Human body skin-friendly POM bracelet bar forming method based on intelligent recognition

Abstract

The invention discloses a human body skin-friendly POM bracelet bar forming method based on intelligent recognition, and particularly relates to the technical field of high polymer material forming and online detection. The POM raw material is subjected to skin-friendly modification and drying treatment, and a bar is prepared in a melt extrusion mode; the method comprises the following steps: collecting bar size and surface images through an image recognition system, and extracting cross section size features and defect features; synchronously acquiring environment temperature and humidity information, and calculating an environment thermal drift factor; combining the dimensional deviation mean value, the batch offset slope coefficient and the thermal drift factor to construct a deviation severity scoring model; when the score value exceeds a threshold value, a camera calibration adjustment strategy is automatically generated according to the trend direction, and correction operation is fed back and executed; according to the method, high-precision online detection and self-adaptive calibration control of the POM bar in the forming process can be achieved, the product size consistency and the surface quality stability are improved, and the method is suitable for quality monitoring and optimization control in the continuous and high-precision polymer product manufacturing process.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of polymer material forming and online detection technology, in particular to a human skin-friendly POM bracelet rod forming method based on intelligent identification. BACKGROUND

[0002] Human skin-friendly POM bracelet rod forming refers to the process of using a POM (polyoxymethylene) material that is friendly to human skin to form a bracelet product through rod forming process. POM material has excellent mechanical properties and wear resistance, and can realize skin-friendly properties after special modification, suitable for long-term contact with skin without causing allergy or discomfort. Rod forming is to process POM material into round bar, and then cut, process or hot press into the specific shape of bracelet. This process ensures the stability, comfort and durability of the bracelet, and is widely used in wearable devices and other fields.

[0003] In the process of continuous batch production, the detection system may not be calibrated in real time due to factors such as slight movement of camera position, optical distortion, temperature change, etc., resulting in that although a single product is within the tolerance range and is judged to be qualified, over time, the size measurement value gradually deviates from the standard, eventually causing a batch of "critical substandard products" to flow into qualified products. This trend error is usually not found by static templates, and the system also lacks dynamic calibration and offset warning mechanism, thereby burying quality risks and affecting downstream assembly or functional matching. SUMMARY

[0004] The purpose of the present application is to provide a human skin-friendly POM bracelet rod forming method based on intelligent identification to solve the problems in the background art.

[0005] In order to achieve the above purpose, the present application provides the following technical solution: a human skin-friendly POM bracelet rod forming method based on intelligent identification, comprising:

[0006] The skin-friendly polyoxymethylene raw material is dried, and skin-friendly additives and modifiers are added according to a predetermined proportion to obtain a raw material mixture with skin contact safety;

[0007] The raw material mixture is processed into POM rod with a predetermined diameter by heating, plasticizing, extruding, shaping and cooling;

[0008] Image acquisition is performed on each rod segment to extract its cross-sectional size features and surface integrity features to form a sample detection data set;

[0009] The ambient temperature and humidity data corresponding to the image acquisition time are synchronously acquired, and the environmental thermal drift factor is calculated to reflect the potential influence of environmental factors on the measurement system;

[0010] The continuous size features obtained by image acquisition are analyzed in time sequence to calculate a batch offset slope coefficient for judging the trend accumulation direction and rate of size deviation;

[0011] Based on the current product size deviation mean, the environmental thermal drift factor and the batch offset slope coefficient, a comprehensive evaluation model is constructed to calculate a deviation severity score value.

[0012] When the score value exceeds a preset threshold, a corresponding camera calibration adjustment scheme is automatically generated according to the change direction of the environmental thermal drift factor and the batch offset slope coefficient, and the POM bracelet rod forming detection is continuously optimized.

[0013] Preferably, the sample detection data set is formed by: performing grayscale and edge enhancement processing on the collected rod images, extracting the rod cross-section edge curve using the Canny edge detection algorithm, and calculating the rod diameter size parameter based on the minimum circumscribed circle fitting method. The calculation result is recorded as a cross-section size feature value and added to the sample detection data set.

[0014] Preferably, the environmental thermal drift factor is calculated by:

[0015] The environmental temperature value and the relative humidity value corresponding to each image acquisition time are obtained in real time.

