A method and system for monitoring the tightening force of a banding machine

Abstract

The present application relates to the technical field of data processing, and discloses a method and system for monitoring the tightening force of a banding machine. The method includes: obtaining stress-deformation data through a high-frequency sampling array to form a time series feature matrix; reconstructing the signal with an adaptive filter to obtain a key feature point set; constructing a prediction engine with a deep tensor network based on the feature point set; using the prediction engine to identify waveform anomalies to obtain a deviation feature spectrum; generating a compensation strategy map through a hybrid decision system according to the feature spectrum; and obtaining a steady-state control scheme through a parameter self-correction algorithm based on the strategy map. The present application realizes precise tightening force control for different materials by constructing a high-frequency tightening force monitoring method based on multi-sensor fusion and an adaptive control technology, and improves the working efficiency and bundling quality of the banding machine.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a method and system for monitoring the tightening force of a strapping machine. Background Art

[0002] With the continuous development of automated packaging technology, strapping machines, as important packaging equipment, are widely used in fields such as logistics, warehousing, and manufacturing. Traditional strapping machine tightening force control mainly relies on mechanical fixed force devices or simple closed-loop electrical control systems to achieve the strapping of different materials by setting fixed tightening force parameters. This type of control method usually uses a tension sensor to monitor the strap tension in real time, and compares the measured value with a preset value through a comparator. When the preset tension value is reached, the cutting and heat sealing processes are triggered to complete the strapping operation. In recent years, some advanced strapping machines have begun to introduce PLC control systems and touch screen human-machine interfaces, allowing operators to manually adjust the tightening force setting value according to different materials, improving the adaptability of the equipment.

[0003] However, there are still many deficiencies in the existing technology for monitoring the tightening force of strapping machines. First, the traditional single-sensor monitoring method cannot comprehensively capture the complex interaction between the strap and the object being strapped, resulting in low tightening force control accuracy; second, the control strategy with fixed parameters is difficult to adapt to the changing characteristics of different materials, easily causing problems such as over-tightening or under-tightening of the strapping; third, there is a lack of an intelligent fault diagnosis mechanism, making it difficult to quickly locate the cause of the fault and automatically adjust when an abnormality occurs during the tightening process; fourth, the response lag and oscillation problems during the control process have not been effectively solved, affecting the strapping efficiency and quality. These problems are particularly obvious when dealing with high-value, easily deformable, or special-shaped materials, severely limiting the application of strapping machines in the fields of high-end manufacturing and precision packaging. Summary of the Invention

[0004] This application provides a method and system for monitoring the tightening force of a strapping machine, which is used to solve the technical problems of low tightening force control accuracy, poor material adaptability, and lack of intelligent fault diagnosis and automatic adjustment ability in the existing technology for strapping machines. By constructing a high-frequency tightening force monitoring method based on multi-sensor fusion and adaptive control technology, precise tightening force control for different materials is achieved, improving the working efficiency and strapping quality of the strapping machine.

[0005] In a first aspect, the present application provides a method for monitoring the tightening force of a strapping machine. The method for monitoring the tightening force of the strapping machine includes: collecting stress-deformation coupling data during the operation of the strapping machine through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix; according to the multi-dimensional tightening force time series feature matrix, performing signal reconstruction and dynamic threshold segmentation through an adaptive filtering algorithm to obtain a set of key feature points in the tightening process; based on the set of key feature points in the tightening process, constructing a tightening force dynamic response model through a cross-validated deep tensor network to obtain an accurate prediction engine with material adaptability; based on the accurate prediction engine, performing multi-modal abnormal pattern recognition on the real-time collected tightening force waveform to obtain a hierarchical tightening deviation feature spectrum; according to the hierarchical tightening deviation feature spectrum, performing intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system to obtain a tightening force deviation compensation strategy map; based on the tightening force deviation compensation strategy map, generating a closed-loop control instruction for the strapping machine through a parameter self-correcting iterative algorithm to obtain a steady-state control scheme for the tightening force.

[0006] In a second aspect, the present application provides a system for monitoring the tightening force of a strapping machine. The system for monitoring the tightening force of the strapping machine includes:

[0007] A coupling module for collecting stress-deformation coupling data during the operation of the strapping machine through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix;

[0008] A reconstruction module for performing signal reconstruction and dynamic threshold segmentation through an adaptive filtering algorithm according to the multi-dimensional tightening force time series feature matrix to obtain a set of key feature points in the tightening process;

[0009] A construction module for constructing a tightening force dynamic response model through a cross-validated deep tensor network based on the set of key feature points in the tightening process to obtain an accurate prediction engine with material adaptability;

[0010] An identification module for performing multi-modal abnormal pattern recognition on the real-time collected tightening force waveform based on the accurate prediction engine to obtain a hierarchical tightening deviation feature spectrum;

[0011] A diagnosis module for performing intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system according to the hierarchical tightening deviation feature spectrum to obtain a tightening force deviation compensation strategy map;

[0012] A generation module for generating a closed-loop control instruction for the strapping machine through a parameter self-correcting iterative algorithm based on the tightening force deviation compensation strategy map to obtain a steady-state control scheme for the tightening force.

[0013] A third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory to cause the computer device to execute the above-mentioned method for monitoring the tightening force of the strapping machine.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the instructions are run on a computer, the computer is caused to execute the above-mentioned method for monitoring the tightening force of the strapping machine.

[0015] In the technical solution provided by this application, the stress-deformation coupling data during the operation of the strapping machine is collected through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix, which not only realizes the comprehensive monitoring of the physical state during the tightening process, but also provides a high-quality raw data basis for subsequent analysis, overcoming the limitations of the traditional single-sensor monitoring method; an adaptive filtering algorithm is used for signal reconstruction and dynamic threshold segmentation to obtain a set of key feature points during the tightening process, effectively filtering out environmental interference and mechanical vibration noise, greatly improving the signal quality and the accuracy of feature extraction, making the identification of key operating conditions more accurate and reliable; a tightening force dynamic response model is constructed through a cross-validated deep tensor network to obtain an accurate prediction engine with material self-adaptability. This deep learning model can automatically learn the mechanical properties and response laws of different materials, enabling the strapping machine to intelligently adapt to the tightening requirements of various materials, significantly improving the versatility and adaptability of the equipment; based on the accurate prediction engine, multi-modal anomaly pattern recognition is performed on the real-time collected tightening force waveform to obtain a hierarchical tightening deviation feature spectrum, realizing the early detection and accurate classification of abnormal situations, providing a reliable basis for fault diagnosis; intelligent fault diagnosis is carried out through a hybrid decision tree-fuzzy inference system to obtain a tightening force deviation compensation strategy map. This hybrid intelligent algorithm fully combines the clear logical reasoning ability of the decision tree and the uncertainty processing ability of the fuzzy inference system, can accurately identify the cause of the fault and generate targeted compensation strategies, greatly improving the accuracy and efficiency of fault handling; finally, a closed-loop control instruction for the strapping machine is generated through a parameter self-correction iterative algorithm to obtain a steady-state control scheme for the tightening force, realizing the automatic optimization and real-time adjustment of the control parameters, ensuring the stability and consistency of the tightening force under various working conditions. It is particularly worth emphasizing that the deep tensor network algorithm applied in this solution innovatively combines tensor decomposition and deep learning technologies. It can not only efficiently process multi-dimensional data structures and capture the complex coupling relationships between different physical quantities, but also significantly enhance the generalization ability of the model through the K-fold cross-validation technique, enabling the prediction engine to accurately handle new materials and working conditions that have not been seen before; the introduction of the genetic-simulated annealing hybrid algorithm cleverly combines the complementary advantages of the two optimization algorithms, achieving a good balance between global search and local fine-tuning, providing an efficient and stable calculation method for the optimization of control parameters, improving the intelligent level and working efficiency of the tightening force monitoring and control of the strapping machine, and realizing the accurate tightening force control of different materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 Schematic diagram of an embodiment of the method for monitoring the tightening force of a belt tying machine in an embodiment of the present application;

[0018] Figure 2 Schematic diagram of an embodiment of the system for monitoring the tightening force of a belt tying machine in an embodiment of the present application;

[0019] Figure 3 is a structural schematic block diagram of a computer device in an embodiment of the present invention. Specific embodiments

[0020] The embodiments of the present application provide a method and a system for monitoring the tightening force of a belt tying machine. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and the above-mentioned drawings of the present application are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order different from that shown or described herein. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these process, method, product or device.

