampling ratio is reduced based on a preset first proportional coefficient. When the false alarm rate is less than the third percentage and the high-precision detection load rate is greater than the fourth percentage, the preset sampling ratio is reduced based on the preset second proportional coefficient. When the false alarm rate is greater than a first percentage and the high-precision detection load rate is greater than a fourth percentage, the preset sampling ratio is updated, and the updated sampling ratio is... The formula for calculation is: In the formula, For proportional update coefficients, This is the minimum sampling proportion; The percentages are arranged in descending order as follows: fourth percentage, second percentage, first percentage, and third percentage.

9. The method for high-precision detection of the dimensions of a motorcycle engine cylinder block according to claim 1, characterized in that, The real-time collected production data is matched with production data feature templates for various defect types, including: Extract a data segment of a single cylinder block from the production data, and standardize the value sequence of each associated parameter in the data segment to obtain the current feature vector; Calculate the difference between the current feature vector and the pre-maintained baseline feature vector to obtain the feature to be matched; Calculate the cosine similarity and normalized distance between the feature to be matched and production data feature templates of various defect types. If the cosine similarity between any production data feature template and the feature to be matched is greater than or equal to a preset similarity threshold and the normalized distance is less than or equal to a preset distance threshold, it is determined that the production data matches the production data feature template and there is a corresponding type of defect risk. Otherwise, it is determined that there is no defect risk.

10. A high-precision dimensional detection system for a motorcycle engine block, characterized in that, include: The acquisition module is used to acquire historical production process data and test results of multiple historical batches of the motorcycle engine cylinder block production line within the most recent time window, and to acquire real-time production data of the motorcycle engine cylinder block production line. The feature extraction module is used to extract defective batches and normal batches from multiple historical batches based on the detection results, and to extract reference baselines for various production parameters from the historical production process data of the normal batches. Extract correlation parameters and their weights that are associated with the increase in defect rate from historical production process data of defective batches; The risk calculation module is used to calculate a comprehensive risk index based on the production data, the reference baseline of the related parameters, and the weights of the related parameters. The sampling inspection module is used to perform high-precision sampling inspection of the cylinder block size of the real-time production batch based on a preset sampling ratio when the comprehensive risk index is greater than or equal to a preset risk threshold, so as to obtain high-precision inspection results. The template construction module is used to construct production data feature templates for various defect types based on high-precision detection results and production data from multiple real-time production batches when the high-precision detection results indicate the presence of defects; and to adjust the preset sampling ratio based on feedback when the high-precision detection results indicate the absence of defects. The dual-path detection module is used to match the real-time collected production data with the production data feature templates of multiple defect types when constructing production data feature templates of multiple defect types. When the production data matches the production data feature template of any defect type, the cylinder is sampled and the corresponding defect type of high-precision size detection is performed. Otherwise, return to the risk calculation module.

Images