On-line automatic test method for compressive capacity of corrugated carton

Abstract

The invention provides an on-line automatic test method for the compressive capacity of a corrugated carton, and belongs to the technical field of corrugated carton production. A micro-deformation loading system is constructed to apply a test load which does not exceed 10% of the expected maximum bearing capacity, and a laser displacement sensor array is adopted to measure micro-deformation displacement data of six surfaces of the carton; and synchronously acquiring acoustic emission signals and extracting spectrum characteristic parameters. Micro-deformation data is analyzed based on a carton fiber structure physical mechanism model, stress field distribution is calculated, analysis is carried out in combination with a hierarchical attention carton strength evaluation model, a rigidity coefficient is calculated and compared with a reference threshold value, a structural weak area is identified through a strain energy distribution diagram, and a damage risk index is quantified. And comprehensively evaluating the compressive strength prediction value and the confidence interval of the corrugated carton, and outputting a prediction result, thereby realizing accurate evaluation of the compressive strength of the corrugated carton under a non-destructive condition.

Description

Technical Field

[0001] The invention belongs to the technical field of corrugated paper box production, and in particular relates to an online automatic testing method for the compression resistance of a corrugated paper box. Background Art

[0002] Corrugated paperboard is one of the most widely used packaging materials in the field of modern logistics and packaging. Its compressive resistance is directly related to the safety of packaged goods. Traditionally, the destructive test method is mainly used to evaluate the compressive resistance of corrugated paperboard, that is, a special pressure testing device is used to apply gradually increasing pressure to the paperboard until its structure is destroyed, and the maximum load-bearing capacity is recorded as the compressive strength index. In addition, there are also calculation methods based on empirical formulas, such as the Mackey formula, which estimate the compressive strength of the paperboard through the basic parameters of the paperboard, but the accuracy is limited.

[0003] However, traditional destructive testing not only wastes samples, but also cannot be applied to real-time quality control of production lines; while the method based on empirical formulas does not fully consider the microstructure characteristics and local deformation behavior of the material, making it difficult to accurately evaluate the actual compressive performance, especially when facing cartons of different specifications and materials, the prediction error is large. For the packaging of high-end precision products, this error may lead to serious economic losses.

[0004] In addition, existing technologies have difficulty effectively identifying weak areas in the carton structure and cannot provide early warning of potential damage risks, which is crucial in production line quality control. How to accurately predict the compressive strength of corrugated cartons and identify structural weak areas without damaging them has become a key technical problem that the industry needs to solve urgently. Summary of the invention

[0005] In view of this, the present invention provides a method for online automatic testing of the compressive strength of a corrugated paper box, which can solve the technical problem in the prior art that it is impossible to accurately predict the compressive strength of the corrugated paper box without destroying it.

[0006] The present invention is implemented as follows: The present invention provides a method for online automatic testing of the compressive strength of a corrugated paper box, including: constructing a micro-deformation loading system, applying a test load to the top of the corrugated paper box; using a high-precision laser displacement sensor array to measure micro-deformation displacement data; synchronously collecting acoustic emission signals and extracting spectrum characteristic parameters; analyzing micro-deformation displacement data based on a physical mechanism model of the paper box fiber structure, and applying a stress distribution calculation equation to calculate the stress field distribution of the corrugated paper box wall; inputting the micro-deformation displacement data and spectrum characteristic parameters into a pre-trained hierarchical attention paper box strength assessment model for analysis; calculating the stiffness coefficient and comparing it with a benchmark threshold; identifying structural weak areas and quantifying the damage risk index; using a multi-source data fusion algorithm to comprehensively assess the predicted value and confidence interval of compressive strength; and outputting the test results.

[0007] The test load refers to the pressure applied to the top of the corrugated box by the micro-deformation loading system, and its magnitude is within 10% of the expected maximum bearing capacity.

[0008] Among them, the micro-deformation loading system refers to a pressure-applying device driven by a high-precision servo motor, which ensures a smooth and precise loading process through closed-loop feedback control. The loading force ranges from 50N to 2000N, and the accuracy is not less than 0.5%.

[0009] Among them, the high-precision laser displacement sensor array refers to a measurement network composed of 24 triangulation principle laser displacement sensors, with a measurement accuracy of 1 micron and a sampling frequency of 1000 Hz, which is used to capture tiny deformations on the surface of the carton.

[0010] The micro-deformation displacement data refers to a set of deformation displacement values ​​of six surfaces of the corrugated box under the test load collected by a high-precision laser displacement sensor array.

[0011] The acoustic emission signal refers to the high-frequency sound wave signal generated by the internal fibers of the corrugated box under the test load, with a frequency range of 20kHz to 200kHz, which is collected by a piezoelectric sensor array.

[0012] The frequency spectrum characteristic parameters refer to a set of characteristic parameters extracted after performing frequency spectrum analysis on the acoustic emission signal, including energy distribution of each frequency band, peak frequency, center frequency and bandwidth.

[0013] Among them, the physical mechanism model of the carton fiber structure refers to a stress-strain analysis model based on the anisotropic mechanical properties of paper materials, taking into account the interlayer structural characteristics and fiber arrangement direction of corrugated paper.

[0014] Among them, the stress distribution calculation equation is based on Hooke's law and is used to calculate the stress distribution state of the corrugated box wall under the condition of micro-deformation displacement data. The input includes micro-deformation displacement data, material elastic modulus tensor, Poisson's ratio parameter, corrugated structure geometric parameters and test load size, and the output is three-dimensional stress field distribution and strain energy density distribution.

[0015] Among them, the hierarchical attention cardboard strength assessment model refers to a deep learning model that combines spatial hierarchical feature extraction and attention mechanism, which is used to predict the compressive strength of corrugated cardboard boxes from micro-deformation displacement data and spectral feature parameters; its hierarchical structure takes into account the multi-layer composite structure characteristics of corrugated cardboard boxes, and forms a corresponding relationship with the corrugated wave distance and corrugated wall thickness parameters in the corrugated structure geometric parameters through the feature extraction capabilities of different scales.

[0016] The present invention combines a high-precision micro-deformation loading system with a laser displacement sensor array to accurately capture the micro-deformation characteristics of cardboard boxes under low loads (within 10% of the maximum bearing capacity). By combining the analysis of acoustic emission signals, a comprehensive non-destructive evaluation system is constructed to ensure the integrity of the cardboard box structure during the testing process.

[0017] This method breaks through the limitation of traditional technologies that require destructive testing. Through a hierarchical attention cardboard box strength evaluation model, it fuses micro-deformation displacement data and acoustic spectrum characteristics to achieve high-precision prediction of the compressive strength of cardboard boxes. At the same time, based on the stress field distribution and strain energy analysis, the present invention can accurately identify weak structural areas and quantify the damage risk index, providing an early warning mechanism for production line quality control and successfully solving the technical problem of accurately predicting the compressive strength without damaging the cardboard box. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of the method of the present invention.

[0019] Figure 2 It is a schematic diagram of the composition of the on-line automatic testing system for the compressive capacity of corrugated cardboard boxes in Example 2.

[0020] Figure 3 It is a schematic diagram of the structure of the micro-deformation loading system in Example 2.

[0021] Figure 4 It is a schematic diagram of the structure of the acoustic emission signal acquisition system in Example 2.

[0022] Figure 5 It is the acoustic emission spectrum characteristics of different types of corrugated cardboard boxes in Example 2.

