Automatic selection system for laser cutting and cloth paving

Abstract

The invention discloses a laser cutting cloth laying automatic selection system, and belongs to the technical field of automobile safety air bag production equipment, the laser cutting cloth laying automatic selection system comprises a cloth laying dynamic characteristic sensing module used for collecting fabric elasticity recovery rate, tearing strength and other mechanical parameters; the airbag model-fabric characteristic mapping database is used for storing characteristic quantitative standards of different airbags; the AI intelligent selection decision module is used for dynamically adjusting the weight through an improved AHP-TOPSIS algorithm and outputting optimal and alternative paving cloth; the laser cutting-cloth paving state closed-loop cooperation module is used for monitoring wrinkles and stretching in real time and adjusting parameters or switching cloth paving; and the full-life-cycle tracing module is used for constructing a full-process data chain through RFID and a QR code and feeding back an algorithm. Precise matching of fabric mechanical characteristics and airbag safety requirements is achieved, the fabric selecting efficiency and the cutting quality stability are improved, the industrial compliance requirement is met, and the device is suitable for lean production of automobile airbags.

Description

Technical Field

[0001] This invention relates to the field of automotive airbag production equipment technology, and in particular to an automatic selection system for laser cutting and fabric laying. Background Technology

[0002] As a core component of vehicle passive safety, the performance of automotive airbags directly determines the occupant protection effect in a collision. The quality of the airbag fabric selection and the precision of laser cutting are key factors affecting airbag safety performance and production stability. Currently, the technical solutions for fabric selection and laser cutting in automotive airbag production have significant limitations, making it difficult to meet the industry's high requirements for safety, efficiency, and quality control. Specific problems are as follows:

[0003] In existing technologies, fabric selection often relies on the appearance parameters of the fabric for screening, without linking the mechanical properties of the fabric to the safety requirements of the airbag model. However, the airbag's resistance to rupture and its deformation recovery ability during deployment are core indicators to ensure its safety performance. Traditional selection methods ignore key mechanical parameters such as the fabric's elastic recovery rate and tear strength, and determine the fabric to be used based solely on manual visual judgment or simple size comparison. This can easily lead to the misuse of fabrics that meet appearance standards but fail to meet mechanical performance standards, thereby causing safety hazards such as airbag rupture and excessive deformation during deployment, and failing to meet the performance requirements in collision tests of different airbag models.

[0004] Current fabric selection decisions often rely on single-dimensional ranking or fixed-weight calculations, failing to comprehensively consider the synergistic optimization of characteristic matching, inventory adaptability, and historical loss rate. For example, some solutions only focus on inventory clearance needs, selecting fabrics with remaining lengths close to cutting requirements but high historical loss rates, leading to increased cutting waste. Other solutions rely on manual experience to judge fabric suitability, which is not only inefficient but also prone to judgment errors due to differences in operator experience, resulting in the waste of high-quality fabrics or the influx of substandard fabrics into production. At the same time, traditional decision-making lacks a dynamic weight adjustment mechanism, making it impossible to prioritize delivery cycles when facing urgent orders, and making it difficult to balance the dual needs of quality and efficiency. The level of automation is far below the lean requirements of modern automotive parts production.

[0005] In existing technologies, fabric selection and laser cutting are separate processes: once the fabric is selected, the laser cutting parameters are fixed, without considering the dynamic changes in the fabric during the fabric conveying process. In actual production, the fabric is prone to wrinkles and stretching due to uneven tension under the traction of the conveyor rollers. Traditional solutions lack real-time monitoring of these dynamic deformations. Wrinkles can cause uneven distribution of laser energy in the fabric stacking area, resulting in incomplete cutting or rough edges. Stretching can cause deviations between the actual cutting size and the design size. Due to the lack of a coordinated mechanism for state monitoring and parameter adjustment, traditional solutions require subsequent manual quality inspection to screen out defective products, which not only increases quality inspection costs but also reduces the stability of the production cycle.

[0006] Current quality traceability in automotive airbag production is mostly limited to the finished product stage, failing to establish a complete data chain encompassing "fabric characteristics - selection decision - cutting process - finished product information." On one hand, it's impossible to trace the mechanical properties and selection records of the fabric used through the finished product's serial number (SN). When quality problems occur in the finished product, it's difficult to pinpoint the root cause as either substandard fabric characteristics or abnormal cutting parameters. On the other hand, traditional traceability systems lack the accumulation and analysis of data such as historical fabric loss rates and characteristic matching degrees. This makes it impossible to optimize fabric selection strategies based on historical data or provide data support for the iteration of laser cutting parameters. Consequently, process optimization relies on trial and error, resulting in low efficiency and failing to meet the compliance requirements of quality management systems for full-process traceability and data-driven optimization.

[0007] In summary, there is an urgent need for a technical solution that can achieve precise matching of learning characteristics, multi-dimensional intelligent decision-making, closed-loop collaboration in selection and cutting, and full life-cycle traceability, in order to improve the safety performance, production efficiency, and quality control level of airbags. Summary of the Invention

[0008] One objective of this invention is to propose an automatic selection system for laser cutting and fabric laying. This invention can achieve precise matching with airbag models by dynamically sensing the mechanical properties of airbag fabric, and complete multi-dimensional intelligent fabric laying selection with the help of an improved AHP-TOPSIS algorithm. It constructs a closed-loop collaboration between fabric laying status and laser cutting to compensate for dynamic deformation. At the same time, it realizes data feedback and process optimization through a full life cycle traceability chain, thereby improving the safety, automation efficiency and quality control level of airbag production, and meeting the lean production needs of automotive safety components.

[0009] According to an embodiment of the present invention, an automatic selection system for laser cutting and fabric laying includes a fabric laying dynamic characteristic sensing module, an airbag model and fabric characteristic mapping database, an AI intelligent selection decision module, a laser cutting and fabric laying status closed-loop collaborative module, and a full life cycle traceability module.

[0010] The fabric dynamic characteristic sensing module is used to collect the mechanical and physical parameters of the automotive airbag fabric. The mechanical parameters include elastic recovery rate and tear strength, and the physical parameters include texture density, length, and width.