[0016] Based on the thermal expansion coefficient of the material used by the key components of the detection equipment, a mathematical model reflecting the physical size deviation of the measurement system caused by temperature and humidity changes is constructed. The mathematical model uses a linear thermal expansion formula: environmental thermal drift factor = material expansion coefficient x temperature change value x effective optical path length.

[0017] Preferably, the batch offset slope coefficient is calculated by:

[0018] The cross-section size feature values of each POM rod collected are cached in a sliding data window in time sequence, and the data window length is preset as N sampling periods.

[0019] Each size feature value is subjected to difference operation with the standard size value set by the product to generate a set of deviation sequences, and the sequence represents the small error of each sampling point relative to the target diameter, and is stored in a deviation vector set.

[0020] The deviation vector set is subjected to least square linear fitting to obtain the slope value of the fitting straight line, which is defined as the batch offset slope coefficient.

[0021] Preferably, the deviation severity score value is calculated by:

[0022] arithmetic average processing is performed on N size deviation values in the current sliding window, to obtain a size deviation mean value Δ-, the size deviation value being a difference between a cross-sectional dimension of each product and a standard value;

[0023] The size deviation mean value Δ-, the environmental thermal drift factor ETD and the batch offset slope coefficient BOSC are respectively linearly normalized according to a set normalization interval, to obtain three standardized index values Δ_norm, ETD_norm and BOSC_norm;

[0024] The three normalized indexes are substituted into a comprehensive score function, and weighted summation is performed according to set weight coefficients α, β and γ, to obtain a deviation severity score value Rs, and the calculation formula is: Rs = α × Δ_norm + β × BOSC_norm + γ × ETD_norm, wherein α, β and γ are weight coefficients, and the sum of the three is equal to 1.

[0025] Preferably, the score value Rs is compared with a preset risk level threshold value, and is divided into three levels of low risk, medium risk and high risk, wherein the Rs value less than 0.3 is defined as low risk, between 0.3 and 0.6 is defined as medium risk, and greater than 0.6 is defined as high risk, and accordingly triggers a corresponding early warning or calibration action.

[0026] Preferably, the camera calibration adjustment scheme is automatically generated according to the change direction of the environmental thermal drift factor and the batch offset slope coefficient, including:

[0027] The positive and negative change trends of the current batch offset slope coefficient BOSC and the environmental thermal drift factor ETD are respectively analyzed, if BOSC is a positive value, it indicates that the size deviation increases with the batch, and if ETD is a positive value, it indicates that the current temperature is higher than the reference environmental temperature; according to the trend vector result, the optimal scheme is matched, if ETD is the dominant factor, the structure displacement compensation strategy is preferentially executed; if the BOSC trend is dominant, the image measurement compensation parameter update is preferentially performed.

[0028] In the above technical solution, the technical effects and advantages provided by the present application are:

[0029] 1、The present application realizes quantitative evaluation of the POM hand ring bar size stability by constructing a deviation severity score model taking the size deviation mean value, the environmental thermal drift factor and the batch offset slope coefficient as the core input, breaking through the limitation of traditional quality detection methods relying on single point error judgment and being unable to identify trend deviation. The score model cooperates with the trend vector analysis mechanism, can intelligently judge the deviation source and match the optimal calibration strategy, so that the detection system has self-diagnosis and self-adaptive adjustment capability, and the stability and consistency of the production process are significantly improved.

[0030] 2、The application realizes a data-driven, logic-closed-loop image detection calibration control system through technical means such as parameter normalization, weight configuration, risk level division and control instruction generation, and can complete the whole process from abnormality identification to automatic compensation without relying on manual intervention. The method is particularly suitable for continuous forming products with high detection accuracy requirements, and has good engineering adaptability, system universality and expansion and upgrading capability. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings described below are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0032] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0034] Embodiment, please refer to Figure 1 As shown in the figure, the human skin-friendly POM bracelet rod forming method based on intelligent identification described in the present embodiment comprises:

[0035] The skin-friendly polyformaldehyde raw material is dried, and skin-friendly additives and modifiers are added according to a predetermined proportion to obtain a raw material mixture with skin contact safety;

[0036] The raw material mixture is processed into POM rod material with a predetermined diameter by heating, plasticizing, extruding, shaping and cooling;

[0037] Image acquisition is performed on each section of the rod material, and cross-sectional size features and surface integrity features are extracted to form a sample detection data set;