[0023] Figure 6 It is a three-dimensional model of the stress-strain distribution of corrugated cardboard boxes in Example 2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0025] As Figure 1 shown, it is a flowchart of a method for on-line automatic testing of the compressive capacity of corrugated cardboard boxes provided by the present invention. This method includes the following steps: A method for on-line automatic testing of the compressive capacity of corrugated cardboard boxes includes the following steps: S01. Construct a micro-deformation loading system and apply a test load not exceeding 10% of the expected maximum bearing capacity to the top of the corrugated cardboard box; S02, using a high-precision laser displacement sensor array to measure micro-deformation displacement data of six surfaces of the corrugated box under the test load; S03, synchronously collecting the acoustic emission signals generated during the test load application process and extracting spectrum characteristic parameters; S04. Analyze the micro-deformation displacement data based on the physical mechanism model of the carton fiber structure, and calculate the stress field distribution of the corrugated carton wall using the stress distribution calculation equation; S05, inputting the micro-deformation displacement data and the spectrum feature parameters into a pre-trained hierarchical attention carton strength assessment model for analysis; S06, calculating the stiffness coefficient of the corrugated box based on the slope of the elastic deformation curve calculated based on the micro-deformation displacement data and comparing it with a reference threshold; S07, generating a strain energy distribution map through the stress field distribution to identify weak areas of the corrugated box structure and quantify a potential damage risk index; S08. Comprehensively evaluate the predicted value of compressive strength of corrugated boxes and its confidence interval using multi-source data fusion algorithm; S09. Outputting the predicted compressive strength value, the confidence interval, the stress field distribution and the damage risk index as test results.

[0026] Among them, the micro-deformation loading system refers to a pressure-applying device driven by a high-precision servo motor. Through closed-loop feedback control, it ensures a smooth and precise loading process. The loading force ranges from 50N to 2000N, and the accuracy is not less than 0.5%.

[0027] The test load refers to the pressure applied to the top of the corrugated box by the micro-deformation loading system, and its magnitude is within 10% of the expected maximum bearing capacity.

[0028] Among them, the high-precision laser displacement sensor array refers to a measurement network composed of 24 triangulation principle laser displacement sensors with a measurement accuracy of 1 micron and a sampling frequency of 1000Hz, which is used to capture tiny deformations on the surface of the carton.

[0029] The micro-deformation displacement data refers to a set of deformation displacement values ​​of six surfaces of the corrugated box under the test load collected by the high-precision laser displacement sensor array.

[0030] Among them, the acoustic emission signal refers to the high-frequency sound wave signal generated by the internal fibers of the corrugated box under the test load, with a frequency range of 20kHz to 200kHz, which is collected by a piezoelectric sensor array.

[0031] Among them, the spectral feature parameters refer to the set of feature parameters extracted after performing spectral analysis on the acoustic emission signal, including the energy distribution in each frequency band, the peak frequency, the center frequency, and the bandwidth.

[0032] Among them, the physical mechanism model of the carton fiber structure refers to the stress-strain analysis model based on the anisotropic mechanical properties of paper materials, considering the interlayer structure characteristics of corrugated paper and the fiber arrangement direction.

[0033] Among them, the stress distribution calculation equation is based on Hooke's law and is used to calculate the stress distribution state of the corrugated carton wall under the condition of the micro-deformation displacement data. The inputs include the micro-deformation displacement data, the material elastic modulus tensor, the Poisson's ratio parameter, the geometric parameters of the corrugated structure, and the magnitude of the test load. The outputs are the three-dimensional stress field distribution and the strain energy density distribution. The material elastic modulus tensor refers to the second-order tensor describing the elastic properties of the corrugated carton material and is obtained through experimental determination. The Poisson's ratio parameter refers to the absolute value of the ratio of the transverse strain to the axial strain under uniaxial stress. The geometric parameters of the corrugated structure include the corrugation height, the corrugation pitch, the corrugation wall thickness, and the corrugation shape factor.

[0034] Among them, the stress field distribution refers to the three-dimensional stress distribution state of the corrugated carton wall calculated by the stress distribution calculation equation and is used for subsequent analysis and evaluation.

[0035] Among them, the hierarchical attention carton strength evaluation model refers to a deep learning model that combines spatial hierarchical feature extraction and attention mechanism and is used to predict the compressive strength of corrugated cartons from the micro-deformation displacement data and the spectral feature parameters.

[0036] Among them, the elastic deformation curve refers to the curve formed by the micro-deformation displacement data changing with the load under the action of the test load.

[0037] Among them, the stiffness coefficient refers to the reciprocal of the displacement change rate caused by a unit load increment in the elastic deformation stage, with the unit of N / mm, and is obtained by calculating the slope of the elastic deformation curve.

[0038] Among them, the reference threshold refers to the standard value of the stiffness coefficient preset according to the specifications and uses of corrugated cartons and is used for comparison with the stiffness coefficient.

[0039] Among them, the strain energy distribution diagram refers to the visual representation of the internal strain energy density distribution calculated from the stress field distribution.

[0040] Among them, the structurally weak area refers to the area with a higher strain energy density identified from the strain energy distribution diagram, indicating the part of the corrugated carton structure that is more likely to be damaged.

[0041] Among them, the damage risk index is a dimensionless parameter obtained based on the analysis of the structurally weak areas, ranging from 0 to 100, indicating the probability of the corrugated cardboard box being damaged under the standard load.

[0042] Among them, the multi-source data fusion algorithm is a calculation method that integrates the micro-deformation displacement data, the spectral feature parameters, the stiffness coefficient, and the damage risk index using a Bayesian inference framework, improving the prediction accuracy and reliability.

[0043] Among them, the predicted value of the compressive strength is the estimated value of the final compressive strength of the corrugated cardboard box calculated by the multi-source data fusion algorithm, with the unit of N.

[0044] Among them, the confidence interval is the range of uncertainty of the predicted value of the compressive strength, calculated by statistical methods, and expressed as a percentage range of the predicted value.

[0045] The specific structure of the hierarchical attention corrugated cardboard box strength evaluation model is a fusion architecture of a multi-level spatial feature extraction network and a temporal information processing network. Among them, the spatial feature extraction network uses six convolutional layers to encode the micro-deformation displacement data of the six faces of the corrugated cardboard box. The number of convolutional kernels in each layer is 64, 128, 256, 256, 512, 512, the size of the convolutional kernel is 3×3, the stride is 1, and the ReLU activation function is used; the temporal information processing network uses a bidirectional long short-term memory network to process the temporal features of the spectral feature parameters, and the number of neurons in the hidden layer is 512; the hierarchical attention corrugated cardboard box strength evaluation model introduces a multi-head cross-attention mechanism in the feature fusion layer, with the number of attention heads being 8 and the embedding dimension being 64, used to capture the correlation between the micro-deformation displacement data and the spectral feature parameters; the output layer uses a fully connected layer to map the fused features to the predicted value of the compressive strength and its uncertainty estimate, and the number of neurons in the fully connected layer is 256, 128, 64, 2 in sequence; the multi-head cross-attention mechanism can adaptively focus on the deformation patterns and acoustic features most relevant to the structurally weak areas by learning the correlation weights between the micro-deformation displacement data and the spectral feature parameters, and the multi-head cross-attention mechanism is closely related to the process of identifying the structurally weak areas in step S07; the hierarchical structure design of the hierarchical attention corrugated cardboard box strength evaluation model particularly considers the multi-layer composite structure characteristics of the corrugated cardboard box, and forms a corresponding relationship with the corrugation pitch and corrugation wall thickness parameters in the geometric parameters of the corrugated structure through the feature extraction capabilities of different scales.

[0046] The steps for establishing the training dataset of the hierarchical attention corrugated box strength evaluation model specifically include: First, a total of 20,000 corrugated box samples with different specifications, different corrugation types, and different raw material characteristics are extracted from the production line; then, each sample is subjected to a micro-deformation test and the complete micro-deformation displacement data and the acoustic emission signal are recorded; next, each sample is sent to a standard compressive strength tester for a destructive test and its final failure strength value is recorded as the label data; subsequently, the dataset is divided into a training set, a validation set, and a test set according to a ratio of 8:1:1; finally, all feature data are normalized so that the mean is 0 and the standard deviation is 1, and the principal component analysis method is used to reduce the dimension of the high-dimensional features, and the principal components with an explained variance ratio reaching 99% are retained.

