Intelligent monitoring and fault early warning system for hollow glass production

Abstract

The application discloses a kind of hollow glass production intelligent monitoring and fault early warning system, it is related to glass production monitoring technical field, the technical means of this application is adopted digital twin simulation, multi-physical dimension feedback diagnosis and process knowledge graph causal closed loop update, can realize the advanced intervention of sealing glue quality hidden danger and self-learning optimization of process parameter, break the hysteresis of traditional process dependence after detection, so that production system can implement targeted pressure and heat pre-compensation to potential failure point, improve the reliability of hollow glass initial sealing from mechanism level, multidimensional real-time feedback not only improves the robustness evaluation precision of production process, also provides non-contact fault early induction factor positioning ability for processing site, reduces the shutdown risk caused by equipment sudden failure, based on the adaptive learning ability of measured data chain, ensure that process decision path is constantly optimized with the growth of service time.

Description

Technical Field

[0001] This invention relates to the field of glass production monitoring technology, specifically to an intelligent monitoring and fault early warning system for insulated glass production. Background Technology

[0002] Insulating glass, a key structural component of modern energy-efficient buildings, relies heavily on the quality of the sealant layer at the cavity edges and the bonding durability of the interface. High-quality sealant not only effectively prevents external moisture from entering the cavity but also prevents the leakage of internal inert gases, thus ensuring the long-term effectiveness of its thermal and sound insulation properties. In the current deep-processing industry, automated sealing equipment is widely used in production lines, and its execution precision and status monitoring level directly determine the quality standards of the final product.

[0003] Existing technologies, such as the invention patent applications related to glass production monitoring with announcement numbers CN112414634B, CN108122226B, and CN121049386B, demonstrate that current monitoring methods for insulating glass production primarily rely on proportional-integral-derivative (PID) controllers to linearly adjust the pumping pressure of the extrusion system, coupled with photoelectric sensors to monitor the movement trajectory of the glue gun. During processing, the monitoring system, using preset process parameter templates, drives a robotic arm to perform glue injection along the glass frame, and a vision sensor performs geometric contour checks on the width and surface flatness of the finished glue strip. Some advanced systems attempt to collect pump pressure fluctuation data from the supply system to simply determine whether there are large-scale flow interruptions in the glue path.

[0004] The aforementioned techniques exhibit significant limitations in real-world, complex production environments. Current solutions generally lack physical simulations of the non-Newtonian fluid dynamics of colloids, making it impossible to anticipate molding defects caused by stress concentration within narrow cavities. Existing monitoring dimensions are limited to single macroscopic physical quantities, making it difficult to identify hidden quality risks caused by subtle attenuations in the equipment's transmission chain or batch-to-batch variations in the physical properties of the colloid. Furthermore, the lack of an efficient causal chain alignment and feedback iteration mechanism between the sealing process execution parameters and the subsequent finished product airtightness testing data prevents the system from effectively implementing closed-loop self-evolution to address long-term process deviations, resulting in production stability being highly susceptible to equipment wear and environmental fluctuations. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the present invention aims to provide an intelligent monitoring and fault early warning system for insulating glass production.

[0006] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides an intelligent monitoring and fault early warning system for insulating glass production, including: a multi-source working condition sensing module, used to acquire the specification and size parameters of the glass to be processed, the physical property parameters of the colloidal material and the environmental parameters of the production site, and generate multi-source working condition parameters.

[0007] The digital twin simulation module is used to simulate and calculate the evolution of multi-source operating parameters using a physical field coupling algorithm, generating a digital colloid representing the colloid forming state and a predicted map of sealing weak points indicating the spatial location of defects.

[0008] The process target planning module is used to extract failure assessment factors from the sealing weak point prediction map and perform calculations in combination with the ideal sealing index stored in the system to generate a dynamic target curve.

[0009] The composite parameter modulation module is used to analyze the dynamic target curve to obtain the drive response command, and generate a composite parameter cooperative modulation strategy by performing pressure and speed control and local heat compensation actions on the sealing equipment.