[0038] The ambient temperature and humidity data corresponding to the image acquisition time are synchronously acquired, and the environmental thermal drift factor is calculated to reflect the potential influence of environmental factors on the measurement system;

[0039] The continuous size features obtained by image acquisition are analyzed in time sequence, and the batch offset slope coefficient is calculated to judge the trend accumulation direction and rate of size deviation;

[0040] Based on the current product size deviation mean, environmental thermal drift factor and batch offset slope coefficient, a comprehensive evaluation model is constructed to calculate the deviation severity score value;

[0041] When the score value exceeds the preset threshold, according to the change direction of the environmental thermal drift factor and the batch offset slope coefficient, the corresponding camera calibration adjustment scheme is automatically generated, and the POM bracelet rod forming detection is continuously optimized.

[0042] POM material is extremely easy to absorb moisture due to the aldehyde end group in its molecular structure. After absorbing moisture, not only does it affect the melt stability of the material, but it can also undergo pyrolysis reaction under high temperature melting conditions, leading to problems such as bubbles, stripes, and size instability during extrusion, which seriously affect the subsequent forming precision and detection accuracy.

[0043] Therefore, before using the raw materials, strict drying treatment must be carried out. The specific drying parameters are as follows:

[0044] The drying temperature is controlled between 80℃ and 100℃;

[0045] The drying time is controlled between 2 to 4 hours;

[0046] The drying method can use hot air circulating drying box or vacuum drying device.

[0047] After the drying process is completed, the moisture content of the raw materials should be controlled below 0.2% to ensure their processing stability.

[0048] In order to improve the skin-friendly performance and thermal comfort of POM materials during wear, skin-friendly additives and modifiers need to be added to the base body in proportion to the quality. The addition content and purpose are as follows:

[0049] Skin-friendly additives are surface-active additives that can form a soft and smooth micro-layer structure on the surface of the formed product. Common additives include polyether ester elastomers, silicone modified polymers, and natural plant-based coating particles.

[0050] Recommended addition ratio: 0.5% to 3% (based on the mass percentage of POM base);

[0051] Addition method: uniformly mix the dried POM particles with a high-speed mixer;

[0052] Function: enhance wear comfort, reduce surface friction coefficient, and reduce skin irritation.

[0053] In order to improve the toughness, thermal stability and extrusion consistency of POM rods, modifiers need to be introduced. Anti-thermal oxidants, light stabilizers and partially copolymerized modified polymers can be used.

[0054] Recommended addition ratio: 1% to 5%.

[0055] The modifier and the POM need to be uniformly dispersed by melt blending;

[0056] Specifically recommended modifiers: copolymerized PBT, SEBS elastomer, antioxidant 1010, etc.

[0057] After the raw materials are mixed, they must be uniformly stirred once and dried twice (60°C for 1 hour) to ensure that all components do not separate or locally aggregate during the subsequent melting process.

[0058] The POM raw materials that have been pretreated and mixed with components enter the second key step of the present application, i.e., rod forming. The rod forming process uses a single-screw extrusion device, which mainly consists of the following processes:

[0059] The pre-mixed raw materials are fed into the extruder barrel through a feeding system and are melted and plasticized under segmented heating control:

[0060] The extrusion temperature zones are set as follows: the front section is 180°C-190°C, the middle section is 195°C-205°C, and the end section is 210°C-215°C.

[0061] The screw rotation speed is 30-50 revolutions per minute.

[0062] The melt viscosity should be maintained within the range of 800-1200 Pa·s to ensure stable forming pressure.

[0063] The POM in the molten state is extruded into an initial rod shape through a sizing die. The rod diameter is controlled to be between 6-12 millimeters, and the inner wall of the die must be made of high-hardness polished alloy material to prevent the melt from adhering and deviating radially.

[0064] A preliminary laser detection link (non-imaging) is provided at this stage to coarsely measure the rod diameter to exclude obvious abnormalities caused by uneven melt or bubbles.

[0065] The formed rod is rapidly cooled through multiple cooling water tanks to prevent internal stress concentration caused by uneven crystal structure, and the cooling water temperature is controlled within the range of 15°C-25°C.

[0066] The rod pulling speed needs to be real-time matched with the screw discharge speed, which is adjusted through a tension sensor to prevent periodic fluctuations in size caused by unstable pulling speed.