[0047] The steps for training the hierarchical attention corrugated box strength evaluation model specifically include: First, initialize the model parameters using a Gaussian distribution with a mean of 0 and a standard deviation of 0.01; then, use the Adam optimizer for model training, set the initial learning rate to 0.001, and adopt the cosine annealing learning rate scheduling strategy with a minimum learning rate of 0.00001; next, set the batch size to 64 and the number of training epochs to 200; after each epoch of training, evaluate the model performance on the validation set, and use the mean absolute percentage error as the evaluation index; when the performance of the validation set has not improved for 10 consecutive epochs, start the early stopping mechanism to terminate the training; finally, evaluate the performance of the best model on the test set to ensure that the average prediction error is less than 5%; during the entire training process, adopt the mixed-precision training strategy to improve the computational efficiency, and use the L2 regularization method with a weight decay value of 0.0001 to prevent overfitting; during the training process, monitor the performance of the model on different types of corrugated boxes at the same time to ensure that the model has good generalization ability for various corrugated structures; the training process pays special attention to the sensitivity of the model to the high-value regions of the stress field distribution calculated in step S04, and by designing a special loss function weighting method, improve the model's ability to identify the structurally weak regions.

[0048] The following describes the specific implementation manners of the above steps in detail.

[0049] The specific implementation of constructing the micro-deformation loading system in step S01 is as follows: First, install a high-precision servo motor and its drive controller. The controller uses a 32-bit microprocessor to achieve closed-loop control, with a control frequency of 1000 Hz. Then, configure a loading force sensor with a range of 0 - 2000 N, a sensitivity of 0.01% of full scale, and a signal-to-noise ratio of not less than 85 dB. Next, install a balanced loading platform with the flatness error of the platform controlled within 0.01 mm to ensure uniform distribution of force during loading. Subsequently, design a closed-loop force feedback control algorithm. The algorithm adopts a proportional-integral-derivative (PID) control structure, where the proportional gain is set to 0.85, the integral time constant is 0.05 s, the derivative time constant is 0.02 s, and the integral saturation threshold is set to ±15% of the control output to prevent control overshoot caused by integral saturation. Finally, establish a loading curve planning module, which uses a fifth-order polynomial curve to achieve smooth loading, ensuring that the loading rate is always controlled within the range of 10 N / s to 50 N / s, and the acceleration is limited within 20 N / within. The loading system is connected to the main control computer through industrial Ethernet, with a data transmission bandwidth of 100 Mbps and a transmission delay of less than 1 ms. The function of step S01 is to provide a stable and accurate micro-load loading environment to ensure obtaining the micro-deformation characteristics of the cardboard box without damaging its structure. The closed-loop control technology in this step can accurately control the load size and avoid the risk of irreversible deformation of the cardboard box caused by excessive force.

[0050] The specific implementation of using a high-precision laser displacement sensor array to measure micro-deformation displacement data in step S02 is as follows: First, build a laser triangulation sensor support system. The support is made of carbon fiber composite material with a thermal expansion coefficient lower than 5× / °C to ensure stability during the measurement process. Then, arrange 24 laser displacement sensors reasonably. Six sensors are arranged on the top surface of the carton, four on the bottom surface, and 3 - 4 on each of the four side surfaces. The spacing between sensors is dynamically adjusted according to the carton size but not more than 150 mm. The measurement points are distributed to cover typical areas of each surface of the carton, especially the corner joints and the central area. Then, configure a high-speed synchronous data acquisition system, using time-division multiplexing technology, with the clock synchronization error controlled within 1 μs. The sampling accuracy of the data acquisition card is 16 bits, and the input range is ±10 V. Subsequently, design a data processing algorithm, using the sliding window median filtering method to remove measurement noise, with the window length of 5, and use the bicubic spline interpolation algorithm to construct the complete deformation field of the carton surface, with the interpolation grid density set to 4 nodes per square centimeter. Finally, implement a three-dimensional deformation field visualization module to generate a pseudo-color cloud map of the carton deformation, with the color mapping range from blue (minimum deformation) to red (maximum deformation). The purpose of this step is to obtain the deformation distribution data of each surface of the corrugated carton under a small load, which is the basis for subsequent analysis, capturing the subtle deformation characteristics of the carton through high-precision measurement and reflecting its structural integrity and mechanical properties.

[0051] The specific implementation method of synchronously collecting acoustic emission signals and extracting spectral characteristic parameters in step S03 is as follows: First, paste 8 high-sensitivity piezoelectric sensors on the surface of the carton. The sensor sensitivity is 75 dB (reference 0 dB = 1 V / μbar), and the frequency response range is 20 kHz - 200 kHz. The spacing between sensors does not exceed 200 mm. Then, configure an acoustic emission signal amplification and filtering circuit. The preamplifier gain is 40 dB, and the passband of the band-pass filter is 20 kHz - 200 kHz, with a roll-off slope of 48 dB / octave. Next, design the signal triggering and acquisition logic, using the floating threshold adaptive triggering method. The initial triggering threshold is set to 3 times the root mean square value of the background noise, and the sampling rate is set to 500 kHz to meet the requirements of the Nyquist sampling theorem. Subsequently, apply the short-time Fourier transform (STFT) to perform time-frequency analysis on the collected acoustic signals. The time window length is set to 2 ms, and the window overlap rate is 50%. Use the Hanning window to reduce spectral leakage. Finally, extract spectral characteristic parameters, including energy parameters (energy distribution in each frequency band, the energy ratios of the three frequency bands of 20 - 50 kHz, 50 - 100 kHz, and 100 - 200 kHz), frequency parameters (peak frequency, center frequency, weighted frequency, and bandwidth), and statistical parameters (spectral entropy, spectral skewness, and spectral kurtosis). The acoustic emission parameter extraction algorithm uses the wavelet packet decomposition method, with the decomposition level set to 5, and uses the db4 wavelet as the basis function. The purpose of this step is to obtain the acoustic signal characteristics generated by the internal fibers of the carton under a small load, and these acoustic characteristics can reflect the microscopic changes of the internal fiber structure of the carton, which is an important basis for predicting the macroscopic compressive performance.

[0052] The specific implementation method of analyzing the micro-deformation displacement data based on the physical mechanism model of the carton fiber structure in step S04 is as follows: First, an anisotropic constitutive model of the carton material is established. The corrugated carton material is regarded as an orthotropic material, and the stress-strain relationship is described by the generalized Hooke's law. The material properties are determined by 9 independent elastic constants (three elastic moduli , , , three Poisson's ratios , , , and three shear moduli , , ). Then, a geometric model of the corrugated structure is established. The corrugation is described as a sine wave, and its shape is defined by the wave height, wave pitch, and wall thickness parameters. The bonding characteristics between the face paper and the corrugated paper are considered in the model. Next, the carton is discretized into finite element meshes, and the mesh density is set to no less than 8 nodes per corrugation pitch to accurately capture the deformation characteristics of the corrugated structure. Subsequently, the micro-deformation displacement data obtained in step S02 is input into the finite element model as boundary conditions, and the displacement method is used to calculate the full-field stress distribution. Finally, the material parameters are optimized through iterative calculations to make the root mean square error between the deformation field predicted by the model and the measured deformation field less than 2%. The calculation process uses the nonlinear finite element method, considering the large deformation effect. The element type is selected as 20-node hexahedral elements, and the integration points use 3×3×3 Gaussian integration points. The function of this step is to reverse the stress field distribution of the carton wall surface based on the physical mechanics principle through the micro-deformation displacement data, providing a theoretical basis for the subsequent evaluation of the carton structure strength.