[0011] The airbag model and fabric property mapping database is used to store quantitative standards for fabric properties corresponding to different car airbag models. The quantitative standards include minimum elastic recovery rate, minimum tear strength, and texture density range.

[0012] The AI ​​intelligent selection decision module is communicatively connected to the fabric dynamic characteristic perception module and the mapping database. It is used to select and output the optimal fabric and at least two sets of alternative fabrics from the inventory fabrics based on mechanical characteristic parameters, physical parameters and quantitative standards through a multi-dimensional decision algorithm.

[0013] The laser cutting and fabric laying status closed-loop collaborative module is communicatively connected to the AI ​​intelligent selection decision module, including a fabric laying status monitoring unit and a parameter adjustment unit. The fabric laying status monitoring unit is used to detect the wrinkles and stretching during the fabric laying and conveying process in real time. The parameter adjustment unit is used to trigger the alternative fabric switching and adjust the laser cutting parameters when the wrinkles and stretching exceed the preset threshold.

[0014] The full lifecycle traceability module is connected to the fabric dynamic characteristic perception module, the AI ​​intelligent selection decision module, and the laser cutting and fabric laying status closed-loop collaboration module. It is used to assign a unique identifier to each roll of fabric and bind its full-process data. The full-process data includes warehousing characteristic parameters, selection decision records, cutting process parameters, and finished product association information.

[0015] Furthermore, the fabric dynamic characteristic sensing module includes a miniature pressure sensor array, a laser tear strength tester, a high-resolution industrial camera, and an RFID reader / writer. The miniature pressure sensor array is embedded in the surface of the fabric conveying roller and is used to collect the tension and deformation curves of the fabric under a preset pretension. The laser tear strength tester is used to calculate the fabric tear strength by converting the energy attenuation rate of the laser beam reflection spectrum. The high-resolution industrial camera is used to capture images of the fabric surface texture. The RFID reader / writer is used to bind a unique electronic tag UID to each roll of fabric.

[0016] Furthermore, the fabric dynamic characteristic sensing module also includes a data processing unit, which calculates the elastic recovery rate based on the tension and deformation curve. The calculation formula is:

[0017]

[0018] in, This is the initial length of the fabric. This is the maximum deformation length under a 5N pretension. The length of residual deformation within 3 seconds after unloading;

[0019] An improved Canny edge detection algorithm is used to extract yarn contours from the texture image. The texture density is obtained by counting the number of yarn interlacing points per unit area using Hough transform. The calculation formula is:

[0020]

[0021] in, This represents the total number of yarn interlacing points within a 1cm² area. For statistical purposes, this refers to the area.

[0022] Furthermore, the airbag model and fabric characteristic mapping database uses MySQL 8.0 and the InnoDB engine, including the airbag_model table, the property_standard table, and the fabric_inventory table;

[0023] The airbag_model table stores fields including airbag model identifier model_id, safety level safety_level, and application part, with the application part selected from front, side, and knee airbags;

[0024] The property_standard table stores fields including the minimum elasticity recovery rate. Minimum tear strength and texture density range ;

[0025] The fabric_inventory table stores fields including the unique fabric identifier (UID) and elasticity recovery rate. Tear strength Texture density The remaining length (remaining_length) and the historical clipping loss rate (loss_rate) are also considered.

[0026] Furthermore, the mapping database also includes a dynamic update unit. When a new airbag model is added, the dynamic update unit calculates the fabric stress distribution and reverse elasticity parameter thresholds in the impact test of that airbag model using ABAQUS finite element simulation, including the tear strength threshold. The reverse formula is:

[0027]

[0028] in, The value represents the maximum stress on the fabric during the simulation, and 1.1 is the safety factor.

[0029] The dynamic update unit can also optimize monthly based on the cutting yield data using a gradient descent algorithm. To ensure that the coverage rate of qualified fabrics remains above 98%, the formula for calculating the coverage rate is as follows:

[0030] .

[0031] Furthermore, the multi-dimensional decision-making algorithm of the AI ​​intelligent selection decision module includes three decision indicators: feature matching degree, inventory adaptability, and historical loss rate, with the weights of the three decision indicators being 40%, 30%, and 30%, respectively.

[0032] The matching degree of the characteristics is determined by calculating the Euclidean distance between the fabric parameters and the model standard, using the following formula:

[0033]

[0034] in, This represents the actual elastic recovery rate of the fabric. This represents the minimum elastic recovery rate required by the model. The actual tear strength of the fabric. This is the minimum tear strength required by the model. The actual texture density of the fabric. This is the average value of the texture density range required by the model, and The smaller the value, the higher the match.

[0035] The inventory adaptability assessment prioritizes fabric with a remaining length ≥ 1.2 times the cutting requirement. The adaptability scoring formula is as follows:

[0036]

[0037] in, This refers to the remaining length of the fabric. This refers to the total length of the cut in a single batch. The larger the value, the higher the adaptability;

[0038] The historical loss rate is selected from fabrics with a cutting loss rate of ≤3% in the past 3 months. The loss rate is calculated using the following formula:

[0039]

[0040] in, This is the initial length of the fabric. This represents the total length of all finished products that have been cut from the fabric.

[0041] Specifically, when receiving urgent orders, the weight of inventory suitability is adjusted through a dynamic adjustment factor. Adjusted to 40%, The calculation formula is:

[0042]

[0043] in, For delivery timeframes required by urgent orders, For the standard delivery cycle, .

[0044] Furthermore, the multi-dimensional decision-making algorithm is an improved AHP-TOPSIS algorithm, specifically including the following steps:

[0045] Step 1: Calculate the dynamic weights of the three decision indicators using the Analytic Hierarchy Process (AHP), and the weight vector. ,in For feature matching degree weight, For inventory fit weight, Weighted by historical loss rate, and ;

[0046] Step 2: Normalize the indicator data:

[0047] Feature matching degree normalization:

[0048]

[0049] in, For the first The Euclidean distance of the rolled fabric. This represents the maximum Euclidean distance among all fabrics. It is the minimum value;

[0050] Inventory fit normalization:

[0051]

[0052] in, For the first Inventory fit rating for roll fabrics. To achieve the highest score;

[0053] Historical loss rate normalization:

[0054]

[0055] in, For the first The loss rate of rolled fabric. For the maximum loss rate, It is the minimum value;

[0056] Step 3: Construct the weighted normalized matrix ,in ;

[0057] Step 4: Determine the ideal solution and negative ideal solution :

[0058]

[0059]

[0060] Step 5: Calculate the closeness of each fabric layer to the ideal solution. :

[0061]

[0062] in, For the first The Euclidean distance between the rolled fabric and the ideal solution. The Euclidean distance to the negative ideal solution;

[0063] Step 6: Press Sort the layouts from largest to smallest, and output the top 3 as the optimal and alternative layouts.