[0021] For ease of understanding, the specific process of the embodiments of the present application is described below. Please refer to Figure 1 , an embodiment of the method for monitoring the tightening force of a belt tying machine in an embodiment of the present application includes:

[0022] Step S101: Collect stress-deformation coupling data during the operation of the belt tying machine through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix;

[0023] Step S102: According to the multi-dimensional tightening force time series feature matrix, perform signal reconstruction and dynamic threshold segmentation through an adaptive filtering algorithm to obtain a set of key feature points in the tightening process;

[0024] Step S103: Based on the set of key feature points in the tightening process, construct a tightening force dynamic response model through a cross-validated deep tensor network to obtain an accurate prediction engine with material adaptability;

[0025] Step S104: Based on the accurate prediction engine, perform multi-modal abnormal pattern recognition on the real-time collected tightening force waveform to obtain a hierarchical tightening deviation feature spectrum;

[0026] Step S105: According to the hierarchical tightening deviation feature spectrum, perform intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system to obtain a tightening force deviation compensation strategy diagram;

[0027] Step S106: According to the bundling force deviation compensation strategy diagram, generate a closed-loop control instruction for the bundling machine through a parameter self-tuning iterative algorithm to obtain a steady-state control scheme for the bundling force.

[0028] It can be understood that the execution subject of this application can be the bundling force monitoring system of the bundling machine, or it can also be a terminal or a server, and it is not specifically limited here. In this embodiment of the application, the server is taken as the execution subject for illustration.

[0029] Specifically, high-frequency sampling sensor array data collection is carried out. By arranging pressure sensors, displacement sensors and angle sensors on the bundling device of the bundling machine to form a high-frequency sampling sensor array, these sensors synchronously collect the contact stress and deformation data between the bundling belt and the object to be bundled at a frequency of 1000 Hz. The collected original data will be added with unified timestamp information to form a time-series related multi-source bundling force parameter stream, and then analog-to-digital conversion processing is carried out to generate a digital stress-deformation sampling lattice. Tensor arrangement and matrix reconstruction are performed on these lattices to construct a bundling state cube containing three-dimensional information of pressure, displacement and angle. Finally, dimension reduction and data enhancement are carried out on the bundling state cube through a feature extraction method to obtain a multi-dimensional bundling force time-series feature matrix. Based on the multi-dimensional bundling force time-series feature matrix, signal reconstruction and dynamic threshold segmentation are carried out through an adaptive filtering algorithm. In this stage, multi-scale decomposition of wavelet transform is carried out on the bundling force time-series feature matrix to obtain the frequency-domain component spectrum of the bundling force signal, the noise energy distribution map is calculated according to the frequency-domain component spectrum, and the signal-to-noise ratio parameters of each frequency band are determined. An adaptive band-pass filter bank is constructed according to the signal-to-noise ratio parameters, a multi-band filtering template is generated and applied to the bundling force time-series feature matrix to form a denoised smooth curve of the bundling force. Then, sliding window analysis is carried out on the smooth curve, the change rate of the curve slope is calculated, the bundling force mutation points are marked, and finally the key working condition points in the bundling process are extracted through a dynamic threshold segmentation algorithm to form a set of key feature points in the bundling process.

[0030] Construct a dynamic response model of the tying force through a cross-validated deep tensor network. Convert the key feature point set of the tying process into a tensor representation form, construct a training matrix of tying force data, perform data augmentation and normalization processing on this matrix to generate a model training sample library. Design the structure of the deep tensor network according to the model training sample library to form a neural hierarchy architecture of the tying force, and train and validate the neural hierarchy architecture of the tying force through the K-fold cross-validation method to obtain a tying force model with optimized parameters. Construct a material property compensation function based on the tying force model with optimized parameters, generate a table of adaptation coefficients for multiple types of materials, and integrate the table of adaptation coefficients for multiple types of materials into the tying force model with optimized parameters to obtain an accurate prediction engine with material self-adaptive ability. Conduct multi-modal anomaly pattern recognition on the tying force waveform collected in real time. Calculate the difference between the standard tying force prediction value generated by the accurate prediction engine and the tying force waveform collected in real time to obtain the original tying force deviation sequence, extract the waveform features from this sequence through time-frequency analysis methods to form a tying force deviation feature vector. Construct a multi-dimensional anomaly index system based on the tying force deviation feature vector, generate a tying deviation evaluation matrix, classify the anomaly patterns through clustering analysis of this matrix to obtain a tying anomaly type library. Match the similarity between the tying anomaly type library and the current tying force deviation to identify the current anomaly pattern category, and perform severity grading and spatial distribution analysis on the tying force deviation according to the anomaly pattern category to obtain a hierarchical tying deviation feature spectrum. Conduct intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system based on the hierarchical tying deviation feature spectrum. Input the hierarchical tying deviation feature spectrum into the fault feature parser to extract the tying fault symptom feature set, construct a decision tree structure according to the tying fault symptom feature set to form a preliminary diagnosis path map of the tying fault. Quantify the uncertain nodes in the preliminary diagnosis path map of the tying fault through a fuzzy membership function to obtain a tying fault probability distribution table, perform Bayesian probability inference based on the tying fault probability distribution table, calculate the occurrence probability ranking of each fault type, identify the main fault source from it, generate the tying force fault diagnosis result, and perform compensation parameter calculation and strategy mapping on the tying force fault diagnosis result to obtain a tying force deviation compensation strategy map.

[0031] According to the bundling force deviation compensation strategy diagram, the closed-loop control instruction of the strapping machine is generated through the parameter self-correction iterative algorithm. The torque transfer curve and response delay parameters are extracted from the bundling force deviation compensation strategy diagram, and a tension-relaxation dual-state control mapping table is constructed. The physical constraint boundary of this mapping table is tested to form a parameter adjustment interval. The genetic-simulated annealing hybrid algorithm is used to globally search and locally fine-tune the control variables within the parameter adjustment interval to obtain the multi-scale bundling force correction amount. This correction amount is dynamically smoothed through a non-linear response compensation function to generate an anti-vibration progressive adjustment curve. According to the progressive adjustment curve, a feedback gain adaptive adjustment mechanism is designed to output a parameter correction control quantity with time-varying characteristics. After the time-domain and frequency-domain joint analysis and optimization integration of the parameter correction control quantity and historical bundling force data, a steady-state control scheme for the bundling force is obtained.

[0032] For example, in the operation of strapping cartons by the strapping machine in a certain packaging factory, the original data collected by the high-frequency sampling sensor array includes the pressure change curve between the strap and the carton recorded by the pressure sensor, the strap tightening displacement recorded by the displacement sensor, and the data of the angle change of the strapping machine chuck recorded by the angle sensor. These data form a three-dimensional data matrix after timestamp marking and analog-to-digital conversion. Wavelet decomposition is used to analyze the noise, and it is identified that the effective signal is below 40 Hz. A band-pass filter bank of 40 - 60 Hz is constructed to filter out the vibration noise in the working environment, making the bundling force curve smoother. The smoothed curve is analyzed by a sliding window. When the change rate of the bundling force exceeds 0.5 N / ms, it is marked as a key feature point. These feature points are input into a deep tensor network trained by K = 5-fold cross-validation to construct a dynamic response model, which can automatically adjust the bundling parameters for different materials such as corrugated cartons and plastic boxes. When there is a deviation between the real-time bundling force and the predicted value, the deviation characteristics are extracted through time-frequency analysis, and it is found that the friction force of the strap increases abnormally by matching with the abnormal library. The intelligent diagnosis system determines that the strap tension controller is worn through decision tree and fuzzy inference, generates a corresponding compensation strategy, and adjusts the motor output torque through the parameter self-correction algorithm to achieve stable bundling force control.