[0053] In step S05, the specific implementation of inputting the micro-deformation displacement data and spectral feature parameters into the hierarchical attention carton strength evaluation model is as follows: First, preprocess the micro-deformation displacement data, including removing outliers (using the improved Z-score method with a threshold of 3.5), normalizing (using the standard score method to normalize the data so that the mean is 0 and the standard deviation is 1), and spatially downsampling (using the adaptive grid method, maintaining high resolution in key areas and appropriately reducing the resolution in other areas). Then, preprocess the spectral feature parameters, including feature standardization (using the min-max normalization method to scale the features to the interval [0, 1]) and feature selection (using the recursive feature elimination method to retain the top 80% of the features ranked by importance). Next, input the processed micro-deformation displacement data into a six-layer convolutional neural network, the convolutional kernel parameters of which are set according to the aforementioned requirements, and the output is a 512-dimensional spatial feature vector. Subsequently, input the processed spectral feature parameters into a bidirectional long short-term memory network, which contains two layers with 512 neurons in each layer, and the output is a 512-dimensional temporal feature vector. Finally, fuse the spatial feature vector and the temporal feature vector through the multi-head cross-attention mechanism. The attention mechanism is implemented using the scaled dot-product attention algorithm, and the temperature parameter is set to 0.1. The fused features are mapped to the compressive strength prediction value and confidence interval through a fully connected layer. The role of this step is to use the deep learning model to extract high-level features from multi-source feature data and focus on the most relevant feature combinations through the attention mechanism to achieve accurate strength evaluation.

[0054] The specific implementation method of calculating the stiffness coefficient based on the slope of the elastic deformation curve calculated from the micro-deformation displacement data in step S06 is as follows: First, extract the corresponding relationship data points of load and displacement from the micro-deformation displacement data. For each measurement point, record the displacement values at different load levels to form a load-displacement data set. Then, perform noise filtering on the load-displacement data using a Savitzky-Golay filter with a window length of 7 and a polynomial order of 3 to retain the main trend of the data while removing high-frequency noise. Next, use the weighted least squares method to fit the load-displacement curve, with the fitting function using a first-order polynomial (linear) model and the weight set to the reciprocal of the displacement value of the data point to enhance the fitting accuracy in the small deformation region. Subsequently, calculate the slope of the fitted straight line, and the reciprocal of this slope is the stiffness coefficient, with the unit of N / mm. Finally, compare the calculated stiffness coefficient with a preset reference threshold, which is determined according to the carton specifications and is usually 90% of the standard carton stiffness. For example, for a 5-layer corrugated carton, the reference threshold is usually set to 800 N / mm - 1200 N / mm; for a 3-layer corrugated carton, the reference threshold is usually set to 400 N / mm - 700 N / mm. The comparison results are divided into three levels: "qualified" (not less than 95% of the reference threshold), "warning" (between 85% and 95% of the reference threshold), and "unqualified" (less than 85% of the reference threshold). The role of this step is to quantitatively evaluate the elastic stiffness of the carton. The stiffness coefficient is an important indicator for measuring the structural stability of the carton and is of great significance for predicting the compressive performance of the carton.

[0055] The specific implementation of generating the strain energy distribution map through the stress field distribution in step S07 is as follows: First, based on the stress field distribution calculated in step S04, calculate the strain energy density of each element node. The calculation formula is that the strain energy density is equal to half of the inner product of the stress component and the strain component. Then, map the element node strain energy density information to the surface of the cardboard box three-dimensional model, and use the node averaging method to process the transition of the strain energy density between elements. Next, apply the adaptive color mapping algorithm to generate the strain energy distribution contour map. The color range is from blue (low strain energy) to red (high strain energy), and the number of color gradations is set to 10 levels. The grading method uses percentile segmentation to ensure that each color level covers the same number of data points. Subsequently, identify the structurally weak areas based on the strain energy distribution. Mark the areas with a strain energy density higher than the 90th percentile as high-risk areas, the areas between the 75th percentile and the 90th percentile as medium-risk areas, and the areas below the 75th percentile as low-risk areas. Finally, calculate the failure risk index. Use the weighted summation method, with the weight distribution being the proportion of high-risk areas × 100, the proportion of medium-risk areas × 60, and the proportion of low-risk areas × 20. At the same time, consider the concentration of high-risk areas. The higher the concentration, the greater the risk index. The risk index is normalized to the range of 0 - 100. The role of this step is to identify the weak links of the cardboard box structure through the energy method, predict the possible failure positions, and quantify the structural failure risk, providing a basis for improving the cardboard box design.

[0056] The specific implementation of comprehensively evaluating the predicted value of the compressive strength using the multi-source data fusion algorithm in step S08 is as follows: First, establish a Bayesian network model. The network structure includes four input nodes (micro-deformation displacement characteristics, spectral feature parameters, stiffness coefficient, and failure risk index) and one output node (predicted value of the compressive strength). Then, define the conditional probability distribution of each node. Use the Gaussian mixture model to describe the probability distribution of continuous variables. The number of mixture components is set to 3, and the model parameters are estimated through the expectation maximization algorithm. Next, use the Monte Carlo Markov chain (MCMC) method for Bayesian inference. The Metropolis - Hastings algorithm is used to implement the sampling process, and the number of samplings is set to 10000. The first 2000 samplings are regarded as the warm-up period and discarded. Subsequently, calculate the posterior distribution of the predicted value of the compressive strength based on the sampling results. Take the median of the posterior distribution as the final predicted value, and take the 2.5% quantile and 97.5% quantile as the lower and upper limits of the 95% confidence interval. Finally, calculate the prediction reliability index. Calculate the mean absolute percentage error of the prediction model based on historical data. This error is usually controlled within 5%. If it exceeds this threshold, the system will issue a calibration prompt. The role of this step is to comprehensively consider various test indicators, obtain a reliable predicted result of the compressive strength through probability statistics methods, and at the same time give the uncertainty range of the prediction, providing a scientific decision-making basis for practical applications.

[0057] The specific implementation of outputting the test results in step S09 is to first generate a compressive strength prediction report, including the predicted value (unit: N), relative prediction error (statistically based on historical data, usually ±3% - ±5%), 95% confidence interval (expressed as a percentage of the predicted value, usually in the range of ±5% - ±10% of the predicted value), and test condition parameters (ambient temperature, humidity, etc.). Then generate a stress distribution visualization report, including a three-dimensional color stress cloud diagram, principal stress vector diagram, and stress concentration area identification. The image resolution is not less than 1024×768 pixels, the color mapping uses a rainbow color scale, and a numerical scale is attached. Next, generate a damage risk assessment report, including a risk index value (a dimensionless value from 0 to 100), risk level classification (0 - 30 is low risk, 30 - 70 is medium risk, 70 - 100 is high risk), and a description of the location of the weak area (identified using the partition numbers on the surface of the cardboard box). Finally, implement the function of exporting the test results, supporting the export of reports in three formats: PDF, Excel, and JSON, and providing a historical data comparison function, which can be compared and analyzed with the historical test results of the same type of cardboard box. The role of this step is to present the complex test analysis results to the user in an intuitive and easy-to-understand way, facilitating the user to make quick judgments and decisions, and at the same time providing a standardized data interface to support subsequent data analysis and management.

[0058] Specifically, the principle of the present invention is as follows: The core of the technical principle of the present invention lies in combining the micro-deformation test technology with the multi-source data fusion algorithm. Based on the material mechanics theory and deep learning method, a mapping relationship between the microscopic deformation behavior and the macroscopic compressive strength of the corrugated cardboard box is established. When the corrugated cardboard box is under low-load conditions, its structure will produce small but measurable elastic deformations. Although these deformations are difficult to detect by the naked eye, they contain rich material and structural information.