[0064] Furthermore, the fabric laying status monitoring unit of the laser cutting and fabric laying status closed-loop collaborative module includes a high-speed linear array camera and a laser contour sensor.

[0065] The high-speed linear array camera is mounted 50cm upstream of the cutting station to photograph the fabric surface and identify wrinkles and wrinkle heights by measuring changes in grayscale values. The calculation formula is:

[0066]

[0067] in, This represents the grayscale difference between the wrinkled and smooth areas. The average gray value of the flat area. Define the camera calibration coefficient. This is abnormal;

[0068] The laser profile sensor is used to measure the stretch in the fabric width direction. The calculation formula is:

[0069]

[0070] in, The width of the fabric to be laid out in real time. This refers to the nominal width of the fabric as specified by the manufacturer.

[0071] The parameter adjustment unit includes a servo motor, a tension controller, and a carbon dioxide laser cutting head.

[0072] Furthermore, the closed-loop logic of the laser cutting and fabric laying state closed-loop collaborative module is as follows:

[0073] When the line scan camera detects the wrinkle height Or the laser contour sensor detects Furthermore, after three consecutive sampling cycles, the alternative fabric laying method is switched, and the conveying tension is adjusted to the standard value via the servo motor. The tension adjustment formula is:

[0074]

[0075] in, This is the tension adjustment coefficient. This is the real-time tension value;

[0076] When detected or At this time, only the laser cutting parameters are adjusted:

[0077] Laser speed in wrinkled scenes Laser power ;

[0078] In stretching scenarios: Cutting length compensation Actual cutting length ,in, To preset the cutting speed, To preset the laser power, To design the cutting length.

[0079] Furthermore, the end-to-end data chain of the full lifecycle traceability module includes:

[0080] Data during the warehousing stage: UID, batch number, characteristic parameters, and supplier information written via RFID tags;

[0081] Select stage data: Select decision time, AI algorithm version, optimal layout and alternative layout list;

[0082] Cutting stage data: cutting start / end time, actual laser parameters, number of finished products, and waste length, including waste length. , This refers to the total length of fabric taken from the fabric in this batch;

[0083] Finished product stage data: Fabric UID and airbag finished product SN are bound by a QR code, and the QR code contains production date and workstation information;

[0084] The traceability module supports both forward and reverse traceability, and updates the correlation model between characteristics and loss rate based on the traceability data using a random forest algorithm. The model output is the predicted loss rate. The calculation formula is:

[0085]

[0086] in, For the first The weights of each decision tree For the first The model predicts the loss rate using a decision tree, and then feeds back into the AI ​​intelligent selection decision module to optimize the weight calculation of historical loss rate indicators.

[0087] The beneficial effects of this invention are:

[0088] 1. In this invention, a fabric dynamic characteristic sensing module is used to collect tension and deformation curves using a micro pressure sensor array to calculate the elastic recovery rate, and a laser tear strength tester is used to convert the tear strength through spectral energy attenuation rate. Combined with the quantitative standards of different airbags in the airbag model and fabric characteristic mapping database, the mechanical properties of the fabric and the safety requirements of the airbag are accurately matched. This design fundamentally eliminates the selection of fabrics with substandard mechanical properties, ensuring that the performance of each roll of fabric is highly consistent with the safety requirements of the corresponding airbag model, and significantly improving the safety and reliability of the airbag during use.

[0089] 2. The AI ​​intelligent selection decision module in this invention adopts an improved AHP-TOPSIS algorithm. It dynamically adjusts the weights of characteristic matching degree, inventory adaptability, and historical loss rate through the analytic hierarchy process. Combined with the TOPSIS algorithm, it calculates the closeness between the fabric layout and the ideal solution, thereby achieving multi-objective optimal decision-making. This algorithm not only overcomes the one-sidedness of traditional decision-making, such as avoiding the selection of high-loss fabrics through historical loss rate indicators and reducing waste generation, but also replaces manual intervention with fully automated calculation, avoiding human experience errors, significantly improving the efficiency of fabric selection and the scientific nature of resource allocation, and reducing material waste in the production process.

[0090] 3. The closed-loop collaborative module for laser cutting and fabric laying in this invention collects the dynamic status of the fabric laying in real time through a high-speed linear array camera and a laser contour sensor, and feeds the data back to the parameter adjustment unit. If wrinkles or stretching exceed the threshold, the alternative fabric laying is switched. If there is a slight abnormality, the laser parameters are adjusted accordingly. This closed-loop mechanism of real-time monitoring, dynamic feedback and precise adjustment breaks the disconnect between selection and cutting, compensates for the error caused by fabric deformation in real time, ensures the consistency of the flatness and dimensional accuracy of the cutting edge, reduces the generation of defective products and reduces the cost of subsequent quality inspection.

[0091] 4. The full lifecycle traceability module in this invention binds the characteristic parameters of the fabric upon warehousing and the AI ​​selection record to RFID tags, and combines QR codes to associate laser cutting parameters with the finished product serial number, constructing a complete data chain of fabric characteristics, selection decisions, cutting process, and finished product information. This mechanism not only achieves bidirectional traceability in both forward and reverse directions, but also updates the characteristic and loss rate correlation model based on traceability data, feeding back into the AI ​​selection decision module to adjust indicator weights, forming a virtuous cycle of data collection, traceability analysis, and process optimization. This not only meets industry compliance requirements, but also provides data support for continuous improvement of production processes, thereby enhancing the overall quality control level. Attached Figure Description

[0092] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0093] Figure 1 This is a schematic diagram of the overall framework of an automatic selection system for laser cutting and fabric laying proposed in this invention;

[0094] Figure 2 This is a flowchart of a multi-dimensional decision-making algorithm for an automatic selection system for laser cutting and fabric laying proposed in this invention. Detailed Implementation

[0095] To make the technical means and objectives and effects of the present invention easier to understand, the present invention will be further described in detail below with reference to specific embodiments.