[0033] In the embodiments of the present application, stress-deformation coupling data during the operation of the strapping machine is collected through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time-series feature matrix, which not only realizes the comprehensive monitoring of the physical state during the tightening process but also provides a high-quality raw data basis for subsequent analysis, overcoming the limitations of traditional single-sensor monitoring methods; an adaptive filtering algorithm is used for signal reconstruction and dynamic threshold segmentation to obtain a key feature point set of the tightening process, effectively filtering out environmental interference and mechanical vibration noise, greatly improving the signal quality and the accuracy of feature extraction, and making the identification of key working condition points more accurate and reliable; a tightening force dynamic response model is constructed through a cross-validated deep tensor network to obtain an accurate prediction engine with material self-adaptability. This deep learning model can automatically learn the mechanical properties and response laws of different materials, enabling the strapping machine to intelligently adapt to the tightening requirements of various materials, significantly improving the versatility and adaptability of the equipment; based on the accurate prediction engine, multi-modal anomaly pattern recognition is performed on the real-time collected tightening force waveform to obtain a hierarchical tightening deviation feature spectrum, realizing the early detection and accurate classification of abnormal situations and providing a reliable basis for fault diagnosis; intelligent fault diagnosis is carried out through a hybrid decision tree-fuzzy inference system to obtain a tightening force deviation compensation strategy map. This hybrid intelligent algorithm fully combines the clear logical reasoning ability of the decision tree and the uncertainty processing ability of the fuzzy inference system, can accurately identify the cause of the fault and generate targeted compensation strategies, greatly improving the accuracy and efficiency of fault handling; finally, a closed-loop control instruction for the strapping machine is generated through a parameter self-correction iterative algorithm to obtain a steady-state control scheme for the tightening force, realizing the automatic optimization and real-time adjustment of control parameters and ensuring the stability and consistency of the tightening force under various working conditions. It is particularly worth emphasizing that the deep tensor network algorithm applied in this solution innovatively integrates tensor decomposition and deep learning technologies, which can not only efficiently process multi-dimensional data structures, capture the complex coupling relationships between different physical quantities, but also significantly enhance the generalization ability of the model through the K-fold cross-validation technology, enabling the prediction engine to accurately handle unseen new materials and working conditions; the introduction of the genetic-simulated annealing hybrid algorithm cleverly combines the complementary advantages of the two optimization algorithms, achieving a good balance between global search and local fine-tuning, providing an efficient and stable calculation method for the optimization of control parameters, and improving the intelligent level and working efficiency of the tightening force monitoring and control of the strapping machine, realizing the precise tightening force control of different materials.

[0034] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0035] (1) Arrange pressure sensors, displacement sensors, and angle sensors on the strapping device of the strapping machine to form a high-frequency sampling sensor array, and collect the contact stress and deformation data between the strap and the object to be strapped;

[0036] (2) Synchronously sample the output signals of the high-frequency sampling sensing array through the signal acquisition circuit at a frequency of 1000 Hz to obtain the original stress-deformation coupled data;

[0037] (3) Add unified timestamp information to the original stress-deformation coupled data to form a multi-source tightening force parameter stream with temporal correlation;

[0038] (4) Process the multi-source tightening force parameter stream through analog-to-digital conversion to generate a digital stress-deformation sampling lattice;

[0039] (5) Perform tensor arrangement and matrix reconstruction on the digital stress-deformation sampling lattice to construct a tightening state cube containing three-dimensional information of pressure, displacement, and angle;

[0040] (6) Reduce the dimension and enhance the data of the tightening state cube through feature extraction methods to obtain a multi-dimensional tightening force temporal feature matrix.

[0041] Specifically, reasonably arrange pressure sensors, displacement sensors, and angle sensors on the tightening device of the strapping machine to form a high-frequency sampling sensing array. The pressure sensor is mainly used to measure the contact stress between the strap and the object to be strapped, the displacement sensor is used to measure the displacement change of the strap during the tightening process, and the angle sensor is used to measure the angle change of the chuck of the strapping machine. The cooperation of these three sensors can comprehensively reflect the physical state changes during the tightening process. Synchronously sample the output signals of the high-frequency sampling sensing array through the signal acquisition circuit at a frequency of 1000 Hz. This high-frequency synchronous sampling can ensure capturing the rapid change details during the strapping machine tightening process and obtaining accurate original stress-deformation coupled data. The sampling frequency of 1000 Hz means that 1000 data points are collected per second, which can effectively capture the transient changes during the high-speed operation of the strapping machine.

[0042] Add unified timestamp information to the original stress-deformation coupling data. By adding accurate time markers to each sampling point, data from different sensors can be accurately aligned in the time dimension, forming a multi-source binding force parameter stream with temporal correlation. This ensures that the relationship between physical quantities changing over time can be accurately restored during the data analysis process. Process the multi-source binding force parameter stream through analog-to-digital conversion to convert the analog signal into a digital signal, generating a digital stress-deformation sampling lattice. During the analog-to-digital conversion process, a high-precision ADC converter is used to ensure data accuracy, and anti-aliasing filtering technology is applied to avoid high-frequency interference. Perform tensor arrangement and matrix reconstruction on the digital stress-deformation sampling lattice. Rearrange the discrete data points according to the spatial and time dimensions to construct a binding state cube containing three-dimensional information of pressure, displacement, and angle. This three-dimensional data structure can comprehensively reflect the state changes during the binding process. Perform dimensionality reduction and data enhancement on the binding state cube through feature extraction methods to obtain a multi-dimensional binding force time series feature matrix. During the feature extraction process, dimensionality reduction techniques such as principal component analysis (PCA) are applied to reduce data redundancy, extract key features, and at the same time enrich the training data through data enhancement technology to enhance the generalization ability of the model. The generated multi-dimensional binding force time series feature matrix will contain key feature information of the binding process of the strapping machine.

[0043] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0044] (1) Perform multi-scale decomposition on the multi-dimensional binding force time series feature matrix through wavelet transform to obtain the frequency domain component spectrum of the binding force signal;

[0045] (2) Calculate the noise energy distribution map according to the frequency domain component spectrum to determine the signal-to-noise ratio parameters of each frequency band;

[0046] (3) Construct an adaptive band-pass filter bank based on the signal-to-noise ratio parameters to generate a multi-band filtering template;

[0047] (4) Apply the multi-band filtering template to the multi-dimensional binding force time series feature matrix to form a denoised binding force smooth curve;

[0048] (5) Perform sliding window analysis on the binding force smooth curve, calculate the curve slope change rate, and mark the binding force mutation points;

[0049] (6) According to the binding force mutation points, extract the key working condition points during the binding process through a dynamic threshold segmentation algorithm to obtain a set of key feature points of the binding process.

[0050] Specifically, the multi-dimensional tightening force time-series feature matrix is decomposed by wavelet transform at multiple scales. Wavelet transform is a time-frequency analysis tool that can analyze signal features at different scales. In the tightening force monitoring of the strapping machine, the discrete wavelet transform is used to decompose the tightening force time-series feature matrix into sub-signals of different frequency bands, thereby obtaining the frequency-domain component spectrum of the tightening force signal. The specific operation is to select a wavelet basis function suitable for the characteristics of the tightening force signal, such as the Daubechies wavelet or the Symlet wavelet, and decompose the tightening force time-series data into high-frequency detail coefficients and low-frequency approximation coefficients through high-pass and low-pass filters, recursively decompose the low-frequency approximation coefficients to form a multi-layer decomposition structure, and obtain the frequency-domain component spectrum. Calculate the noise energy distribution map according to the frequency-domain component spectrum, and identify the effective components and noise components in the signal by analyzing the energy distribution characteristics of each frequency band. In the tightening force monitoring of the strapping machine, the high-frequency part usually contains noises such as environmental interference and mechanical vibration, while the low-frequency part contains the effective information of the tightening force. During the calculation process, the energy of each frequency band is statistically analyzed and compared with the known tightening force signal pattern to calculate the signal-to-noise ratio parameters of each frequency band. The signal-to-noise ratio represents the ratio of the effective signal energy to the noise energy. The frequency band with a high signal-to-noise ratio indicates that it contains more effective information. According to the signal-to-noise ratio parameters of each frequency band, a set of band-pass filters is designed, and each filter corresponds to a specific frequency band. In the tightening force monitoring of the strapping machine, the center frequency and bandwidth of the filter are dynamically adjusted according to the signal-to-noise ratio. A narrower passband is used for the frequency band with a high signal-to-noise ratio to retain more details, while a wider passband is used for the frequency band with a low signal-to-noise ratio for smoothing. This set of filters constitutes a multi-band filtering template, which can perform adaptive processing according to the noise characteristics under different environments and working conditions. Apply the multi-band filtering template to the multi-dimensional tightening force time-series feature matrix, filter out the irrelevant noise components by performing convolution operations on each frequency band of the time-series feature matrix with the corresponding filtering template, and retain the effective tightening force signal. The filtered signals of each frequency band are then reconstructed and synthesized into a denoised smoothing curve of the tightening force. This smoothing curve retains the main characteristics of the original tightening force change, while significantly reducing the influence of noise interference, providing a high-quality data basis for subsequent analysis.