[0059] First, the present invention uses a high-precision laser displacement sensor array to accurately capture the micro-deformation data of the six surfaces of the cardboard box. The measurement accuracy reaches the micron level, which can reflect the local stress concentration points and deformation patterns of the material. At the same time, the acoustic emission signals generated by the fiber structure inside the cardboard box under stress are collected. These high-frequency acoustic wave signals (20 kHz to 200 kHz) can reflect the changes in the internal microscopic structure of the material and the precursor characteristics of potential damage.

[0060] Secondly, based on the physical mechanism model of the cardboard box fiber structure, the present invention applies the stress distribution calculation equation to convert the micro-deformation displacement data into a three-dimensional stress field distribution, and then calculates the strain energy distribution. This process follows Hooke's law and the principle of energy conservation in material mechanics. The non-uniformity of the strain energy density distribution directly reflects the weak areas of the structure, that is, the positions most likely to be damaged under actual load conditions.

[0061] At the same time, the hierarchical attention carton strength assessment model designed by the present invention adopts a fusion architecture of a multi-level spatial feature extraction network and a temporal information processing network, combined with a multi-head cross-attention mechanism, which can adaptively identify the patterns in micro-deformation data and acoustic signal features that are most relevant to compressive strength. The hierarchical structure design of the model takes into account the multi-layer composite structure characteristics of corrugated carton boxes and can effectively capture deformation characteristics of different scales.

[0062] Finally, through the multi-source data fusion algorithm of the Bayesian reasoning framework, the micro-deformation displacement data, spectral characteristic parameters, stiffness coefficient and damage risk index are integrated to comprehensively evaluate the predicted value of the compressive strength of the corrugated box and its confidence interval. The entire prediction process has a solid physical and statistical basis, and has shown high accuracy and reliability in a large number of experimental verifications.

[0063] A specific embodiment 1 of the present invention is provided below, and the specific implementation method of each step in this embodiment 1 is described in detail as follows.

[0064] The specific implementation method of constructing the micro-deformation loading system in step S01 is to first install a high-precision servo motor and its drive controller. The controller uses a 32-bit microprocessor to achieve closed-loop control with a control frequency of 1000Hz. Then configure the loading force sensor with a range of 0 to 2000N, a sensitivity of 0.01% of full scale, and a signal-to-noise ratio of not less than 85dB. Then install a balanced loading platform, and control the platform flatness error within 0.01mm to ensure uniform distribution of force during loading. Subsequently, a closed-loop force feedback control algorithm is designed. The algorithm adopts a proportional integral differential (PID) control structure, in which the proportional gain is set to 0.85, the integral time constant is 0.05s, the differential time constant is 0.02s, and the integral saturation threshold is set to ±15% of the control output to prevent control overshoot caused by integral saturation. Finally, a loading curve planning module is established, and a quintic polynomial curve is used to achieve smooth loading to ensure that the loading rate is always controlled within the range of 10N / s to 50N / s, and the acceleration is limited to 20N / The loading system is connected to the main control computer via industrial Ethernet, with a data transmission bandwidth of 100Mbps and a transmission delay of less than 1ms. The function of step S01 is to provide a stable and precise micro-load loading environment to ensure that the micro-deformation characteristics of the carton are obtained without damaging the carton structure. The closed-loop control technology in this step can accurately control the load size to avoid the risk of irreversible deformation of the carton due to excessive force.

[0065] In step S02, a specific implementation method of using a high-precision laser displacement sensor array to measure micro-deformation displacement data is to first build a laser triangulation sensor bracket system, the bracket is made of carbon fiber composite material, and the thermal expansion coefficient is less than 5× / °C to ensure stability during the measurement process. Then, arrange 24 laser displacement sensors reasonably. Among them, 6 are arranged on the top surface of the carton, 4 on the bottom surface, and 3 - 4 on each of the four side surfaces. The spacing between the sensors is dynamically adjusted according to the size of the carton, but not more than 150 mm. The measurement points are distributed to cover the typical areas of each surface of the carton, especially the corner joints and the central area. Then, configure a high-speed synchronous data acquisition system, adopt time-division multiplexing technology, control the clock synchronization error within 1 μs, the sampling accuracy of the data acquisition card is 16 bits, and the input range is ±10 V. Subsequently, design a data processing algorithm, use the sliding window median filtering method to remove measurement noise, the window length is 5, and use the bicubic spline interpolation algorithm to construct the complete deformation field of the carton surface, and set the interpolation grid density to 4 nodes per square centimeter. Finally, implement a three-dimensional deformation field visualization module to generate a pseudo-color cloud map of the carton deformation, and the color mapping range is from blue (minimum deformation) to red (maximum deformation). The function of this step is to obtain the deformation distribution data of each surface of the corrugated carton under the action of a small load. These data are the basis for subsequent analysis, capture the subtle deformation characteristics of the carton through high-precision measurement, and reflect its structural integrity and mechanical properties.

[0066] The specific implementation method of synchronously collecting acoustic emission signals and extracting spectral characteristic parameters in step S03 is as follows: First, paste 8 high-sensitivity piezoelectric sensors on the surface of the carton. The sensitivity of the sensor is 75 dB (reference 0 dB = 1 V / μbar), the frequency response range is 20 kHz - 200 kHz, and the spacing between the sensors does not exceed 200 mm. Then, configure an acoustic emission signal amplification and filtering circuit. The gain of the preamplifier is 40 dB, the passband of the band-pass filter is 20 kHz - 200 kHz, and the roll-off slope is 48 dB / octave. Next, design a signal trigger and acquisition logic, adopt a floating threshold adaptive trigger method, set the initial trigger threshold to 3 times the root mean square value of the background noise, and set the sampling rate to 500 kHz to meet the requirements of the Nyquist sampling theorem. Subsequently, apply the short-time Fourier transform (STFT) to perform time-frequency analysis on the collected acoustic signals. The length of the time window is set to 2 ms, the window overlap rate is 50%, and the Hanning window is used to reduce spectral leakage. Finally, extract spectral characteristic parameters, including energy parameters (energy distribution in each frequency band, the energy ratios of the three frequency bands of 20 - 50 kHz, 50 - 100 kHz, and 100 - 200 kHz), frequency parameters (peak frequency, center frequency, weighted frequency, and bandwidth), and statistical parameters (spectral entropy, spectral skewness, and spectral kurtosis). The acoustic emission parameter extraction algorithm uses the wavelet packet decomposition method, and the decomposition level is set to 5, and the db4 wavelet is used as the basis function. The function of this step is to obtain the acoustic signal characteristics generated by the internal fibers of the carton under the action of a small load. These acoustic characteristics can reflect the microscopic changes of the internal fiber structure of the carton and are an important basis for predicting the macroscopic compressive performance.

[0067] The specific implementation of analyzing the micro-deformation displacement data based on the physical mechanism model of the carton fiber structure in step S04 is as follows: First, an anisotropic constitutive model of the carton material is established. The corrugated carton material is regarded as an orthotropic material, and the generalized Hooke's law is used to describe the stress-strain relationship. The material properties are determined by 9 independent elastic constants (three elastic moduli , , , three Poisson's ratios , , , and three shear moduli , , ). Then, a geometric model of the corrugated structure is established. The corrugation is described as a sine wave, and its shape is defined by the wave height, wave pitch, and wall thickness parameters. The bonding characteristics between the face paper and the corrugated paper are considered in the model. Next, the carton is discretized into finite element meshes, and the mesh density is set to no less than 8 nodes per corrugation pitch to accurately capture the deformation characteristics of the corrugated structure. Subsequently, the micro-deformation displacement data obtained in step S02 is input into the finite element model as boundary conditions, and the displacement method is used to calculate the full-field stress distribution. Finally, the material parameters are optimized through iterative calculations to make the root mean square error between the deformation field predicted by the model and the measured deformation field less than 2%. The calculation process uses the nonlinear finite element method, considering the large deformation effect. The element type is selected as 20-node hexahedral elements, and the integration points use 3×3×3 Gauss integration points. The role of this step is to reverse the stress field distribution of the carton wall surface based on the physical mechanics principle through the micro-deformation displacement data, providing a theoretical basis for the subsequent evaluation of the carton structure strength.