[0096] like Figure 1-2 As shown, this embodiment provides an automatic selection system for laser cutting and fabric laying of automotive airbags. This system enables intelligent selection and collaborative cutting of fabric throughout the entire airbag production process, from warehousing to cutting, thereby improving production efficiency and product quality. Specifically, the system includes a fabric laying dynamic characteristic sensing module, an airbag model and fabric characteristic mapping database, an AI intelligent selection decision module, a laser cutting and fabric laying status closed-loop collaboration module, and a full lifecycle traceability module. These modules work together to complete the automatic selection and cutting process of the fabric.

[0097] The fabric dynamic characteristic sensing module is used to collect the mechanical and physical parameters of automotive airbag fabric. The mechanical parameters include elastic recovery rate and tear strength, while the physical parameters include texture density, length, and width.

[0098] The hardware of this module includes a miniature pressure sensor array, a laser tear strength tester, a high-resolution industrial camera, and an RFID reader.

[0099] Specifically, the miniature pressure sensor array uses an FSS1500N sensor with an accuracy of ±0.01N. It needs to be embedded in three evenly distributed circumferential positions on the fabric feeding conveyor roller. Each sensor is flush with the surface of the conveyor roller to avoid scratching during fabric conveying. It is used to collect the tension and deformation curves of the fabric under a preset pretension of 5N. The sensor connection line uses a shielded cable, which passes through the hollow shaft inside the conveyor roller and connects to the data acquisition card. The sampling frequency is set to 1kHz to ensure accurate capture of the deformation changes of the fabric during the stress process.

[0100] The laser tear strength tester is installed 1.2m above the inbound conveyor path. The distance between its laser emitter and the edge of the fabric must be fixed at 50mm. Before testing, it must be calibrated with a standard tear strength sample. By adjusting the spectral receiving sensitivity of the tester, the testing error can be controlled within ≤0.5N. Zero-point calibration must be performed once before production each day to avoid the impact of ambient temperature changes on the testing accuracy.

[0101] High-resolution industrial cameras require 8mm focal length lenses with the lens axis at a 30° angle to the fabric surface to prevent fabric reflections from blurring the texture image. A polarizing filter is installed in front of the camera to filter out more than 90% of stray light. The texture images captured by the camera need to be transmitted to an industrial control computer in real time. An improved Canny edge detection is implemented using the OpenCV open-source library. Specifically, the image is first subjected to a 5×5 Gaussian filter to remove noise, then the edge detection threshold is set to [50, 150]. Finally, the yarn interlacing points within a 1cm² range are counted using Hough transform. During the count, invalid areas within 2cm of the fabric edge must be excluded to ensure texture density. The detection accuracy reaches ±5 points / cm²;

[0102] The RFID reader is installed at the inlet end of the inbound conveyor roller. The distance between the reader antenna and the end cover of the fabric roll is controlled at 10-15cm. The RFID tag for each roll of fabric needs to be pasted on the inside of the end cover of the roll. The UID written on the tag adopts the format of "year + batch + serial number", such as 202505-F008, and is also associated with the initial test report provided by the fabric supplier.

[0103] The fabric dynamic characteristic sensing module also includes a data processing unit, which calculates the elastic recovery rate based on the tension and deformation curves collected by the micro pressure sensor array. The calculation formula is:

[0104]

[0105] in, This is the initial length of the fabric. This is the maximum deformation length under a 5N pretension. This represents the residual deformation length within 3 seconds after unloading; abnormal data points need to be removed during the calculation.

[0106] Meanwhile, the data processing unit uses an improved Canny edge detection algorithm to extract yarn contours from texture images captured by an industrial camera. First, the image is subjected to a 5×5 Gaussian filter. Then, the edge detection threshold is set to [50, 150]. Finally, the yarn interlacing points within a 1cm² range are statistically analyzed using a Hough transform. During the statistical analysis, invalid areas within 2cm of the fabric edge are excluded to obtain the texture density. Its accuracy is ±5 points / cm², and the calculation formula is:

[0107]

[0108] in, This represents the total number of yarn interlacing points within a 1cm² area. For statistical purposes, this refers to the area.

[0109] The airbag model-to-fabric characteristic mapping database uses MySQL 8.0 and the InnoDB engine, deployed on an industrial server in the workshop, to store the quantitative standards of fabric characteristics corresponding to different automotive airbag models. The database requires master-slave backup to prevent data loss and includes three main tables:

[0110] The airbag_model table stores fields including model_id (airbag model identifier), safety_level (safety level), application (application location, selected from front, side, and knee airbags), design_length (single-piece design length, such as 800mm), and design_width (single-piece design width, such as 600mm).

[0111] The `property_standard` table stores fields including... (Minimum elastic recovery rate) (Minimum tear strength) (Texture density range), max_thickness (maximum thickness, e.g., 0.5mm);

[0112] The fabric_inventory table stores fields including UID (unique identifier for fabric placement). (Elastic recovery rate) (Tear strength) (Texture density), remaining_length, loss_rate, storage_location (e.g., shelf number 3 in area A), in_stock_time (e.g., 2025-05-08 09:15:32).

[0113] The mapping database also includes a dynamic update unit. When a new airbag model is added, a deployment simulation model of that airbag model is created using ABAQUS 2023 finite element software.

[0114] The collision speed was set at 50 km / h, the airbag deployment time was 30 ms, and the fabric material parameters were obtained based on actual sample testing; the maximum stress value of the fabric was obtained through simulation. According to the tear strength threshold The inverse formula Calculated Rounded down to 38N;

[0115] in, The value represents the maximum stress on the fabric during the simulation, and 1.1 is the safety factor.