[0067] During the pulling process, the system records the difference between the actual pulling speed value and the set value of each meter of rod and calculates the pulling offset index (referred to as PI value). The calculation method is as follows: the actual pulling speed is subtracted from the set speed, and the relative value is taken as the offset ratio. If the continuous offset ratio exceeds ±3%, the pulling feedback adjustment is triggered.

[0068] The image acquisition system is the basic building block of the entire intelligent recognition module, primarily using a high-resolution industrial-grade camera for real-time image acquisition. To ensure the accuracy and consistency of the imaging, geometric distortion correction and pixel ratio calibration of the camera are required to ensure that any unit of length in the image corresponds to the actual physical size.

[0069] Camera parameters:

[0070] Imaging resolution: no less than 2048×2048 pixels;

[0071] Imaging frame rate: no less than 30 frames per second;

[0072] Lens type: Industrial telecentric lens;

[0073] Installation angle: facing the cross-section or side of the POM bar, with an installation error of no more than ±1 degree.

[0074] When the equipment is started, an optical calibration target (including crosshair marks and a multi-point grid calibration plate) is used in conjunction with image calibration software to complete the calibration process. The steps are as follows:

[0075] Place the calibration target at the center of the imaging area;

[0076] Capture calibration images and calculate camera intrinsic parameters using a multi-point perspective correction model;

[0077] Establish a pixel-to-physical length mapping relationship;

[0078] The mapping error between unit pixels and millimeters is controlled to within five per thousand.

[0079] After image acquisition, the image enters the preprocessing module. The purpose is to remove noise, enhance edge information, and improve the accuracy and robustness of subsequent image recognition. The preprocessing steps are as follows:

[0080] Grayscale conversion: Converts RGB images into single-channel grayscale images to reduce computational burden;

[0081] Gaussian filtering: uses a 3×3 Gaussian kernel to smooth the image and remove background and grainy noise;

[0082] Histogram equalization: enhances image contrast and highlights edge contours;

[0083] Edge enhancement: The Laplacian operator is used to enhance the edge contour, which facilitates subsequent feature extraction.

[0084] All processing operations are performed in real time on the embedded image processing module, with processing latency controlled within 20 milliseconds.

[0085] One of the key quality parameters of POM bars is their cross-sectional dimensions (especially diameter or side length). To extract these features, edge detection and geometric fitting algorithms need to be applied based on the processed image. The specific implementation is as follows:

[0086] The Canny edge detection algorithm is used to extract the outline boundary of the bar from the processed image. This algorithm is implemented in the following three steps:

[0087] Calculate the image gradient and perform non-maximum suppression;

[0088] Set high and low thresholds (50 and 150 gray levels respectively) for edge concatenation;

[0089] Output a binary edge image that preserves the complete outline.

[0090] The extracted edge shape is then sized using the minimum circumcircle fitting method, as follows:

[0091] Perform Hough circle detection on the edge point set;

[0092] The center coordinates and radius of the circle were obtained through fitting.

[0093] Multiply the radius by the pixel-to-millimeter mapping ratio to obtain the actual diameter size D.

[0094] The obtained dimensional data is recorded as "cross-sectional dimension characteristic values" and stored in a structured manner with a timestamp and product number. The dimensional measurement error is controlled within ±0.02 mm.

[0095] Besides dimensional inspection, the presence of defects such as scratches, bubbles, indentations, and discoloration on the bar surface is also a key factor affecting product quality. This invention uses a model trained based on a deep convolutional neural network (CNN) to automatically identify surface defects in the bar.

[0096] An improved ResNet-34 network structure is used, with an input image size of 256×256 pixels. The output consists of multi-class defect labels and corresponding confidence scores. Defect categories include:

[0097] Category 1: Scratches;

[0098] Category 2: Bubbles;

[0099] Category 3: Different colored dots;

[0100] Category 4: Surface indentation.

[0101] The model training samples should be no less than 5,000 images, and data augmentation methods should include rotation, brightness perturbation, and noise superposition. The training rounds should be no less than 200.

[0102] Each defect result output includes three metrics:

[0103] Defect type;

[0104] Coordinate region (rectangle);

[0105] Confidence score (range 0 to 1, a value greater than or equal to 0.85 is considered valid).

[0106] The identification results are recorded in the defect data table in JSON format and bound to the corresponding size information and bar number.