[0051] The fifth step is to perform a sliding window analysis on the smooth curve of the tightening force. A window of a fixed size is set and slid point by point along the time axis to calculate the slope of the tightening force curve within the window and the rate of change of the slope. The slope represents the rate of change of the tightening force over time, and the rate of change of the slope reflects the acceleration of the change in the tightening force. During the tightening process of the strapping machine, when the rate of change of the slope exceeds a certain threshold, it indicates that a sudden change in the tightening force has occurred. These mutation points usually correspond to key state changes during the tightening process, such as the start of tightening, reaching the preset tension, and completing the tightening. Finally, based on the mutation points of the tightening force, key operating condition points during the tightening process are extracted through a dynamic threshold segmentation algorithm. The dynamic threshold segmentation algorithm sets different threshold criteria for different stages according to the statistical characteristics of the tightening force data and the historical operating condition characteristics. The algorithm performs clustering analysis on the marked mutation points of the tightening force, identifies groups of mutation points with similar characteristics, and then sets corresponding threshold conditions according to the characteristics of each group of mutation points to screen out the key points that truly represent the operating condition changes, forming a set of key characteristic points for the tightening process. These sets of key characteristic points accurately characterize each key stage of the tightening process of the strapping machine.

[0052] For example, when strapping a packing box, the obtained multi-dimensional tightening force time series feature matrix contains 30,000 data points collected by three sensors at a sampling rate of 1000 Hz during a 10-second tightening process. Using 5-level Daubechies-4 wavelet decomposition through wavelet transform, these data are decomposed into five frequency bands: 0-10 Hz, 10-20 Hz, 20-40 Hz, 40-80 Hz, and 80-160 Hz. Analysis shows that the 0-40 Hz frequency band mainly contains the effective signal of the tightening force, with a signal-to-noise ratio of about 15 dB, while the 40-160 Hz frequency band mainly contains mechanical vibration and electrical interference, with a signal-to-noise ratio of about 3 dB. Based on this, an adaptive band-pass filter bank is constructed. A narrower passband is used for the 0-40 Hz frequency band to retain details, and intensity suppression is performed on the 40-160 Hz frequency band. After applying the filter bank, the smooth curve of the tightening force clearly shows three stages of the tightening process: initial tightening, maintaining tension, and completing locking. Using a sliding window analysis with a window size of 200 ms, the rate of change of the slope is calculated. When the rate of change exceeds 0.5 N / ms, it is marked as a mutation point, and a total of 7 mutation points are detected. By setting dynamic thresholds, 4 key operating condition points are extracted: the strap contacts the packing box (t = 1.2 s), reaches the initial tension (t = 3.5 s), reaches the target tension (t = 6.8 s), and completes locking (t = 8.7 s). These four points accurately describe the key state changes of the entire tightening process.

[0053] In a specific embodiment, the process of performing step S103 may specifically include the following steps:

[0054] (1) Convert the set of key characteristic points of the tightening process into a tensor representation form to construct a training matrix for the tightening force data;

[0055] (2) Perform data augmentation and normalization on the training matrix of the tightening force data to generate a model training sample library;

[0056] (3) Design a deep tensor network structure based on the model training sample library to form a tightening force neural hierarchical architecture;

[0057] (4) Train and validate the tightening force neural hierarchical architecture through the K-fold cross-validation method to obtain a parameter-optimized tightening force model;

[0058] (5) Construct a material property compensation function based on the parameter-optimized tightening force model to generate a multi-type material adaptation coefficient table;

[0059] (6) Integrate the multi-type material adaptation coefficient table into the parameter-optimized tightening force model to obtain an accurate prediction engine with material self-adaptive ability.

[0060] Specifically, convert the key feature point set of the tightening process into a tensor representation form to construct a training matrix of the tightening force data. Tensor representation is a multi-dimensional data structure that can express the information relationships of multiple dimensions simultaneously. In the tightening force monitoring of the bundling machine, convert the key feature point set from the time series form to a third-order tensor structure. The first dimension represents the sample index, the second dimension represents the time window, and the third dimension represents the feature type (such as pressure, displacement, angle, etc.). Through this conversion, each tightening process is encoded as a multi-dimensional feature block, capturing the coupling relationships between different physical quantities and forming a training matrix of the tightening force data.

[0061] Data augmentation and normalization are performed on the training matrix of the tightening force data to generate a model training sample library. Data augmentation refers to the technology of generating more training samples through a series of transformations to enhance the generalization ability of the model. In the tightening force monitoring of the strapping machine, methods such as time window sliding, random noise addition, and signal scale scaling are used to expand the original training matrix of the tightening force data into a larger-scale dataset. Normalization processing is to convert features with different dimensions into a unified scale to eliminate the influence of dimensions. Common normalization methods include Z-score normalization (subtracting the mean and dividing by the standard deviation) and Min-Max normalization (linearly mapping to the 0-1 interval). The data after augmentation and normalization constitute the model training sample library, providing rich training materials for the subsequent deep learning model. Design a deep tensor network structure according to the model training sample library to form a tightening force neural hierarchical architecture. The deep tensor network is a special neural network designed specifically to process data in tensor form and can retain the multi-dimensional structural characteristics of the data. In the tightening force monitoring of the strapping machine, the designed network structure includes a tensor decomposition layer, a tensor convolution layer, and a fully connected layer. The tensor decomposition layer is used to reduce the model complexity and extract key features; the tensor convolution layer is used to capture local patterns in different dimensions; the fully connected layer is used to synthesize features in each dimension to achieve prediction. These layers are organized in a hierarchical manner to form a tightening force neural hierarchical architecture, which has powerful feature extraction and pattern recognition capabilities.

[0062] K-fold cross-validation is a model evaluation method that divides the dataset into K subsets. Each time, K - 1 subsets are used as the training set, and the remaining 1 subset is used as the validation set. This process is repeated K times so that each subset serves as the validation set. In the tightening force monitoring of the strapping machine, 5-fold or 10-fold cross-validation is adopted to comprehensively evaluate the performance of the model under different data partitions and avoid overfitting. During the training process, the backpropagation algorithm is used to adjust the network parameters to minimize the mean square error between the predicted value and the actual tightening force. After multiple rounds of iteration, a parameter-optimized tightening force model with good generalization ability is obtained. Based on the parameter-optimized tightening force model, a material property compensation function is constructed to generate a multi-type material adaptation coefficient table. The material property compensation function is used to adjust the output of the tightening force model to adapt to the characteristics of different materials. The compensation function can be expressed as:

[0063]

[0064] Among them, represents the adjusted tightening force value, represents the originally predicted tightening force value, represents the material property parameter vector, represents the weight coefficient of the i-th compensation term, represents a function about the original predicted value, represents a function regarding material characteristics, and n represents the number of compensation terms. By analyzing the response characteristics of different materials during the bundling process, the characteristic parameters of each material type are determined, and a multi-type material adaptation coefficient table is generated, which includes parameters such as the elastic coefficient, friction coefficient, hardness, etc. of each material and their corresponding compensation coefficients. The multi-type material adaptation coefficient table is integrated into the parameter-optimized bundling force model to obtain an accurate prediction engine with material self-adaptive ability. During the integration process, a material recognition mechanism and an adaptive adjustment mechanism are established. The material recognition mechanism identifies the type of the material being bundled currently by analyzing the characteristics of the force-deformation curve at the initial stage of bundling; the adaptive adjustment mechanism then looks up the corresponding compensation parameters from the material adaptation coefficient table according to the recognition result and applies them to the prediction model to achieve adaptive adjustment for different materials. The accurate prediction engine can accurately predict the mechanical response of different materials during the bundling process based on real-time monitoring data, providing an accurate control basis for the strapping machine.

[0065] For example, in the carton bundling link of a certain packaging production line, the key feature point sets of the bundling processes of 100 groups of cartons with different sizes and materials are converted into a tensor representation form. Each sample contains 20 time points × 3 types of sensor data (pressure, displacement, angle), forming a third-order tensor structure of 100×20×3. Data augmentation is performed by sliding the time window and adding 5% Gaussian noise, expanding the sample size to 500 groups, and using the Z-score method for standardization processing to make the mean of all features 0 and the standard deviation 1. Based on the processed data, a deep tensor network structure is designed, which includes 2 tensor decomposition layers (kernel size 3×2), 3 tensor convolution layers (sliding windows are 3, 2, 1 respectively), and 2 fully connected layers (the number of hidden units is 64 and 32 respectively). This network is trained using the 10-fold cross-validation method, and the average relative error on the validation set is reduced to 3.2%. It is found through analysis that for different materials such as corrugated cartons, plastic boxes, and wooden boxes, the bundling force model needs to be adjusted to varying degrees. Therefore, a material characteristic compensation function is constructed, considering the elasticity, friction, etc. of the materials, and an adaptation coefficient table for 8 common packaging materials is generated. These coefficients are integrated into the prediction model to form an accurate prediction engine. This engine can automatically apply the corresponding flexible compensation coefficient when detecting a corrugated carton and switch to a rigid compensation coefficient when detecting a wooden box, achieving accurate prediction and control of the bundling process for different materials.