[0068] The stress distribution calculation equation is based on the generalized Hooke's law of orthotropic materials, and its matrix form can be expressed as: , where the components of the compliance matrix are: , , , , , , , , , , , .

[0069] Taking the inverse of the above compliance matrix, the stiffness matrix can be obtained, representing the relationship between strain and stress: .

[0070] For the corrugated structure, adopting the laminated composite theory, the equivalent elastic moduli along the corrugated direction (taken as the 1-direction) and perpendicular to the corrugated direction (taken as the 2-direction) are respectively: , , where, and respectively represent the elastic moduli of the face paper and the corrugated paper, and respectively represent the thicknesses of the face paper and the corrugated paper.

[0071] Based on the obtained micro-deformation displacement data and the geometric parameters of the corrugated box, the stress equilibrium equation is solved by the finite element method: , where, represents the divergence operator, represents the stress tensor, represents the body force. Adopting the displacement method, the displacement field is expressed as a combination of the nodal displacement and the shape function : , where, represents the number of element nodes (for a 20-node hexahedral element, ). Substituting the displacement field into the strain-displacement relationship gives the strain field: , where, represents the strain-displacement matrix. Substituting the strain field into the constitutive relationship gives the stress field: , where, represents the above stiffness matrix. Applying the principle of minimum potential energy, for any virtual displacement , it should satisfy: , where, represents the surface force. Expressing the virtual displacement field as , substituting it into the above equation and considering the arbitrariness of , the discrete form of the equilibrium equation is obtained: , where, represents the global stiffness matrix, represents the global load vector, which are respectively: , .

[0072] Solving the above equation can obtain the nodal displacements , and then the strain field can be obtained and the stress field . For the thin-walled structure of corrugated cardboard boxes, the membrane stress and the bending moment are further calculated for evaluating the buckling stability: , , wherein, represents the wall thickness, represents the coordinate along the wall thickness direction. The calculation accuracy of the final stress field distribution is verified by comparing with the measured deformation field, and the material parameters are iteratively optimized until the root mean square error is lower than 2%.

[0073] The specific implementation of inputting the micro-deformation displacement data and the spectral feature parameters into the hierarchical attention corrugated box strength evaluation model in step S05 is to first preprocess the micro-deformation displacement data, including removing outliers (using the improved Z-score method with a threshold set to 3.5), normalization (using the standard score method for normalization to make the data mean 0 and standard deviation 1), and spatial downsampling (using the adaptive grid method, keeping high resolution in key areas and appropriately reducing resolution in other areas). Then preprocess the spectral feature parameters, including feature standardization (using the min-max normalization method to scale the features to the interval [0, 1]) and feature selection (using the recursive feature elimination method to retain the top 80% of the features ranked by importance). Next, input the processed micro-deformation displacement data into a six-layer convolutional neural network, the convolutional kernel parameters of which are set according to the aforementioned requirements, and the output is a 512-dimensional spatial feature vector. Subsequently, input the processed spectral feature parameters into a bidirectional long short-term memory network, which contains two layers with 512 neurons in each layer, and the output is a 512-dimensional temporal feature vector. Finally, fuse the spatial feature vector and the temporal feature vector through the multi-head cross-attention mechanism. The attention mechanism is implemented using the scaled dot-product attention algorithm with the temperature parameter set to 0.1, and the fused features are mapped to the compressive strength prediction value and the confidence interval through a fully connected layer. The role of this step is to use the deep learning model to extract high-level features from multi-source feature data and focus on the most relevant feature combinations through the attention mechanism to achieve accurate strength evaluation.

[0074] The calculation process of the multi-head cross-attention mechanism can be expressed as: , wherein, represents the query matrix (from the spatial feature vector), Denote the key matrix (from the time series feature vector), Denote the value matrix (also from the time series feature vector), Denote the dimension of the key vector, Denote the temperature parameter (set to 0.1). The multi-head attention mechanism calculates 8 parallel attention heads, then concatenates the results and passes them through a linear transformation: , where, , and are parameter matrices to be learned.

[0075] The specific implementation of calculating the stiffness coefficient based on the slope of the elastic deformation curve formed from the micro-deformation displacement data in step S06 is as follows: First, extract the corresponding relationship data points of load and displacement from the micro-deformation displacement data. For each measurement point, record the displacement values at different load levels to form a load-displacement data set. Then, perform noise filtering on the load-displacement data using the Savitzky-Golay filter with a window length of 7 and a polynomial order of 3 to retain the main trend of the data while removing high-frequency noise. Next, use the weighted least squares method to fit the load-displacement curve, with the fitting function using a first-degree polynomial (linear) model and the weight set to the reciprocal of the displacement value of the data point to enhance the fitting accuracy in the small deformation region. Subsequently, calculate the slope of the fitted straight line, and the reciprocal of this slope is the stiffness coefficient, with the unit of N / mm. Finally, compare the calculated stiffness coefficient with a preset reference threshold, which is determined according to the carton specifications and is usually 90% of the standard carton stiffness. For example, for a 5-layer corrugated carton, the reference threshold is usually set to 800 N / mm - 1200 N / mm; for a 3-layer corrugated carton, the reference threshold is usually set to 400 N / mm - 700 N / mm. The comparison results are divided into three levels: qualified (not less than 95% of the reference threshold), warning (between 85% and 95% of the reference threshold), and unqualified (less than 85% of the reference threshold). The role of this step is to quantitatively evaluate the elastic stiffness of the carton. The stiffness coefficient is an important indicator for measuring the structural stability of the carton and is of great significance for predicting the compressive performance of the carton.

[0076] The calculation process of fitting the load-displacement curve by the weighted least squares method can be expressed as: , where, Denote the load value of the th data point, Denote the displacement value of the th data point, Denote the weight of the th data point, Denote the total number of data points, represents the slope of the fitting line. The stiffness coefficient is calculated as: , with the unit of N / mm.

[0077] The specific implementation of generating the strain energy distribution map from the stress field distribution in step S07 is as follows: First, based on the stress field distribution calculated in step S04, calculate the strain energy density of each element node. The calculation formula is that the strain energy density is equal to half of the inner product of the stress component and the strain component. Then map the element node strain energy density information to the surface of the cardboard box three-dimensional model, and use the node averaging method to handle the transition of the strain energy density between elements. Next, apply the adaptive color mapping algorithm to generate the strain energy distribution contour map, with the color range from blue (low strain energy) to red (high strain energy), and the number of color gradations set to 10 levels. The grading method uses percentile segmentation to ensure that each color level covers the same number of data points. Subsequently, identify the structurally weak areas based on the strain energy distribution, mark the areas with a strain energy density higher than the 90th percentile as high-risk areas, the areas between the 75th percentile and the 90th percentile as medium-risk areas, and the areas below the 75th percentile as low-risk areas. Finally, calculate the failure risk index using the weighted summation method, with the weight assignment being the proportion of high-risk areas × 100, the proportion of medium-risk areas × 60, and the proportion of low-risk areas × 20. At the same time, consider the concentration of high-risk areas. The higher the concentration, the greater the risk index, and the risk index is normalized to the range of 0 - 100. The function of this step is to identify the weak links of the cardboard box structure through the energy method, predict the possible failure positions, and quantify the structural failure risk, providing a basis for improving the cardboard box design.