[0107] To support subsequent trend-based dimensional deviation analysis and risk level assessment, this invention constructs a standard sample dataset from the image detection results of each bar in a structured manner. Each sample record includes at least the following fields:

[0108] Product number; image acquisition timestamp; cross-sectional dimension feature value (unit: mm); surface defect type (if any); defect confidence score; camera position and parameter record; current ambient temperature and humidity values ​​(for ETD calculation);

[0109] The dataset is saved in time series format. Every 100 bars collected are automatically summarized and packaged, and used in subsequent batch offset trend analysis models.

[0110] A set of high-precision digital temperature and humidity sensors is placed near the image acquisition device installation area (including the industrial camera, lens, bracket, and lighting module). The specific configuration is as follows:

[0111] The temperature sensor should be an NTC thermistor or a digital temperature module (such as SHT31) with a response time of less than 2 seconds.

[0112] The humidity sensor uses a capacitive digital sensor with a measurement range of 0–100%RH and an accuracy of no less than ±2%RH.

[0113] The sensor sampling period is set to once every 30 seconds, and the current sample value is recorded for each image acquisition.

[0114] The sampled data is input to the image recognition control motherboard via IC or UART bus to achieve data synchronization and binding.

[0115] Each time an image is acquired, the system packages and stores the current timestamp and temperature and humidity values ​​to form an "image-environmental state" pairing record, which serves as the basis for subsequent thermal drift calculations.

[0116] The "Environmental Thermal Drift Factor" (ETD) mentioned in this invention refers to the physical compensation amount for the dimensional deviation caused by the thermal expansion of key components of the detection equipment (such as camera lens brackets, lighting components, and measurement reference frames) due to temperature changes.

[0117] This factor is calculated using the following mathematical model: ETD = s × ΔT × L; where: ETD is the environmental thermal drift factor, in millimeters; s is the linear thermal expansion coefficient of the main materials of the equipment structure, in terms of the length change ratio per degree Celsius (e.g., aluminum alloy is approximately 2.4 × 10⁻⁶). -5 / ℃); ΔT is the difference between the current ambient temperature and the reference temperature (the temperature during system calibration), in degrees Celsius; L is the critical optical path length (distance from the center of the camera lens to the imaging reference surface), in millimeters.

[0118] For example: If the camera mount is made of aluminum alloy, the coefficient of thermal expansion is taken as 2.4 × 10⁻⁶. -5 Given a path length of 300 mm and an ambient temperature increase of 5 degrees Celsius, the thermal drift factor is calculated as follows: ETD = 2.4 × 10⁻⁶. -5 ×5×300=0.036 mm.

[0119] The above results indicate that the current image acquisition system may have a dimensional measurement error of up to 0.036 mm due to thermal expansion, which must be corrected or warning measures are taken.

[0120] In the size recognition module, this invention constructs a measurement correction mechanism based on ETD (Earning Tolerance). Whenever the system calculates the ETD value, it automatically overlays and analyzes it with the size deviation value extracted from the current image recognition result (i.e., the difference between the actual recognized size and the standard design value) to calculate the corrected deviation value. The model is defined as follows: Actual Measurement Error = Original Recognition Deviation + ETD; this value will be directly used in the risk level assessment model (see Section 5 of the specification) for risk score calculation, and will trigger a system prompt or automatic calibration process if it exceeds a set threshold.

[0121] The system's thermal drift threshold is set to 0.03 mm. When the ETD exceeds this value, the system will identify it as a period of significant thermal drift impact, requiring the user to be alerted or a local recalibration to be performed.

[0122] To achieve real-time system status maintenance, this invention introduces an environmental deviation early warning mechanism. The system collects environmental data every 30 seconds and performs the following judgment logic:

[0123] If the ETD calculation value exceeds the set threshold (e.g., 0.03 mm) three times in a row, an "abnormal offset" mark will be recorded.

[0124] At the same time, a prompt message pops up on the detection system interface, reminding the user to check the camera position or perform automatic calibration;

[0125] The environmental parameter record file will be updated synchronously to form a three-dimensional tracking matrix of temperature, humidity, time, and ETD, which can be used for subsequent deviation trend analysis or production environment optimization.

[0126] To support continuous trend analysis, the system needs to construct a dimensional feature data caching mechanism to record and maintain the detection dimensional sequence of POM bars in real time. The specific implementation is as follows:

[0127] After each segment of bar material is identified, the system writes the characteristic value of that cross-sectional dimension (in millimeters) into the data cache.