[0066] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0067] (1) Calculate the difference between the standard bundling force prediction value generated by the accurate prediction engine and the bundling force waveform collected in real time to obtain the original bundling force deviation sequence;

[0068] (2) Extract waveform features from the original binding force deviation sequence through time-frequency analysis to form a binding force deviation feature vector;

[0069] (3) Construct a multi-dimensional anomaly index system based on the binding force deviation feature vector to generate a binding deviation evaluation matrix;

[0070] (4) Classify the anomaly patterns of the binding deviation evaluation matrix through cluster analysis to obtain a binding anomaly type library;

[0071] (5) Match the similarity between the binding anomaly type library and the current binding force deviation to identify the current anomaly pattern category;

[0072] (6) Perform severity grading and spatial distribution analysis on the binding force deviation according to the anomaly pattern category to obtain a hierarchical binding deviation feature spectrum.

[0073] Specifically, calculate the difference between the standard binding force prediction value generated by the precise prediction engine and the real-time collected binding force waveform to obtain the original binding force deviation sequence. The standard binding force prediction value is an ideal binding force curve generated by the precise prediction engine based on historical data and material characteristics, and the real-time collected binding force waveform is the binding force data actually measured during the operation of the strapping machine. The difference calculation is carried out in a point-to-point manner, subtracting the predicted value and the measured value at each time point to form a deviation sequence that changes with time. This original binding force deviation sequence reflects the difference between the actual binding process and the ideal state, and contains various deviation information during the operation of the strapping machine. Extract waveform features from the original binding force deviation sequence through time-frequency analysis to form a binding force deviation feature vector. Time-frequency analysis is a method that simultaneously analyzes the time-domain and frequency-domain characteristics of a signal, and can reveal the frequency characteristics of the signal changing with time. In the monitoring of the binding force of the strapping machine, time-frequency analysis techniques such as short-time Fourier transform (STFT) and wavelet transform are used to extract various characteristic parameters from the original binding force deviation sequence, including average deviation, maximum deviation, standard deviation of deviation, spectral energy distribution, main frequency component, etc. These characteristic parameters are combined into a high-dimensional feature vector to comprehensively describe various characteristics of the binding force deviation.

[0074] Construct a multi-dimensional anomaly index system based on the binding force deviation eigenvector and generate a binding deviation evaluation matrix. The multi-dimensional anomaly index system is a set of indexes used to evaluate the anomaly degree of the system, covering anomaly manifestations from different angles and levels. In the binding force monitoring of the banding machine, the constructed index system includes time-domain indexes (such as deviation amplitude, duration), frequency-domain indexes (such as high-frequency component ratio, spectrum distribution), and time-frequency combined indexes (such as energy concentration, modal distribution), etc. These indexes are arranged according to a certain organizational structure to form a binding deviation evaluation matrix. Each row represents a time point or time window, each column represents an evaluation index, and the value of the matrix element reflects the evaluation result of the corresponding index at a specific time. The fifth step is to classify the anomaly patterns of the binding deviation evaluation matrix through cluster analysis to obtain a binding anomaly type library. Cluster analysis is an unsupervised learning method that classifies similar samples into the same category by calculating the similarity between samples. In the binding force monitoring of the banding machine, algorithms such as K-means clustering, hierarchical clustering, or density clustering are used to classify the row vectors of the binding deviation evaluation matrix, and deviation patterns with similar characteristics are classified into the same category. The classification results form a binding anomaly type library, and each type represents a specific anomaly pattern, such as band slipping, material deformation, mechanical failure, etc.

[0075] Match the similarity between the bundling exception type library and the current bundling force deviation to identify the current exception mode category. Similarity matching is a pattern recognition method that determines the category to which a sample to be recognized belongs by calculating the degree of similarity between the sample to be recognized and the samples of known categories. In the bundling force monitoring of a strapping machine, calculate the similarity between the feature vector of the current bundling force deviation and the representative vectors of each type in the exception type library. Commonly used similarity metrics include Euclidean distance, cosine similarity, Mahalanobis distance, etc. Sort according to the similarity to identify the most likely mode category of the current exception, providing a basis for subsequent fault diagnosis and processing. Analyze the severity grading and spatial distribution of the bundling force deviation according to the exception mode category to obtain a hierarchical bundling deviation feature spectrum. Severity grading divides exceptions into different levels, such as minor, medium, severe, etc., according to factors such as the amplitude, duration, and influence range of the deviation. Spatial distribution analysis studies the distribution characteristics of the deviation at different positions and different stages, identifying the concentrated areas and propagation laws of the deviation. Considering the exception mode category, severity, and spatial distribution comprehensively, a hierarchical bundling deviation feature spectrum is formed. It is a hierarchical exception description structure that contains multi-level information such as exception type, degree, and distribution, providing a comprehensive basis for subsequent fault diagnosis and processing. In the bundling work of a strapping machine on a certain packaging production line, the precise prediction engine generated a standard bundling force curve based on the size and material of the current bundled carton. Theoretically, from the start of contact to the completion of locking, it should present a smooth rising and then stabilizing curve. However, the real-time collected bundling force waveform shows obvious fluctuations and drops in the middle section. By subtracting point by point, the original bundling force deviation sequence is obtained, and it is found that there is a negative deviation of up to 15N between 3.5 seconds and 5.2 seconds during the bundling process. Perform time-frequency analysis on this deviation sequence using wavelet transform, extract features such as the average deviation value, maximum deviation value, deviation duration, and spectral energy distribution, and form a 25-dimensional deviation feature vector. Based on this feature vector, construct a multi-dimensional exception index system including time-domain stability index, frequency-domain purity index, and time-frequency consistency index, and generate a bundling deviation evaluation matrix. Use the K-means clustering algorithm to analyze the historical collected deviation evaluation matrix, set K = 8, and obtain 8 typical bundling exception types, including belt slipping, material compression deformation, mechanical jamming, etc. Calculate the cosine similarity between the feature vector of the current deviation and the center vectors of these 8 exception types, and find that the similarity with the "belt slipping" type is the highest, reaching 0.92, confirming that the current exception is of the belt slipping type. According to the amplitude and duration of the deviation, determine that this slipping belongs to medium severity, and by analyzing the distribution of the deviation at different positions, determine that the slipping occurs in the contact area between the belt and the bottom of the carton, forming a hierarchical bundling deviation feature spectrum including exception type, severity, and location information.

[0076] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0077] (1) Input the hierarchical bundling deviation feature spectrum into a fault feature parser to extract the bundling fault symptom feature set;

[0078] (2) Construct a decision tree structure based on the bundling fault symptom feature set to form a preliminary bundling fault diagnosis path diagram;

[0079] (3) Quantify the uncertain nodes in the preliminary bundling fault diagnosis path diagram through a fuzzy membership function to obtain a bundling fault possibility distribution table;

[0080] (4) Perform Bayesian probability reasoning based on the bundling fault possibility distribution table and calculate the occurrence probability ranking of each fault type;

[0081] (5) Identify the main fault source from the occurrence probability ranking of the fault types to generate a bundling force fault diagnosis result;

[0082] (6) Calculate compensation parameters and perform strategy mapping on the bundling force fault diagnosis result to obtain a bundling force deviation compensation strategy diagram.

[0083] Specifically, input the hierarchical bundling deviation feature spectrum into a fault feature parser to extract the bundling fault symptom feature set. The fault feature parser is a data processing unit specifically used for analyzing fault features and can extract key fault symptoms from complex feature spectra. In the method for monitoring the bundling force of a strapping machine, the fault feature parser extracts symptom parameters directly related to fault diagnosis from the hierarchical bundling deviation feature spectrum through operations such as feature separation, dimensionality reduction, and recombination, including the time characteristics of the deviation (such as duration, occurrence frequency), amplitude characteristics (such as maximum deviation, average deviation), and spectral characteristics (such as main frequency components, harmonic distribution), etc. These parameters constitute the bundling fault symptom feature set and directly reflect the abnormal performance during the bundling process. Construct a decision tree structure based on the bundling fault symptom feature set to form a preliminary bundling fault diagnosis path diagram. A decision tree is a tree-structured classification model that gradually narrows down the fault range through a series of conditional judgments. In the monitoring of the bundling force of a strapping machine, the root node of the decision tree represents the initial fault judgment state, the internal nodes represent the test conditions for a certain fault feature, the branches represent different test results, and the leaf nodes represent the fault diagnosis results. The construction process determines the splitting criteria of the decision tree, and commonly used indicators include information gain, Gini coefficient, etc. Then, according to these criteria, the optimal feature is selected as the splitting node, and the subtrees are recursively constructed. The formed preliminary bundling fault diagnosis path diagram intuitively represents the reasoning path from symptoms to faults, but this path has the limitation of deterministic judgment.