[0078] The strain energy density calculation formula can be expressed as: , where represents the strain energy density, and represent the components of the stress and strain tensors respectively, , , represent the normal stress components, , , represent the shear stress components, , , represent the normal strain components, , , represent the shear strain components.

[0079] The failure risk index calculation formula can be expressed as: , Among them, represents the damage risk index, , , respectively represent the weight coefficients of high, medium, and low risk areas, , , respectively represent the proportions of high, medium, and low risk areas in the total area, represents the concentration factor of the high risk area, and the calculation formula is: , where, represents the number of connected components in the high risk area, represents the possible maximum number of connected components (set to 10% of the total number of units), represents the concentration influence coefficient (set to 0.5).

[0080] In step S08, the specific implementation method of comprehensively evaluating the predicted value of compressive strength using the multi-source data fusion algorithm is to first establish a Bayesian network model. The network structure includes four input nodes (micro-deformation displacement characteristics, spectral feature parameters, stiffness coefficient, and damage risk index) and one output node (predicted value of compressive strength). Then, define the conditional probability distribution of each node. Use the Gaussian mixture model to describe the probability distribution of continuous variables, set the number of mixture components to 3, and estimate the model parameters through the expectation maximization algorithm. Next, use the Monte Carlo Markov chain (MCMC) method for Bayesian inference, and use the Metropolis-Hastings algorithm to implement the sampling process. Set the number of sampling times to 10000, and discard the first 2000 times as the warm-up period. Subsequently, calculate the posterior distribution of the predicted value of compressive strength based on the sampling results, take the median of the posterior distribution as the final predicted value, and take the 2.5% quantile and 97.5% quantile as the lower and upper limits of the 95% confidence interval. Finally, calculate the prediction reliability index, calculate the mean absolute percentage error of the prediction model based on historical data, and this error is usually controlled within 5%. If it exceeds this threshold, the system will issue a calibration prompt. The role of this step is to comprehensively consider various test indicators, obtain reliable predicted results of compressive strength through probability statistics methods, and at the same time give the uncertainty range of the prediction, providing a scientific decision-making basis for practical applications.

[0081] The probability density function of the Gaussian mixture model can be expressed as: , where, represents the observed data, represents the model parameters, represents the number of mixture components, represents the th weight of the mixture component (satisfying ), represents a mean of and a covariance matrix of for a multivariate Gaussian distribution: , where represents the data dimension, represents the determinant of the covariance matrix.

[0082] The specific implementation of outputting the test results in step S09 is to first generate a compressive strength prediction report, including the predicted value (unit: N), the relative prediction error (statistically based on historical data, usually ±3% - ±5%), the 95% confidence interval (expressed as a percentage of the predicted value, usually in the range of ±5% - ±10% of the predicted value), and the test condition parameters (ambient temperature, humidity, etc.). Then generate a stress distribution visualization report, including a three-dimensional color stress cloud map, a principal stress vector map, and an identification of the stress concentration area. The image resolution is not less than 1024×768 pixels, the color mapping uses a rainbow color scale, and a numerical scale is attached. Next, generate a damage risk assessment report, including a risk index value (a dimensionless value from 0 to 100), a risk level classification (0 - 30 is low risk, 30 - 70 is medium risk, 70 - 100 is high risk), and a description of the location of the weak area (identified using the partition numbers on the surface of the cardboard box). Finally, implement the function of exporting the test results, supporting the export of reports in three formats: PDF, Excel, and JSON, and providing a historical data comparison function, which can be used to compare and analyze with the historical test results of the same type of cardboard box. The role of this step is to present the complex test analysis results to the user in an intuitive and understandable way, facilitating the user to make quick judgments and decisions, and at the same time providing a standardized data interface to support subsequent data analysis and management.

[0083] The loss function in the hierarchical attention cardboard box strength assessment model is in the form of a weighted combination, including the mean squared error loss and the weighted region-sensitive loss: , where represents the mean squared error loss, which is used to measure the difference between the predicted strength and the actual strength: , where represents the th sample's actual compressive strength, represents the compressive strength predicted by the model, represents the number of samples. represents the weighted region-sensitive loss, which is used to enhance the model's ability to identify structurally weak areas: , where Indicates the th sample's actual strain energy density of the th region, represents the strain energy density predicted by the model, represents the number of region divisions, Indicates the weight of the th region. The weight is proportional to the failure frequency of this region in the historical failure data: , where represents the failure frequency of the th region in the historical data. and are the weight coefficients of the two loss terms, which are set to 0.7 and 0.3 respectively.

[0084] Through the above implementation method, the on-line automatic test method for the compressive capacity of corrugated cartons realizes the non-destructive prediction of the compressive strength of cartons, and has the advantages of fast test speed, high prediction accuracy, simple operation, etc., and can be effectively applied to the quality control process of corrugated carton production lines.

[0085] To better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: In order to study the damage problem of corrugated cartons during the transportation of electronic products, the method of the present invention is used to conduct on-line tests on the compressive capacity of corrugated cartons of different specifications. Three common types of corrugated cartons are selected for this test: Type A (3-layer single corrugated carton), Type B (5-layer double corrugated carton), and Type C (7-layer triple corrugated carton). 100 samples are selected for each type for testing. The test environment temperature is controlled at 23±1°C, and the relative humidity is controlled at 50±2% to ensure the consistency and reliability of the test results. As Figure 2 shown, it is a schematic diagram of the composition of the on-line automatic test system for the compressive capacity of corrugated cartons.

[0086] First, as Figure 3 shown, a micro-deformation loading system that meets the requirements of step S01 is constructed. The system uses a servo motor with a rated power of 2.5kW, a maximum torque of 12N·m, a 32-bit microprocessor for the controller, and a control frequency set to 1000Hz. The loading force sensor has a range of 0-2000N, a sensitivity of 0.01% of full scale, and a signal-to-noise ratio of 87dB. The flatness error of the balanced loading platform is controlled within 0.008mm. The PID parameters of the closed-loop force feedback control algorithm are shown in Table 1: Table 1 PID control parameter table of the micro-deformation loading system

[0087] Then, according to the requirements of step S02, a high-precision laser displacement sensor array was arranged. The sensor bracket is made of carbon fiber composite material with a thermal expansion coefficient of 4.2× / ℃. The specific arrangement of the 24 laser displacement sensors is shown in Table 2: Table 2 Laser Displacement Sensor Arrangement Scheme Table

[0088] Next, according to the requirements of step S03, acoustic emission signals were collected and spectral characteristic parameters were extracted. Eight piezoelectric sensors were pasted on the surface of the cardboard box. The sensor sensitivity is 75 dB, the frequency response range is 20 kHz to 200 kHz, and the sensor spacing is controlled within 180 mm. The trigger threshold of the acoustic emission signal is set to 3.2 times the root mean square value of the background noise, and the sampling rate is 500 kHz. The short-time Fourier transform was used to analyze the collected acoustic signals. The main spectral characteristic parameters of the three types of cardboard boxes are shown in Table 3: Table 3 Acoustic Emission Spectral Characteristic Parameter Table of Different Types of Cardboard Boxes

[0089] Figure 5 In the figure, the abscissa represents the frequency and the ordinate represents the normalized amplitude. The spectral curves of three different types of cardboard boxes are plotted in the figure. It can be clearly seen from this chart that as the number of corrugated cardboard box layers increases, the peak frequency of the acoustic emission signal gradually moves towards the high-frequency direction, and the energy ratio in the high-frequency band (100 - 200 kHz) gradually increases. According to the requirements of step S04, a constitutive model of the anisotropy of the cardboard box material was established. The material elastic constants of the Type A cardboard box are shown in Table 4: Table 4 Material Elastic Constant Table of Type A Cardboard Box

[0090] After inputting the micro-deformation displacement data into the finite element model, the calculated stress field distribution shows that when the Type A cardboard box bears a test load of 200 N, the maximum stress appears at the connection between the side and the bottom, reaching 1.25 MPa, and the stress concentration coefficient is 2.8.