[0128] The cache uses a sliding window mechanism, which means that only the most recent N sample data are retained, where N is the preset window length, the recommended value range is 30 to 200, and the system default value is 100;

[0129] As new samples arrive, the oldest samples are automatically removed, ensuring that the data window reflects the real-time status of the current production batch.

[0130] This mechanism can dynamically adapt to different production rhythms, ensuring that subsequent calculations are representative and do not excessively trace back to historical data.

[0131] After each update of the sliding window, the system performs a difference calculation between all sample data and the standard diameter value D set for the product, resulting in a continuous sequence of dimensional deviations, denoted as the "deviation vector," which is defined as follows:

[0132] Let the data in the window be D1, D2, ..., D;

[0133] The standard size is D0 (e.g., 10.00 mm);

[0134] The deviation vectors are then Δ1 = D1 – D0, Δ2 = D2 – D0, …, Δ = D – D0;

[0135] The resulting Δ sequence is stored in the "part number-deviation" format and bound with a timestamp for subsequent fitting and modeling.

[0136] This deviation vector can intuitively reflect the size drift of each product and serves as the input basis for subsequent fitting slope analysis.

[0137] To extract the trend of dimensional deviations over time, the system uses a least squares linear fitting algorithm to process the deviation vector. This method fits a trend line by using the product number as the independent variable and the deviation value as the dependent variable; the slope of this trend line is the BOSC value, defined as follows:

[0138] Let the deviation vector be (x1,Δ1), (x2,Δ2), ..., (x,Δ), where x is the product number (or collection time index);

[0139] The linear regression equation Δ=a·x+b was used for fitting;

[0140] The obtained slope a is the batch offset slope coefficient BOSC, with the unit of "millimeter per piece".

[0141] A positive BOSC indicates that the deviation increases with the part number, while a negative BOSC indicates a decrease. The larger the absolute value, the more obvious the trend.

[0142] For example, among 100 products, if the fitted slope a is +0.0007 millimeter per piece, it means that for each manufactured product, the diameter increases by an average of 0.0007 millimeter. After accumulating 100 pieces, the increase is 0.07 millimeter, which may exceed the allowable tolerance.

[0143] To determine whether the current BOSC value constitutes a quality risk, the present invention sets a trend offset judgment threshold Tt, whose value is obtained based on the statistics of historical stable production data, as follows:

[0144] The threshold Tt is defined as the limit of the acceptable trend change rate, with the unit of "millimeter per piece";

[0145] It is set according to the process control ability, material characteristics, and detection error tolerance, and generally takes a value of 0.0005 millimeter per piece;

[0146] The judgment rule is as follows:

[0147] If BOSC≥Tt or BOSC≤-Tt, it is determined as "trend offset occurs", and it is necessary to enter the warning or calibration stage;

[0148] If |BOSC|<Tt, it is determined as "stable state", and the system continues to run but the monitoring frequency is increased.

[0149] This threshold is adjustable, allowing the system administrator to optimize the configuration according to the actual product sensitivity.

[0150] A scoring model is constructed, and this scoring model uses the detection parameters based on the following three dimensions as input variables:

[0151] Mean dimension deviation (Δ - ):

[0152] It is defined as the average value of all product cross-sectional dimension deviation values within the current sliding data window;

[0153] Each deviation value is the difference between the actual measured dimension of a single product and the design standard dimension D; <​​​​Since the above three parameters have different physical dimensions and data distribution characteristics, they must be normalized first for fusion in a unified evaluation model. The normalization process uses the linear ratio scaling method to convert all parameters to the range of [0, 1]. Δ_norm = Δ - / Δ_max; BOSC_norm = |BOSC| / BOSC_max; ETD_norm = ETD / ETD_max.

[0156] Where: Δ_max, BOSC_max, and ETD_max are the maximum reference values of the three parameters in the historical qualified data, and the default values are: Δ_max = 0.10 mm; BOSC_max = 0.001 mm / piece; ETD_max = 0.05 mm; if the actual parameter value exceeds the reference maximum value, the normalization result is truncated to 1; taking the absolute value for BOSC is to uniformly process the intensity of the trend without distinguishing the direction.