[0084] Quantify the uncertain nodes in the preliminary diagnosis path diagram of the binding failure through the fuzzy membership function to obtain the binding failure possibility distribution table. The fuzzy membership function is a function used in fuzzy set theory to represent the membership degree of an element to a fuzzy set, and its value ranges from 0 to 1. In the binding force monitoring of the strapping machine, for the nodes with uncertainty in the decision tree, such as the relatively fuzzy concept of "high vibration frequency", the fuzzy membership function is introduced for quantification. Commonly used membership functions include triangular function, trapezoidal function, Gaussian function, etc. Select the appropriate function type according to the specific characteristic distribution. By calculating the membership degree of each eigenvalue to each failure type, the binding failure possibility distribution table is obtained. Each element in the table represents the possibility of a specific characteristic value leading to a specific failure. Based on the binding failure possibility distribution table, Bayesian probability inference is carried out to calculate the occurrence probability ranking of each failure type. Bayesian probability inference is a statistical inference method based on Bayes' theorem, which can combine prior knowledge and observed data to calculate the posterior probability. In the binding force monitoring of the strapping machine, the Bayesian probability inference process can be expressed as:

[0085]

[0086] Among them, represents the posterior probability of the failure type under the given symptom set S, represents the likelihood probability that the failure type causes the symptom set S to appear, represents the prior probability of the failure type , represents the total number of failure types. The likelihood probability can be obtained from the binding failure possibility distribution table, and the prior probability It comes from historical failure statistical data. Calculate the posterior probability of each failure type through the above formula, and sort them according to the probability size to obtain the occurrence probability ranking of each failure type. The main failure source refers to one or several failure types with the highest probability, which are the most likely reasons for the current abnormal bundling force. In the bundling force monitoring of the strapping machine, set a probability threshold (such as 0.6), and identify the failure types with probabilities exceeding the threshold as the main failure sources. If the probabilities of no failure types exceed the threshold, select the combination of the first few failure types with the highest probabilities as the composite failure source. The identified main failure sources, together with information such as corresponding symptoms, positions, and severities, constitute the bundling force fault diagnosis result. Perform compensation parameter calculation and strategy mapping on the bundling force fault diagnosis result to obtain the bundling force deviation compensation strategy diagram. Compensation parameter calculation is to determine the adjustment amount of control parameters required to correct the bundling force deviation according to the failure type and severity. Strategy mapping is to convert the compensation parameters into a specific sequence of control operation instructions. In the bundling force monitoring of the strapping machine, different compensation strategies are adopted according to different failure types, such as adjusting the motor output torque, modifying the strap pre-tightening force, and adjusting the bundling speed. These compensation strategies are arranged in chronological order and spatial positions to form the bundling force deviation compensation strategy diagram, providing intuitive operation guidance for subsequent closed-loop control.

[0087] For example, during the bundling process of a strapping machine on an industrial packaging line, the hierarchical bundling deviation characteristic spectrum shows that there is a periodic decrease in the force value in the middle of the bundling process. The decrease range is between 10 - 15 N, and the frequency is about 2 Hz. Inputting these characteristics into the fault characteristic parser, 10 key symptom parameters including periodicity (2 Hz), amplitude (10 - 15 N), duration (2.5 seconds), occurrence stage (middle of bundling), etc. are extracted to form a bundling fault symptom characteristic set. According to these characteristics, a decision tree is constructed to judge whether the deviation is periodic, then judge the frequency range, then judge the amplitude size, and finally judge the occurrence stage, forming a preliminary diagnostic path diagram containing 15 nodes. For uncertain nodes such as "frequency is about 2 Hz", a Gaussian-type fuzzy membership function is used for quantification. The membership degrees of the actual frequency of 1.8 Hz to "low-frequency vibration (1 - 3 Hz)" and "medium-frequency vibration (3 - 10 Hz)" are calculated as 0.85 and 0.15 respectively. Similarly, other uncertain nodes are processed to obtain a possibility distribution table containing 8 possible fault types. According to the prior probabilities of each fault type in historical data (such as motor vibration 0.25, belt slipping 0.20, mechanical looseness 0.15, etc.) and the likelihood probability of the current symptoms, the posterior probability of motor vibration is calculated as 0.68, belt slipping is 0.22, and the probabilities of other types are relatively low. It is determined that the motor vibration is the main fault source, and combined with its occurrence location (drive motor) and severity (medium), a detailed fault diagnosis result is generated. For the motor vibration fault, the compensation parameters that need to be adjusted are calculated, including reducing the motor speed by 10%, increasing the damping coefficient of the PID controller, adjusting the initial belt tension, etc., to form a bundling force deviation compensation strategy diagram containing time-sequential operation instructions.

[0088] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0089] (1) Extract the torque transmission curve and response delay parameters from the bundling force deviation compensation strategy diagram to construct a tension-relaxation two-state control mapping table;

[0090] (2) Conduct a physical constraint boundary test on the tension-relaxation two-state control mapping table to form a parameter adjustment interval;

[0091] (3) Through a genetic-simulated annealing hybrid algorithm, globally search and locally fine-tune the control variables within the parameter adjustment interval to obtain a multi-scale bundling force correction amount;

[0092] (4) Dynamically smooth the multi-scale bundling force correction amount through a non-linear response compensation function to generate an anti-vibration progressive adjustment curve;

[0093] (5)Design a feedback gain adaptive adjustment mechanism according to the progressive adjustment curve, and output a parameter correction control quantity with time-varying characteristics;

[0094] (6)Perform time-domain and frequency-domain joint analysis on the parameter correction control quantity and historical binding force data, and then optimize and integrate them to obtain a steady-state control scheme for the binding force.

[0095] Specifically, extract the torque transfer curve and response delay parameters from the binding force deviation compensation strategy diagram, and construct a tension-relaxation dual-state control mapping table. The torque transfer curve is a function curve that describes the relationship between the output torque of the driving motor of the strapping machine and the actual binding force, reflecting how the torque is converted into the actual binding effect. The response delay parameter represents the time required for the system to generate an actual response from receiving a control signal, including motor startup delay, mechanical transmission delay, and strapping deformation delay, etc. Tension-relaxation dual-state control is two basic states in the binding process of the strapping machine. The tension state is responsible for increasing the binding force, and the relaxation state is used to release excessive binding force. By analyzing the adjustment instructions in the binding force deviation compensation strategy diagram, extract the torque input values and corresponding binding force output values under different working conditions, as well as the time delay of state conversion, and construct a tension-relaxation dual-state control mapping table, providing a basic data structure for subsequent precise control. Conduct a physical constraint boundary check on the tension-relaxation dual-state control mapping table to form a parameter adjustment interval. Physical constraint boundaries refer to the physical limitations that must be followed in parameter adjustment in the actual system, including the performance limits of the equipment itself, the safe operating range, and the material bearing capacity, etc. In the monitoring of the binding force of the strapping machine, the physical constraints to be checked include the maximum motor output torque, the maximum tension that the strapping can withstand, the maximum binding speed, the minimum response time, etc. During the inspection process, determine the physical limit values of each parameter, and then compare the parameters in the tension-relaxation dual-state control mapping table with these limit values to ensure that all parameters are within a safe and reasonable range. For parameters outside the range, perform truncation or scaling processing to form a parameter adjustment interval that conforms to physical constraints, ensuring that unrealistic control parameters will not be generated in the subsequent optimization process.