[0116] Elastic recovery rate Through 100 physical unfolding tests, it was determined that when When the airbag coverage rate is ≥92%, there is no permanent deformation after deployment, meeting the usage requirements. Additionally, it automatically executes on the 1st of each month. Optimize tasks:

[0117] Extract the cutting data of all AB-S202 models from last month and statistically analyze the differences. The yield rate corresponding to the value:

[0118] The yield is highest when the density is 125-135 points / cm². When the density is <125 points / cm², the yield drops to 97.5%. When the density is >135 points / cm², the yield rate is 98.3%.

[0119] Meanwhile, based on monthly cutting yield data, optimization is performed using the gradient descent algorithm. To ensure that the coverage rate of qualified fabrics remains above 98%, the formula for calculating the coverage rate is as follows:

[0120]

[0121] After optimization, a total of 118 rolls of fabric met the criteria for the month, with a total inventory of 120 rolls, representing a coverage rate of 98.33%.

[0122] The AI-powered intelligent selection and decision-making module communicates with the fabric dynamic characteristic perception module and the mapping database. Based on the collected mechanical and physical parameters and the quantitative standards in the mapping database, it uses a multi-dimensional decision-making algorithm to select and output the optimal fabric layout and a list of at least two alternative fabric layouts from the inventory. The module runs on an industrial control computer, developed in a Python 3.9 environment, and uses the Scikit-learn 1.2.2 library to implement the algorithm, with a computation cycle of ≤2 seconds. The module's interaction with the database uses the SQLAlchemy ORM framework, avoiding direct manipulation of database tables and improving security.

[0123] This multi-dimensional decision-making algorithm includes three decision metrics:

[0124] The characteristic matching degree, with a weight of 40%, is determined by calculating the Euclidean distance between the fabric parameters and the model standard. The formula is as follows:

[0125]

[0126] in, This represents the actual elastic recovery rate of the fabric. This represents the minimum elastic recovery rate required by the model. The actual tear strength of the fabric. This is the minimum tear strength required by the model. The actual texture density of the fabric. This is the average value of the texture density range required by the model, and The smaller the value, the higher the matching degree; during calculation, each parameter needs to be dimensionless.

[0127] Inventory suitability, weighted at 30%, prioritizes fabric with remaining length ≥ 1.2 times the cutting requirement. The suitability scoring formula is:

[0128]

[0129] in, This refers to the remaining length of the fabric. This refers to the total length of the cut in a single batch. The larger the value, the higher the adaptability; if If so, then the fabric should be excluded.

[0130] Historical loss rate, weighted at 30%, selects fabrics with a cutting loss rate ≤3% in the past 3 months. The loss rate calculation formula is as follows:

[0131]

[0132] in, This is the initial length of the fabric. This represents the total length of all finished products that have been cut from this fabric. If there is no historical cutting data for a particular fabric, the default wastage rate is 2.5%.

[0133] When receiving urgent orders, the weight of inventory suitability is adjusted using a dynamic adjustment factor. The weighting has been adjusted to 40%, and the feature matching score weight has been reduced to 30%. The calculation formula is:

[0134]

[0135] in, For delivery timeframes required by urgent orders, For the standard delivery cycle, .

[0136] The multi-dimensional decision-making algorithm is specifically the improved AHP-TOPSIS algorithm, and its steps are as follows:

[0137] Step 1: Calculate the dynamic weights of the three decision indicators using the Analytic Hierarchy Process (AHP) and construct the judgment matrix to obtain the weight vector. ,in For feature matching degree weight, For inventory fit weight, Weighted by historical loss rate, and Taking a regular order as an example, .

[0138] Step 2: Normalize the indicator data, specifically positive normalization, where a larger value is better. The formula for normalizing the feature matching degree is:

[0139]

[0140] in, For the first The Euclidean distance of the rolled fabric. This represents the maximum Euclidean distance among all fabrics. It is the minimum value; if ,but .

[0141] The formula for inventory fit normalization is:

[0142]

[0143] in, For the first Inventory fit rating for roll fabrics. To achieve the highest score;

[0144] The formula for normalizing historical loss rates is:

[0145]

[0146] in, For the first The loss rate of rolled fabric. For the maximum loss rate, It is the minimum value; if ,but .

[0147] Step 3, construct the weighted normalized matrix ,in ; matrix dimension is Each element Indicates the first roll of fabric in the first... Weighted scores on each indicator;

[0148] Step 4: Determine the ideal solution and negative ideal solution :

[0149] Ideal solution The maximum weighted score for each indicator is:

[0150]

[0151] Negative ideal solution The minimum weighted score for each indicator is:

[0152]

[0153] Step 5: Calculate the closeness of each fabric layer to the ideal solution. :

[0154]

[0155] in, For the first The Euclidean distance between the rolled fabric and the ideal solution. ;

[0156] The Euclidean distance to the negative ideal solution, ;

[0157] The closer the value is to 1, the better the fabric is.

[0158] Step 6: Press Sort the fabrics from largest to smallest, and output the top 3 as the optimal and alternative fabrics; if there are fewer than 3 candidate fabrics, output all of them and prompt "Insufficient stock, replenishment required". This is based on the AB-S202 side airbag cutting task (500 pieces) on May 20, 2025. For example, there are 5 candidate fabric rolls, with the following parameters:

[0159] Volume 1 (UID=202505-F008, =93%, =40N, =130 points / cm², remaining_length=450m, loss_rate=2.5%), Volume 2 (UID=202505-F009, =92.5%, =39N, =128 points / cm², remaining_length=420m, loss_rate=2.8%), Volume 3 (UID=202505-F010, =94%, =41N, =132 points / cm², remaining_length=380m, loss_rate=2.2%, Volume 4 (UID=202505-F011, =92%, =38.5N, =126 points / cm², remaining_length=480m, loss_rate=3.0%), Volume 5 (UID=202505-F012, =93.5%, =39.5N, =131 points / cm², remaining_length=410m, loss_rate=2.6%).