[0157] After normalization, the three standardized parameters will be input into a unified weighted scoring function. The function is as follows: Rs = α×Δ_norm + β×BOSC_norm + γ×ETD_norm; where: Rs is the final deviation severity score value, and the value range is between 0 and 1; α, β, and γ are the weight coefficients of the mean dimension deviation, trend slope coefficient, and thermal drift factor respectively; the weight coefficients need to satisfy α + β + γ = 1; the default settings are: α = 0.5, β = 0.3, γ = 0.2;

[0158] The weight settings are based on the following principles:

[0159] The mean dimension deviation has the greatest impact on product quality and accounts for a high proportion; the trend deviation has a warning effect and accounts for the second; the environmental drift is an indirect influencing factor and accounts for a relatively low proportion; the weight parameters support user-defined adjustment to adapt to different production environments and quality control strategies.

[0160] To apply the scoring result to actual production control, the system introduces a risk level classification mechanism. The scoring value Rs will correspond to the following three risk levels:

[0161] Low risk: Rs ≤ 0.3, continue to run and enter the next detection cycle. Medium risk: 0.3 < Rs ≤ 0.6, pop-up prompt, suggesting manual observation and data review. High risk: Rs > 0.6, automatically trigger calibration suggestions or system control strategies.

[0162] When the scoring value is at the medium risk level, the system records the current sliding window data and generates a warning log once; if the scores are at the high risk level for three consecutive rounds, the system will automatically call the calibration suggestion module or pause the production line and prompt the operator to intervene.

[0163] To accurately determine the dominant trend and causes of deviations, this invention proposes a symbolic combination analysis method combining ETD and BOSC to construct a trend direction vector. The implementation is as follows:

[0164] Get the BOSC and ETD for the current calculation cycle;

[0165] Extract its sign value (positive or negative) as the dimensional feature of the logical vector;

[0166] Construct the trend vector T_vec, as follows:

[0167] [+1,+1]: indicates that the size continues to increase and the temperature rises;

[0168] [+1,-1]: Indicates that the size increases but the temperature decreases;

[0169] [-1,+1]: Size decreases but temperature increases;

[0170] [-1,-1]: Both size and temperature decrease.

[0171] This trend vector is used for the conditional matching logic of subsequent calibration strategies.

[0172] The system includes several pre-set typical calibration strategy templates, each with a different response approach for different trend combinations and deviation types. Template examples are shown below:

[0173]

[0174] Once the trend vector is identified, the system enters the strategy matching module, automatically selects the calibration strategy that best matches the current trend from the template library, and reads its associated adjustment parameters.

[0175] After selecting a calibration strategy, the system needs to perform the following procedures:

[0176] Calculate the corresponding adjustment amount based on the current magnitudes of ETD and BOSC;

[0177] For example, lens axial compensation = ETD × system magnification factor (e.g., 200 times);

[0178] Software correction factor = BOSC × number of samples;

[0179] Forming structured control instructions, including:

[0180] Control type (e.g., focus adjustment, image scaling, pixel calibration);

[0181] Execution value (e.g., +2 step pulse, scaling 0.98 times);

[0182] Execute module number;

[0183] Commands are sent to the corresponding execution units (such as electric focusing modules and image acquisition software control interfaces) via the system's internal communication bus (such as CAN, Modbus, or custom interfaces);

[0184] The execution result is fed back and verified in the control system. If the adjustment is successful, the status is set to "complete"; otherwise, an error code is generated and recorded in the log.

[0185] After each calibration operation is completed, the system packages and stores the following data: current score value Rs; BOSC and ETD values; trend vector; selected strategy ID; actual execution parameter values; calibration result status (success / failure); execution timestamp and operation number. This data will serve as input samples for the subsequent "calibration strategy knowledge base," used for training the intelligent system and optimizing strategy weights.