[0091] According to the requirements of step S05, a hierarchical attention cardboard box strength evaluation model was used to process the micro-deformation displacement data and spectral characteristic parameters. The model structure includes six convolutional layers and a bidirectional long short-term memory network, and a multi-head cross-attention mechanism is used to fuse features. The model configuration and training parameters are shown in Table 5: Table 5 Hierarchical Attention Model Configuration Parameter Table

[0092] According to the requirements of step S06, the stiffness coefficients of three types of cartons were calculated and compared with the reference threshold. The results are shown in Table 6 as follows: Table 6 Comparison Results of Stiffness Coefficients of Different Types of Cartons

[0093] According to the requirements of step S07, a strain energy distribution diagram was generated through the stress field distribution, and the structurally weak areas were identified. The failure risk assessment results of three types of cartons are shown in Table 7 as follows: Table 7 Failure Risk Assessment Results of Different Types of Cartons

[0094] Figure 6 In the form of a three-dimensional surface diagram, the stress distribution on the surface of the carton is shown. The horizontal axis and the vertical axis represent the relative position coordinates on the surface of the carton, and the vertical axis represents the stress value (MPa). The color gradient is used in the figure to represent the stress magnitude, from blue (low stress) to yellow and then to red (high stress). The chart also marks the areas of different risk levels: red dots represent high-risk areas (stress > 1.5 MPa), orange dots represent medium-risk areas (1.0 MPa < stress ≤ 1.5 MPa), and green dots represent low-risk areas (stress ≤ 1.0 MPa). The corner stress concentration area (risk index 76.4) and the central compression area (risk index 58.2) are specially marked. According to the requirements of step S08, a comprehensive assessment of the compressive strength is carried out using a multi-source data fusion algorithm. The Bayesian network model uses a Gaussian mixture model to describe the conditional probability distribution of each node, and the number of mixture components is 3. Through the MCMC method, 10,000 samplings are carried out, and the prediction results are shown in Table 8 as follows: Table 8 Prediction Results of Compressive Strength of Different Types of Cartons

[0095] Finally, according to the requirements of step S09, a test result report is generated, which includes three parts: compressive strength prediction, stress analysis, and risk assessment. For the weak areas of type C cartons, the recommended optimized design solutions are shown in Table 9 as follows: Table 9 Structure Optimization Suggestion Table for Type C Cartons

[0096] The traditional method for testing the compressive strength of corrugated cartons mainly uses destructive testing, that is, applying continuously increasing pressure to the carton through a press until the carton is damaged, and recording the maximum pressure at the time of damage as the compressive strength. This method has the following problems: First, the test process completely destroys the sample, resulting in waste of resources; second, each carton can only be tested once, and it is impossible to conduct a full inspection on the production line; third, it is impossible to predict in advance the weak areas and possible failure modes of the carton, making it difficult to improve the design targeted; fourth, the test results have a large discreteness and are significantly affected by operating conditions. The online automatic testing method for the compressive strength of corrugated cartons proposed by the present invention has the following significant advantages compared with the traditional method: First, it uses the micro-deformation testing principle and causes no damage to the carton; second, it obtains multi-dimensional information through a high-precision sensor array and acoustic emission technology, which can comprehensively reflect the structural characteristics of the carton; third, it uses a method combining physical models and deep learning to achieve high-precision prediction of the compressive strength, and the average prediction error is controlled within 5%; finally, it can identify the weak structural areas and quantify the failure risk, providing a direct basis for optimizing the carton structure.

[0097] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 10 and 11 below.

[0098] Table 10 Variable Explanation Table (First Part)

[0099] Table 11 Variable Explanation Table (Second Part)

[0100] As described above, the above are only the specific implementation manners of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. A method for on-line automatic testing of the compressive strength of corrugated cardboard boxes, characterized in that, include: Construct a micro-deformation loading system to apply the test load to the top of the corrugated box; Use high-precision laser displacement sensor array to measure micro-deformation displacement data; Acoustic emission signals are collected synchronously and spectrum characteristic parameters are extracted; micro-deformation displacement data are analyzed based on the physical mechanism model of carton fiber structure, and stress distribution calculation equation is used to calculate the stress field distribution of corrugated carton wall; The micro-deformation displacement data and spectral feature parameters are input into the pre-trained hierarchical attention carton strength assessment model for analysis; the stiffness coefficient is calculated and compared with the benchmark threshold; the structural weak areas are identified and the damage risk index is quantified; The multi-source data fusion algorithm is used to comprehensively evaluate the predicted value and confidence interval of compressive strength; Output test results.

2. The method for on-line automatic testing of the compressive strength of corrugated cartons according to claim 1, characterized in that, The test load is the pressure applied to the top of the corrugated box by the micro-deformation loading system, which is within 10% of the expected maximum load-bearing capacity.

3. The method for online automatic testing of the compressive strength of corrugated cardboard boxes according to claim 2, characterized in that, The micro-deformation loading system refers to a pressure-applying device driven by a high-precision servo motor, which ensures a smooth and precise loading process through closed-loop feedback control. The loading force ranges from 50N to 2000N, and the accuracy is not less than 0.5%.

4. The method for on-line automatic testing of the compressive strength of corrugated cardboard boxes according to claim 3, characterized in that, The high-precision laser displacement sensor array refers to a measurement network composed of 24 triangulation principle laser displacement sensors, with a measurement accuracy of 1 micron and a sampling frequency of 1000 Hz, which is used to capture tiny deformations on the surface of the carton.

5. The method for on-line automatic testing of the compressive strength of a corrugated cardboard box according to claim 4, characterized in that, The micro-deformation displacement data refers to a set of deformation displacement values ​​of six surfaces of the corrugated box under the test load collected by a high-precision laser displacement sensor array.

6. The method for online automatic testing of the compressive strength of corrugated cartons according to claim 5, characterized in that, The acoustic emission signal refers to the high-frequency sound wave signal generated by the internal fibers of the corrugated box under the test load, with a frequency range of 20kHz to 200kHz, which is collected by a piezoelectric sensor array.

7. The method for on-line automatic testing of the compressive strength of corrugated cartons according to claim 6, characterized in that The frequency spectrum characteristic parameters refer to a set of characteristic parameters extracted after performing frequency spectrum analysis on the acoustic emission signal, including energy distribution in each frequency band, peak frequency, center frequency and bandwidth.

8. The method for on-line automatic testing of the compressive strength of corrugated cartons according to claim 7, characterized in that, The physical mechanism model of the carton fiber structure refers to a stress-strain analysis model based on the anisotropic mechanical properties of paper materials, taking into account the interlayer structural characteristics and fiber arrangement direction of corrugated paper.

9. The method for on-line automatic testing of the compressive strength of corrugated cardboard boxes according to claim 8, characterized in that, The stress distribution calculation equation is based on Hooke's law and is used to calculate the stress distribution state of the corrugated box wall under the condition of micro-deformation displacement data. The input includes micro-deformation displacement data, material elastic modulus tensor, Poisson's ratio parameter, corrugated structure geometric parameters and test load size, and the output is three-dimensional stress field distribution and strain energy density distribution.

10. The method for online automatic testing of the compressive strength of corrugated cardboard boxes according to claim 9, characterized in that, The hierarchical attention carton strength assessment model refers to a deep learning model that combines spatial hierarchical feature extraction and attention mechanism, and is used to predict the compressive strength of corrugated boxes from micro-deformation displacement data and spectral feature parameters; its hierarchical structure takes into account the multi-layer composite structure characteristics of corrugated boxes, and forms a corresponding relationship with the corrugated wave distance and corrugated wall thickness parameters in the corrugated structure geometric parameters through the feature extraction capabilities of different scales.

Images