[0160] Volume 3 was excluded because remaining_length = 380m < 1.2 × 400m = 480m;

[0161] Calculate the remaining 4 volumes ( =130 points / cm²):

[0162] Volume 1 ;

[0163] Volume 2 Volume 4 Volume 5 , , Normalized Volume 1 Volume 2 ≈ 0.428, Volume 4 ≈ 0.648, Volume 5 ≈ 1;

[0164] Inventory fit rating Volume 1 = 450 / 480 ≈ 0.9375, Volume 2 = 420 / 480 = 0.875, Volume 4 = 480 / 480 = 1, Volume 5 = 410 / 480 ≈ 0.8542. After normalization, Volume 1 ≈ 0.9375, Volume 2 ≈ 0.875, Volume 4 = 1, Volume 5 ≈ 0.8542.

[0165] Historical loss rate normalization Volume 1 = 1, Volume 2 = 0.4, Volume 4 = 0, Volume 5 = 0.8;

[0166] Weighted normalized matrices: Volume 1 = [0.4×0, 0.3×0.9375, 0.3×1] = [0, 0.28125, 0.3], Volume 2 = [0.4×0.428, 0.3×0.875, 0.3×0.4] = [0.1712, 0.2625, 0.12], Volume 4 = [0.4×0.648, 0.3×1, 0.3×0] = [0.2592, 0.3, 0], Volume 5 = [0.4×1, 0.3×0.8542, 0.3×0.8] = [0.4, 0.2563, 0.24];

[0167] Ideal solution Negative ideal solution ;

[0168] Proximity Volume 5 ≈ 0.864, Volume 1 ≈ 0.512, Volume 4 ≈ 0.498, Volume 2 ≈ 0.489;

[0169] The final ranking is Volume 5 (best), Volume 1 (alternate 1), and Volume 4 (alternate 2).

[0170] The laser cutting and fabric laying status closed-loop collaborative module is connected to the AI ​​intelligent selection decision module, including a fabric laying status monitoring unit and a parameter adjustment unit.

[0171] The fabric laying status monitoring unit is used to detect the wrinkles and stretching during the fabric laying and conveying process in real time. Its hardware includes a high-speed line scan camera and a laser contour sensor. The high-speed line scan camera has a frame rate of 120fps and a resolution of 4096 pixels. It is mounted on an aluminum profile bracket 50cm upstream of the cutting station, with the lens 80cm vertically to the fabric surface. The camera trigger signal is synchronized with the conveyor roller encoder, triggering one image capture every 1mm of fabric conveyed. The images captured by the camera are transmitted to the image processor via a dedicated cable. Wrinkles are identified by comparing grayscale values. Wrinkles are identified by the reduced grayscale value due to fabric stacking, indicating the wrinkle height. The calculation formula is:

[0172]

[0173] in, This represents the grayscale difference between the wrinkled and smooth areas. The average gray value of the flat area. Define the camera calibration coefficient. This is abnormal;

[0174] Secondly, the laser contour sensor has a scanning frequency of 5kHz, a measurement range of 0-100mm, an accuracy of ±0.05mm, and is mounted on the same side as the camera, at a horizontal distance of 10cm from the fabric edge. The scanning direction is perpendicular to the fabric conveying direction. The sensor outputs real-time width data and stretching amount via an RS485 bus. The calculation formula is:

[0175]

[0176] in, The width of the fabric to be laid out in real time. The width is the nominal width of the fabric as stated by the manufacturer. During testing, the rough edges of the fabric must be excluded, and the effective width is determined by the edge detection algorithm built into the sensor.

[0177] In addition, the parameter adjustment unit includes a servo motor, a tension controller, and a CO2 laser cutting head;

[0178] Specifically, the servo motor has a torque of 1.5 N·m, a rated speed of 3000 rpm, a response time of ≤10 ms, and is connected to the fabric conveying roller through a reducer. The motor driver receives control commands through the EtherCAT bus.

[0179] The tension sensors of the tension controller are installed at both ends of the driven roller to collect the conveying tension in real time and feed it back to the servo motor driver, forming a tension closed-loop control.

[0180] The CO2 laser cutting head has a power range of 50-300W and a spot diameter of 0.1mm. It features automatic focus tracking and uses a laser displacement sensor to detect the fabric surface height in real time, adjusting the Z-axis position of the cutting head to ensure that the focus is always located 0.1mm below the fabric surface. The cutting head is mounted on the linear module of the gantry.

[0181] The closed-loop logic of the laser cutting and fabric laying state closed-loop collaborative module is as follows:

[0182] The system collects the fold height every 10ms. and stretching amount If the abnormal conditions are met for three consecutive sampling periods, an adjustment will be triggered.

[0183] When the line scan camera detects the wrinkle height Or the laser contour sensor detects When the system has been sampling for three consecutive cycles, it triggers a switch to the alternative fabric laying method. First, a signal is sent to the laser cutting head to cut off the laser output, and the servo motor brakes simultaneously. Then, the system sends a command to the fabric laying library to start the conveyor roller of alternative fabric laying method 1, and the tension controller adjusts the tension from the current value to the standard value. The tension adjustment formula is:

[0184]

[0185] in, This is the tension adjustment coefficient. This is the real-time tension value;

[0186] After the first adjustment The second adjustment The third adjustment The fourth adjustment The fifth adjustment The laser cutting head eventually stabilized at 5N within 2 seconds; after the alternative fabric was conveyed to the cutting station, the laser cutting head resumed operation.

[0187] When detected or At this time, only the laser cutting parameters are adjusted:

[0188] Laser speed in wrinkled scenes Laser power ;

[0189] In stretching scenarios: Cutting length compensation Actual cutting length ,in, To preset the cutting speed, To preset the laser power, To design the cutting length.

[0190] The full lifecycle traceability module communicates with the fabric laying dynamic characteristic perception module, the AI ​​intelligent selection decision module, and the laser cutting and fabric laying status closed-loop collaboration module to assign a unique identifier to each roll of fabric and bind its full-process data.

[0191] The module software is integrated into the workshop MES system and interacts with each module through a WebService interface. Operators can query and trace information through the workshop touch screen terminal.