[0096] Perform global search and local fine-tuning on the control variables within the parameter adjustment interval through a genetic-simulated annealing hybrid algorithm to obtain a multi-scale binding force correction quantity. The genetic-simulated annealing hybrid algorithm is an optimization method that combines the global search ability of the genetic algorithm and the local fine-tuning ability of the simulated annealing algorithm. In the monitoring of the binding force of the strapping machine, the optimization process of this hybrid algorithm can be expressed as:

[0097]

[0098] Among them, represents the optimization objective function of the hybrid algorithm, represents the parameter set of the genetic algorithm, Represents the parameter set of the simulated annealing algorithm. Represents the weight coefficient of the i-th control variable. Represents the mixing ratio of the genetic algorithm and the simulated annealing algorithm. Represents when the temperature is The optimization result of the genetic algorithm for the i-th control variable. Represents when the temperature is The optimization result of the simulated annealing algorithm for the i-th control variable. n represents the total number of control variables. The genetic algorithm is used for global search within the parameter adjustment range, and possible optimal solution regions are found through selection, crossover, and mutation operations. Then, the simulated annealing algorithm is used for local fine-tuning within these regions, accepting or rejecting new solutions according to the energy function and temperature parameters, and gradually reducing the temperature to improve the accuracy. The multi-scale tightening force correction amount is obtained, including large-scale overall adjustment and small-scale fine correction, covering the tightening force control requirements at different time and space scales. Dynamically smoothing the multi-scale tightening force correction amount through a non-linear response compensation function to generate an anti-vibration progressive adjustment curve is the fourth step. The non-linear response compensation function is a mathematical function used to adjust the response characteristics of the system, which can process input signals of different magnitudes and frequencies to different extents. In the tightening force monitoring of the banding machine, the non-linear response compensation function is mainly used to smooth the sudden change components in the multi-scale correction amount, reducing the oscillation and impact in the control process. Commonly used non-linear response compensation functions include S-type functions, exponential smoothing functions, and high-order polynomial functions, etc. By dynamically smoothing the multi-scale tightening force correction amount through these functions, the change of the control amount becomes more gentle and continuous, forming an anti-vibration progressive adjustment curve, avoiding sudden changes and oscillations in the control process. Design a feedback gain adaptive adjustment mechanism according to the progressive adjustment curve, and output a parameter correction control amount with time-varying characteristics. The feedback gain adaptive adjustment mechanism is a mechanism that can dynamically adjust control parameters according to the system state and external environment, making the control system have better adaptability and robustness. In the tightening force monitoring of the banding machine, the feedback gain mainly refers to the proportional, integral, and differential coefficients in the PID controller, and these coefficients directly affect the response characteristics of the control system. Analyze the characteristics of the progressive adjustment curve during the design process, determine the required response speed and stability requirements at different stages, and then design a gain adjustment function so that the parameters of the PID controller can be automatically adjusted according to the current state. The adjusted controller outputs a parameter correction control amount with time-varying characteristics, which can provide the most suitable control signal at different stages of the tightening process.

[0099] The parameter correction control quantity and historical tightening force data are optimized and integrated after time-domain and frequency-domain joint analysis to obtain a steady-state control scheme for the tightening force. Time-domain and frequency-domain joint analysis is an analysis method that simultaneously considers the time characteristics and frequency characteristics of signals and can comprehensively evaluate the performance of control strategies. In the tightening force monitoring of the strapping machine, the parameter correction control quantity and historical tightening force data are compared in the time domain to analyze the time characteristics of the control response; then the data is transformed to the frequency domain through Fourier transform to analyze the frequency response characteristics; finally, the control strategy is optimized and adjusted by combining the analysis results in the time domain and frequency domain. In the optimization and integration process, various factors such as control accuracy, stability, response speed, and anti-interference ability are comprehensively considered to form a steady-state control scheme for the tightening force.

[0100] The method for monitoring the tightening force of the strapping machine in the embodiments of the present application has been described above. Next, the tightening force monitoring system of the strapping machine in the embodiments of the present application will be described. Please refer to Figure 2 , an embodiment of the tightening force monitoring system of the strapping machine in the embodiments of the present application includes:

[0101] A coupling module for collecting stress-deformation coupling data during the operation of the strapping machine through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix;

[0102] A reconstruction module for performing signal reconstruction and dynamic threshold segmentation through an adaptive filtering algorithm according to the multi-dimensional tightening force time series feature matrix to obtain a set of key feature points in the tightening process;

[0103] A construction module for constructing a dynamic response model of the tightening force through a cross-validated deep tensor network according to the set of key feature points in the tightening process to obtain an accurate prediction engine with material adaptability;

[0104] An identification module for performing multi-modal abnormal pattern recognition on the real-time collected tightening force waveform based on the accurate prediction engine to obtain a hierarchical tightening deviation feature spectrum;

[0105] A diagnosis module for performing intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system according to the hierarchical tightening deviation feature spectrum to obtain a tightening force deviation compensation strategy diagram;

[0106] A generation module for generating a closed-loop control instruction for the strapping machine through a parameter self-correction iterative algorithm according to the tightening force deviation compensation strategy diagram to obtain a steady-state control scheme for the tightening force.

[0107] Through the collaborative cooperation of the above-mentioned various components, the stress-deformation coupling data during the operation of the strapping machine is collected by the high-frequency sampling sensing array, and a multi-dimensional bundling force time series feature matrix is obtained, which not only realizes the comprehensive monitoring of the physical state of the bundling process, but also provides a high-quality raw data basis for subsequent analysis, overcoming the limitations of the traditional single-sensor monitoring method; The adaptive filtering algorithm is used for signal reconstruction and dynamic threshold segmentation to obtain the key feature point set of the bundling process, effectively filtering out environmental interference and mechanical vibration noise, greatly improving the signal quality and the accuracy of feature extraction, and making the identification of key working condition points more accurate and reliable; The bundling force dynamic response model is constructed by the deep tensor network with cross-validation to obtain an accurate prediction engine with material self-adaptive ability. This deep learning model can automatically learn the mechanical properties and response laws of different materials, enabling the strapping machine to intelligently adapt to the bundling requirements of various materials, significantly improving the versatility and adaptability of the equipment; Based on the accurate prediction engine, multi-modal anomaly pattern recognition is performed on the real-time collected bundling force waveform to obtain a hierarchical bundling deviation feature spectrum, realizing the early detection and accurate classification of abnormal situations, and providing a reliable basis for fault diagnosis; Through the hybrid decision tree-fuzzy inference system for intelligent fault diagnosis, a bundling force deviation compensation strategy map is obtained. This hybrid intelligent algorithm fully combines the clear logical reasoning ability of the decision tree and the uncertainty processing ability of the fuzzy inference system, can accurately identify the cause of the fault and generate targeted compensation strategies, greatly improving the accuracy and efficiency of fault handling; Finally, the closed-loop control instruction of the strapping machine is generated by the parameter self-correction iterative algorithm to obtain a bundling force steady-state control scheme, realizing the automatic optimization and real-time adjustment of the control parameters, and ensuring the stability and consistency of the bundling force under various working conditions. It is particularly worth emphasizing that the deep tensor network algorithm applied in this solution innovatively integrates tensor decomposition and deep learning technologies, which can not only efficiently process multi-dimensional data structures and capture the complex coupling relationships between different physical quantities, but also significantly enhance the generalization ability of the model through the K-fold cross-validation technology, enabling the prediction engine to accurately handle new materials and working conditions that have not been seen before; The introduction of the genetic-simulated annealing hybrid algorithm cleverly combines the complementary advantages of the two optimization algorithms, achieving a good balance between global search and local fine-tuning, providing an efficient and stable calculation method for the optimization of control parameters, and improving the intelligent level and working efficiency of the bundling force monitoring and control of the strapping machine, realizing the accurate bundling force control of different materials.

[0108] Referring to Figure 3 , in the embodiment of the present invention, a computer device is further provided. This computer device can be a server, and its internal structure can be as Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected via a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. The computer program, when executed by the processor, implements the above method.

[0109] Those skilled in the art can understand that Figure 3 the structure shown in is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0110] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0111] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to the memory, storage, database, or other media provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0112] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0113] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0114] As described above, the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A method for monitoring the tightening force of a banding machine, characterized in that, The method for monitoring the tightening force of the strapping machine includes: Collecting stress-deformation coupling data during the operation of the strapping machine through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix; According to the multi-dimensional tightening force time series feature matrix, performing signal reconstruction and dynamic threshold segmentation through an adaptive filtering algorithm to obtain a set of key feature points in the tightening process; Based on the set of key feature points in the tightening process, converting the set of key feature points from a time series form to a third-order tensor structure, where the first dimension represents the sample index, the second dimension represents the time window, and the third dimension represents the feature type. Constructing a tightening force dynamic response model through a cross-validated deep tensor network to obtain an accurate prediction engine with material adaptability, including: converting the set of key feature points in the tightening process into a tensor representation form, constructing a tightening force data training matrix; performing data augmentation and normalization processing on the tightening force data training matrix to generate a model training sample library; designing a deep tensor network structure according to the model training sample library to form a tightening force neural hierarchical architecture; training and validating the tightening force neural hierarchical architecture through the K-fold cross-validation method to obtain a parameter-optimized tightening force model; constructing a material property compensation function based on the parameter-optimized tightening force model to generate a multi-type material adaptation coefficient table; integrating the multi-type material adaptation coefficient table into the parameter-optimized tightening force model to obtain an accurate prediction engine with material adaptability. Among them, during the integration process, a material identification mechanism and an adaptive adjustment mechanism are established. The material identification mechanism identifies the type of the current strapped material by analyzing the force-deformation curve characteristics at the initial stage of strapping, and the adaptive adjustment mechanism searches for corresponding compensation parameters from the material adaptation coefficient table according to the identification result and applies them to the prediction model; Based on the accurate prediction engine, performing multi-modal abnormal pattern recognition on the real-time collected tightening force waveform to obtain a hierarchical tightening deviation feature spectrum; According to the hierarchical tightening deviation feature spectrum, performing intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system to obtain a tightening force deviation compensation strategy map; Based on the tightening force deviation compensation strategy map, generating a closed-loop control instruction for the strapping machine through a parameter self-correction iterative algorithm to obtain a steady-state control scheme for the tightening force.