[0186] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for forming skin-friendly POM bracelet rods based on intelligent recognition, characterized in that: include: The skin-friendly polyoxymethylene raw material is dried, and skin-friendly additives and modifiers are added in a preset ratio to obtain a raw material mixture that is safe for skin contact. The raw material mixture is processed into POM rods of a predetermined diameter by heating and plasticizing, extrusion shaping, and cooling and traction. Images were acquired for each section of the bar, and its cross-sectional size features and surface integrity features were extracted to form a sample detection dataset. Simultaneously acquire ambient temperature and humidity data corresponding to the image acquisition time, and calculate the environmental thermal drift factor to reflect the potential impact of environmental factors on the measurement system; The continuous dimensional features obtained from image acquisition are analyzed in time series to calculate the batch offset slope coefficient, which is used to determine the trend of cumulative dimensional deviation. Based on the current average product size deviation, environmental thermal drift factor, and batch offset slope coefficient, a comprehensive evaluation model is constructed to calculate the deviation severity score. When the score exceeds the preset threshold, the corresponding camera calibration and adjustment scheme is automatically generated based on the changing direction of the environmental thermal drift factor and the batch offset slope coefficient, continuously optimizing the POM wristband bar forming inspection.

2. The method for forming skin-friendly POM bracelet rods based on intelligent recognition according to claim 1, characterized in that: The process of forming the sample detection dataset includes: performing grayscale and edge enhancement processing on the acquired bar images, extracting the edge curves of the bar cross-section using the Canny edge detection algorithm, and calculating the bar diameter parameters based on the minimum circumcircle fitting method. The calculation results are recorded as cross-sectional dimension feature values ​​and added to the sample detection dataset.

3. The method for forming a skin-friendly POM wristband rod based on intelligent recognition according to claim 1, characterized in that: The calculation of the environmental thermal drift factor includes: Real-time acquisition of ambient temperature and relative humidity values ​​corresponding to each image acquisition moment; Based on the thermal expansion coefficient of the materials used in the key components of the detection equipment, a mathematical model is constructed to reflect the physical dimensional shift of the measurement system caused by temperature and humidity changes. The mathematical model adopts the linear thermal expansion formula: environmental thermal drift factor = material expansion coefficient × temperature change value × effective optical path length.

4. The method for forming a skin-friendly POM wristband rod based on intelligent recognition according to claim 3, characterized in that: The calculation of the batch offset slope coefficient includes: The cross-sectional dimension feature values ​​of each POM bar obtained are cached in a sliding data window in chronological order, and the length of the data window is a preset N sampling periods; The difference between each dimensional feature value and the standard dimensional value set for the product is calculated to generate a set of deviation sequences. The sequence represents the small error of each sampling point relative to the target diameter and is stored in the deviation vector set. The least squares method is used to perform linear fitting on the set of deviation vectors to obtain the slope value of the fitted line, which is defined as the batch offset slope coefficient.

5. The method for forming a skin-friendly POM wristband rod based on intelligent recognition according to claim 4, characterized in that: The calculated severity score for the deviation includes: The arithmetic mean of the N size deviation values ​​within the current sliding window is calculated to obtain the average size deviation Δ-, where the size deviation value is the difference between the cross-sectional dimension of each product and the standard value. The mean dimensional deviation Δ-, the environmental thermal drift factor ETD, and the batch offset slope coefficient BOSC were linearly normalized according to the set normalization interval to obtain three standardized index values: Δ_norm, ETD_norm, and BOSC_norm. Substitute the three normalized indicators into the comprehensive scoring function, and sum them according to the set weight coefficients α, β and γ to obtain the severity score Rs. The calculation formula is: Rs=α×Δ_norm+β×BOSC_norm+γ×ETD_norm, where α, β and γ are weight coefficients, and the sum of the three equals 1.

6. The method for forming a skin-friendly POM wristband rod based on intelligent recognition according to claim 5, characterized in that: The score Rs is compared with the preset risk level threshold and divided into three levels: low risk, medium risk and high risk. An Rs value less than 0.3 is defined as low risk, between 0.3 and 0.6 is defined as medium risk, and greater than 0.6 is defined as high risk. Based on this, corresponding warning or calibration actions are triggered.

7. The method for forming a skin-friendly POM wristband rod based on intelligent recognition according to claim 6, characterized in that: Based on the changing direction of the environmental thermal drift factor and the batch offset slope coefficient, the corresponding camera calibration and adjustment scheme is automatically generated, including: The positive and negative trends of the current batch offset slope coefficient BOSC and the environmental thermal drift factor ETD are analyzed separately. If BOSC is positive, it means that the dimensional deviation increases with the batch. If ETD is positive, it means that the current temperature is higher than the reference ambient temperature. The optimal solution is matched according to the trend vector results. If ETD is the dominant factor, the structural displacement compensation strategy is implemented first. If the BOSC trend is dominant, the image measurement compensation parameters are updated first.

Images