[0192] This end-to-end data chain includes: data from the warehousing stage, which is automatically written to RFID tags via RFID readers, including UID, batch number, characteristic parameters, supplier information, and warehousing time;

[0193] Select stage data, which is automatically uploaded to the MES system by the AI ​​module, including the selection decision time, AI algorithm version, optimal deployment UID, alternative 1 UID, alternative 2 UID, and decision basis;

[0194] Data during the cutting stage is recorded in real time by the laser cutting module, including cutting start time, end time, actual laser parameters, number of finished products, and L-value usage. , where the length of the waste material is... , This refers to the total length of fabric taken from the fabric in this batch;

[0195] The finished product stage data is printed into QR codes using a QR code printer. Each finished product corresponds to a unique serial number (SN). The QR code contains information including the SN, fabric UID, production date, workstation information, and operator number. The QR code is affixed to the label bag inside the finished airbag.

[0196] The traceability module supports both forward and reverse traceability. In forward traceability, scanning the finished product SN allows you to query the corresponding fabric UID, characteristic parameters, cutting parameters, and supplier reports. In reverse traceability, entering the fabric UID allows you to query a list of all finished product SNs produced using that fabric, the quality inspection results of each finished product, and the cutting loss rate.

[0197] Furthermore, the module updates the correlation model between characteristics and loss rate based on traceability data using a random forest algorithm. The model employs 100 decision trees, with the input features being... , , The output is the predicted loss rate. The calculation formula is:

[0198]

[0199] in, For the first The weights of each decision tree For the first The predicted loss rate of each decision tree is trained monthly using newly added traceable data, reducing the prediction error from the initial ±0.5% to ±0.2%. This model then feeds back into the AI ​​intelligent selection decision module to optimize the weight calculation of historical loss rate indicators, such as when... If the predicted loss rate is >3% when the density is <125 points / cm², the AI ​​module will reduce the weight of the historical loss rate for this type of fabric in subsequent selections, prioritizing the selection of fabrics with higher loss rates. Value fabric.

[0200] In addition, the system is also equipped with a preprocessing unit and a security protection unit:

[0201] The pre-processing unit is a fabric dust removal device before the fabric is put into storage, to prevent dust from affecting the detection accuracy of the sensors;

[0202] The safety protection unit includes a safety light curtain in the laser cutting area, which immediately cuts off the laser output and servo motor power when a person or object enters the cutting area;

[0203] Each module is equipped with a red emergency stop button with a response time of ≤100ms to ensure operator safety.

[0204] The system's routine maintenance cycle is as follows: the miniature pressure sensor array is calibrated with standard weights every 7 days; the laser tear strength tester is calibrated with standard samples every 30 days; the servo motor is lubricated with Shell Alvania EP2 grease every 300 hours; the laser cutting head is calibrated for focal position every 500 hours; and the RFID reader's antenna surface is cleaned every 15 days.

[0205] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An automatic selection system for laser-cut fabric laying, characterized in that, It includes a fabric dynamic characteristic perception module, an airbag model and fabric characteristic mapping database, an AI intelligent selection decision module, a laser cutting and fabric laying status closed-loop collaboration module, and a full life cycle traceability module; The fabric dynamic characteristic sensing module is used to collect the mechanical and physical parameters of the automotive airbag fabric. The mechanical parameters include elastic recovery rate and tear strength, and the physical parameters include texture density, length, and width. The airbag model and fabric property mapping database is used to store quantitative standards for fabric properties corresponding to different car airbag models. The quantitative standards include minimum elastic recovery rate, minimum tear strength, and texture density range. The AI ​​intelligent selection decision module is communicatively connected to the fabric dynamic characteristic perception module and the mapping database. It is used to select and output the optimal fabric and at least two sets of alternative fabrics from the inventory fabrics based on mechanical characteristic parameters, physical parameters and quantitative standards through a multi-dimensional decision algorithm. The laser cutting and fabric laying status closed-loop collaborative module is communicatively connected to the AI ​​intelligent selection decision module, including a fabric laying status monitoring unit and a parameter adjustment unit. The fabric laying status monitoring unit is used to detect the wrinkles and stretching during the fabric laying and conveying process in real time. The parameter adjustment unit is used to trigger the alternative fabric switching and adjust the laser cutting parameters when the wrinkles and stretching exceed the preset threshold. The full lifecycle traceability module is connected to the fabric dynamic characteristic perception module, the AI ​​intelligent selection decision module, and the laser cutting and fabric laying status closed-loop collaboration module. It is used to assign a unique identifier to each roll of fabric and bind its full-process data. The full-process data includes warehousing characteristic parameters, selection decision records, cutting process parameters, and finished product association information.

2. The automatic selection system for laser cutting and fabric laying according to claim 1, characterized in that, The fabric laying dynamic characteristic sensing module includes a miniature pressure sensor array, a laser tear strength tester, a high-resolution industrial camera, and an RFID reader / writer. The miniature pressure sensor array is embedded in the surface of the fabric laying conveyor roller and is used to collect the tension and deformation curves of the fabric under a preset pretension. The laser tear strength tester is used to calculate the fabric tear strength by converting the energy attenuation rate of the laser beam reflection spectrum. The high-resolution industrial camera is used to capture images of the fabric surface texture. The RFID reader / writer is used to bind a unique electronic tag UID to each roll of fabric.

3. The automatic selection system for laser cutting and fabric laying according to claim 2, characterized in that, The fabric dynamic characteristic sensing module also includes a data processing unit, which calculates the elastic recovery rate based on the tension and deformation curve. The calculation formula is: in, This is the initial length of the fabric. This is the maximum deformation length under a 5N pretension. The length of residual deformation within 3 seconds after unloading; An improved Canny edge detection algorithm is used to extract yarn contours from the texture image. The texture density is obtained by counting the number of yarn interlacing points per unit area using Hough transform. The calculation formula is: in, This represents the total number of yarn interlacing points within a 1cm² area. For statistical purposes, this refers to the area.

4. The automatic selection system for laser cutting and fabric laying according to claim 1, characterized in that, The airbag model and fabric property mapping database uses MySQL 8.0 and the InnoDB engine, and includes the airbag_model table, property_standard table, and fabric_inventory table; The airbag_model table stores fields including airbag model identifier model_id, safety level safety_level, and application part, with the application part selected from front, side, and knee airbags; The property_standard table stores fields including the minimum elasticity recovery rate. Minimum tear strength and texture density range ; The fabric_inventory table stores fields including the unique fabric identifier (UID) and elasticity recovery rate. Tear strength Texture density The remaining length (remaining_length) and the historical clipping loss rate (loss_rate) are also considered.