2. The method for monitoring the tightening force of the strapping machine according to claim 1, wherein The step of collecting stress-deformation coupling data during the operation of the strapping machine through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix includes: Arranging pressure sensors, displacement sensors, and angle sensors on the strapping device of the strapping machine to form a high-frequency sampling sensor array, and collecting the contact stress and deformation data between the strap and the strapped object; Synchronously sampling the output signals of the high-frequency sampling sensor array at a frequency of 1000 Hz through a signal acquisition circuit to obtain the original stress-deformation coupling data; Adding unified timestamp information to the original stress-deformation coupling data to form a multi-source tightening force parameter stream with time series correlation; Performing analog-to-digital conversion processing on the multi-source tightening force parameter stream to generate a digital stress-deformation sampling dot matrix; Perform tensor arrangement and matrix reconstruction on the digitized stress-deformation sampling lattice points to construct a tightening state cube containing three-dimensional information of pressure, displacement, and angle; Reduce the dimension and enhance the data of the tightening state cube through feature extraction methods to obtain a multi-dimensional tightening force time series feature matrix.

3. The method for monitoring the tightening force of the bundling machine according to claim 1, characterized in that, According to the multi-dimensional tightening force time series feature matrix, perform signal reconstruction and dynamic threshold segmentation through an adaptive filtering algorithm to obtain a set of key feature points in the tightening process, including: Perform multi-scale decomposition on the multi-dimensional tightening force time series feature matrix through wavelet transform to obtain the frequency-domain component spectrum of the tightening force signal; Calculate the noise energy distribution map based on the frequency-domain component spectrum and determine the signal-to-noise ratio parameters for each frequency band; Construct an adaptive band-pass filter bank based on the signal-to-noise ratio parameters to generate multi-band filtering templates; Apply the multi-band filtering templates to the multi-dimensional tightening force time series feature matrix to form a denoised smoothing curve of the tightening force; Perform sliding window analysis on the smoothing curve of the tightening force, calculate the change rate of the curve slope, and mark the mutation points of the tightening force; According to the mutation points of the tightening force, extract the key working condition points in the tightening process through a dynamic threshold segmentation algorithm to obtain a set of key feature points in the tightening process.

4. The method for monitoring the tightening force of the banding machine according to claim 1, characterized in that, Based on the precise prediction engine, perform multi-modal abnormal pattern recognition on the real-time collected tightening force waveform to obtain a hierarchical tightening deviation feature spectrum, including: Calculate the difference between the standard tightening force prediction value generated by the precise prediction engine and the real-time collected tightening force waveform to obtain the original tightening force deviation sequence; Extract waveform features from the original tightening force deviation sequence through time-frequency analysis methods to form a tightening force deviation feature vector; Construct a multi-dimensional abnormal index system based on the tightening force deviation feature vector to generate a tightening deviation evaluation matrix; Classify abnormal patterns through cluster analysis on the tightening deviation evaluation matrix to obtain a tightening abnormal type library; Match the similarity between the tightening abnormal type library and the current tightening force deviation to identify the current abnormal pattern category; Perform severity grading and spatial distribution analysis on the tightening force deviation according to the abnormal pattern category to obtain a hierarchical tightening deviation feature spectrum.

5. The method for monitoring the tightening force of a girdling machine according to claim 1, characterized in that, According to the hierarchical tightening deviation feature spectrum, perform intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system to obtain a tightening force deviation compensation strategy map, including: Input the hierarchical tightening deviation feature spectrum into a fault feature parser to extract a set of tightening fault symptom features; Construct a decision tree structure based on the set of tightening fault symptom features to form a preliminary tightening fault diagnosis path map; Quantify the uncertain nodes in the preliminary tightening fault diagnosis path map through a fuzzy membership function to obtain a tightening fault probability distribution table; Perform Bayesian probability inference based on the tightening fault probability distribution table to calculate the occurrence probability ranking of each fault type; Identify the main fault sources from the occurrence probability ranking of the fault types to generate a tightening force fault diagnosis result; Perform compensation parameter calculation and strategy mapping on the tightening force fault diagnosis result to obtain a tightening force deviation compensation strategy map.

6. The method for monitoring the tightening force of the bundling machine according to claim 1, characterized in that, According to the tightening force deviation compensation strategy diagram, a closed-loop control instruction for the strapping machine is generated through a parameter self-correction iterative algorithm, and a steady-state control scheme for the tightening force is obtained, including: Extract the torque transfer curve and response delay parameters from the tightening force deviation compensation strategy diagram, and construct a tension-relaxation dual-state control mapping table; Conduct a physical constraint boundary test on the tension-relaxation dual-state control mapping table to form a parameter adjustment interval; Through a genetic-simulated annealing hybrid algorithm, globally search and locally fine-tune the control variables within the parameter adjustment interval to obtain a multi-scale tightening force correction amount; Dynamically smooth the multi-scale tightening force correction amount through a non-linear response compensation function to generate an anti-vibration progressive adjustment curve; Design a feedback gain adaptive adjustment mechanism according to the progressive adjustment curve, and output a parameter correction control quantity with time-varying characteristics; After jointly analyzing the parameter correction control quantity and historical tightening force data in the time domain and frequency domain and optimizing the integration, a steady-state control scheme for the tightening force is obtained.

7. A tightening force monitoring system for a strapping machine, which is used to implement the tightening force monitoring method of the strapping machine described in any one of claims 1-6, characterized in that, The tightening force monitoring system of the strapping machine includes: A coupling module, which is used to collect stress-deformation coupling data during the operation of the strapping machine through a high-frequency sampling sensor array to obtain a multi-dimensional tightening force time series feature matrix; A reconstruction module, which is used to perform signal reconstruction and dynamic threshold segmentation on the multi-dimensional tightening force time series feature matrix through an adaptive filtering algorithm to obtain a set of key feature points in the tightening process; A construction module, which is used to convert the set of key feature points from a time series form to a third-order tensor structure according to the set of key feature points in the tightening process. The first dimension represents the sample index, the second dimension represents the time window, and the third dimension represents the feature type. A dynamic response model of the tightening force is constructed through a cross-validated deep tensor network to obtain an accurate prediction engine with material self-adaptive ability; including: converting the set of key feature points in the tightening process into a tensor representation form, and constructing a tightening force data training matrix; performing data augmentation and standardization processing on the tightening force data training matrix to generate a model training sample library; designing a deep tensor network structure according to the model training sample library to form a tightening force neural hierarchical architecture; training and validating the tightening force neural hierarchical architecture through the K-fold cross-validation method to obtain a parameter-optimized tightening force model; constructing a material property compensation function according to the parameter-optimized tightening force model to generate a multi-type material adaptation coefficient table; integrating the multi-type material adaptation coefficient table into the parameter-optimized tightening force model to obtain an accurate prediction engine with material self-adaptive ability. During the integration process, a material recognition mechanism and an adaptive adjustment mechanism are established. The material recognition mechanism identifies the type of the current strapped material by analyzing the force-deformation curve characteristics at the initial stage of strapping, and the adaptive adjustment mechanism searches for corresponding compensation parameters from the material adaptation coefficient table according to the recognition result and applies them to the prediction model; An identification module, which is used to perform multi-modal abnormal pattern recognition on the real-time collected tightening force waveform based on the accurate prediction engine to obtain a hierarchical tightening deviation feature spectrum; A diagnostic module, configured to perform intelligent fault diagnosis through a hybrid decision tree-fuzzy inference system according to the hierarchical bundling deviation feature spectrum, and obtain a bundling force deviation compensation strategy diagram; A generation module, configured to generate a closed-loop control instruction for the strapping machine through a parameter self-tuning iterative algorithm according to the bundling force deviation compensation strategy diagram, and obtain a steady-state control scheme for the bundling force.

8. A computer device, characterized in that, It includes a memory and a processor. The memory stores a computer program that can run on the processor. It is characterized in that when the processor executes the computer program, the bundling force monitoring method of the strapping machine described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, the processor is caused to execute the bundling force monitoring method of the strapping machine described in any one of claims 1 to 6.

Images