5. The automatic selection system for laser cutting and fabric laying according to claim 4, characterized in that, The mapping database also includes a dynamic update unit. When a new airbag model is added, the dynamic update unit uses ABAQUS finite element simulation to calculate the fabric stress distribution and anti-tear stress parameter thresholds in the impact test of that airbag model, including the tear strength threshold. The reverse formula is: in, The value represents the maximum stress on the fabric during the simulation, and 1.1 is the safety factor. The dynamic update unit can also optimize monthly based on the cutting yield data using a gradient descent algorithm. To ensure that the coverage rate of qualified fabrics remains above 98%, the formula for calculating the coverage rate is as follows: 。 6. The automatic selection system for laser cutting and fabric laying according to claim 1, characterized in that, The multi-dimensional decision-making algorithm of the AI ​​intelligent selection decision module includes three decision indicators: feature matching degree, inventory adaptability, and historical loss rate, with the weights of the three decision indicators being 40%, 30%, and 30%, respectively. The matching degree of the characteristics is determined by calculating the Euclidean distance between the fabric parameters and the model standard, using the following formula: in, This represents the actual elastic recovery rate of the fabric. This represents the minimum elastic recovery rate required by the model. The actual tear strength of the fabric. This is the minimum tear strength required by the model. The actual texture density of the fabric. This is the average value of the texture density range required by the model, and The smaller the value, the higher the match. The inventory adaptability assessment prioritizes fabric with a remaining length ≥ 1.2 times the cutting requirement. The adaptability scoring formula is as follows: in, This refers to the remaining length of the fabric. This refers to the total length of the cut in a single batch. The larger the value, the higher the adaptability; The historical loss rate is selected from fabrics with a cutting loss rate of ≤3% in the past 3 months. The loss rate is calculated using the following formula: in, This is the initial length of the fabric. This represents the total length of all finished products that have been cut from the fabric. Specifically, when receiving urgent orders, the weight of inventory suitability is adjusted through a dynamic adjustment factor. Adjusted to 40%, The calculation formula is: in, For delivery timeframes required by urgent orders, For the standard delivery cycle, .

7. The automatic selection system for laser cutting and fabric laying according to claim 6, characterized in that, The multi-dimensional decision-making algorithm is an improved AHP-TOPSIS algorithm, which specifically includes the following steps: Step 1: Calculate the dynamic weights of the three decision indicators using the Analytic Hierarchy Process (AHP), and the weight vector. ,in For feature matching degree weight, For inventory fit weight, Weighted by historical loss rate, and ; Step 2: Normalize the indicator data: Feature matching degree normalization: in, For the first The Euclidean distance of the rolled fabric. This represents the maximum Euclidean distance among all fabrics. It is the minimum value; Inventory fit normalization: in, For the first Inventory fit rating for roll fabrics. To achieve the highest score; Historical loss rate normalization: in, For the first The loss rate of rolled fabric. For the maximum loss rate, It is the minimum value; Step 3: Construct the weighted normalized matrix ,in ; Step 4: Determine the ideal solution and negative ideal solution : Step 5: Calculate the closeness of each fabric layer to the ideal solution. : in, For the first The Euclidean distance between the rolled fabric and the ideal solution. The Euclidean distance to the negative ideal solution; Step 6: Press Sort the layouts from largest to smallest, and output the top 3 as the optimal and alternative layouts.

8. The automatic selection system for laser cutting and fabric laying according to claim 1, characterized in that, The fabric laying status monitoring unit of the laser cutting and fabric laying status closed-loop collaborative module includes a high-speed linear array camera and a laser contour sensor. The high-speed linear array camera is mounted 50cm upstream of the cutting station to photograph the fabric surface and identify wrinkles and wrinkle heights by measuring changes in grayscale values. The calculation formula is: in, This represents the grayscale difference between the wrinkled and smooth areas. The average gray value of the flat area. Define the camera calibration coefficient. This is abnormal; The laser profile sensor is used to measure the stretch in the fabric width direction. The calculation formula is: in, The width of the fabric to be laid out in real time. This refers to the nominal width of the fabric as specified by the manufacturer. The parameter adjustment unit includes a servo motor, a tension controller, and a carbon dioxide laser cutting head.

9. The automatic selection system for laser cutting and fabric laying according to claim 8, characterized in that, The closed-loop logic of the laser cutting and fabric laying state closed-loop collaborative module is as follows: When the line scan camera detects the wrinkle height Or the laser contour sensor detects Furthermore, after three consecutive sampling cycles, the alternative fabric laying method is switched, and the conveying tension is adjusted to the standard value via the servo motor. The tension adjustment formula is: in, This is the tension adjustment coefficient. This is the real-time tension value; When detected or At this time, only the laser cutting parameters are adjusted: Laser speed in wrinkled scenes Laser power ; In stretching scenarios: Cutting length compensation Actual cutting length ,in, To preset the cutting speed, To preset the laser power, To design the cutting length.

10. The automatic selection system for laser cutting and fabric laying according to claim 1, characterized in that, The end-to-end data chain of the full lifecycle traceability module includes: Data during the warehousing stage: UID, batch number, characteristic parameters, and supplier information written via RFID tags; Select stage data: Select decision time, AI algorithm version, optimal layout and alternative layout list; Cutting stage data: cutting start / end time, actual laser parameters, number of finished products, and waste length, including waste length. , This refers to the total length of fabric taken from the fabric in this batch; Finished product stage data: Fabric UID and airbag finished product SN are bound by a QR code, and the QR code contains production date and workstation information; The traceability module supports both forward and reverse traceability, and updates the correlation model between characteristics and loss rate based on the traceability data using a random forest algorithm. The model output is the predicted loss rate. The calculation formula is: in, For the first The weights of each decision tree For the first The model predicts the loss rate using a decision tree, and then feeds back into the AI ​​intelligent selection decision module to optimize the weight calculation of historical loss rate indicators.

Images