[0010] The collaborative modulation strategy execution monitoring module collects multi-physical dimension feedback signals from the sealing equipment during the execution of the composite parameter collaborative modulation strategy, and performs time-frequency domain transformation processing on the multi-physical dimension feedback signals to generate process oscillation spectrum characteristics.

[0011] The signal feature diagnosis module is used to compare the process oscillation spectrum characteristics with the simulation expected characteristics derived from the digital colloid, and output process health diagnosis results for evaluating production robustness.

[0012] The closed-loop update module of the process knowledge graph is used to aggregate multi-source operating parameters, process health diagnosis results, and final product sealing test results fed back by back-end testing equipment to generate and update a process knowledge graph to guide subsequent production.

[0013] The beneficial effects of this invention are as follows: This invention constructs a digital colloidal model through a physical field coupling algorithm, enabling the prediction and simulation of the colloidal forming state within a complex and confined space. This allows for the identification of weak points in the seal caused by geometric deformation or fluctuations in material flow resistance before sealing is performed. This prediction mechanism based on digital evolution overcomes the lag of traditional processes that rely on post-process detection, enabling the production system to implement targeted pressure and heat pre-compensation for potential failure points, thereby improving the reliability of the initial seal of insulating glass from a mechanistic perspective.

[0014] This invention integrates stator current waveforms with robotic arm vibration characteristics and implements short-time fast Fourier transform to establish a time-frequency domain-based process health diagnostic system. By comparing the measured physical response centroid with the simulated baseline, this scheme can capture metastable oscillations caused by microbubbles in the flow channel or decreased spindle transmission accuracy. This multi-dimensional real-time feedback not only improves the accuracy of robustness evaluation of the production process but also provides non-contact fault early-stage inducing factors location capabilities for the processing site, reducing the downtime risk caused by sudden equipment failures.

[0015] This invention achieves a deep self-evolving closed loop in production logic by constructing a process knowledge graph that includes multi-source operating parameters and final sealing performance test results. The system can reverse-drive the correction of associated node weights based on the actual performance indicators obtained from helium mass spectrometry leak detection, dynamically eliminating the effects of mechanical wear deviations and seasonal environmental drift caused by long-term equipment service. This adaptive learning capability based on measured data chains ensures that the process decision path is continuously optimized as service time increases, guaranteeing a high degree of consistency in finished product sealing quality under large-scale continuous production conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the system module connections of the present invention;

[0018] Figure 2 This is a flowchart of the method of the present invention;

[0019] Figure 3 Flowchart for generating a prediction map of weak points in the seal;

[0020] Figure 4 This is a flowchart illustrating the generation process of the dynamic target curve. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Reference Figure 1 and Figure 2As shown, this invention provides an intelligent monitoring and fault early warning system for insulating glass production, comprising:

[0023] The multi-source operating condition sensing module is used to acquire the specifications and dimensions of the glass to be processed, the physical properties of the colloidal material, and the environmental parameters of the production site, and generate multi-source operating condition parameters.

[0024] In a specific embodiment of the present invention, the generation of multi-source operating parameters includes: acquiring the cumulative discharge stroke data of the sealing equipment and performing loss quantification processing to generate a wear coefficient that reflects the performance of the pump body.

[0025] Specifically, the physical attenuation of the sealing equipment's actuator is quantitatively assessed to calibrate the mechanical characteristics of the conveying system. The servo controller's underlying layer cyclically reads the cumulative discharge stroke data at a sampling period. After the actuator unit resets, the actual displacement value of the spindle drive mechanism—the measured value of the cumulative discharge stroke data—is extracted. The algorithm encapsulated within the servo controller quantifies this data to generate a wear coefficient reflecting pump performance. By introducing a dimensionless exponential degradation model, the physical blind spot in measuring the internal frictional deformation of the glue pump is avoided. The structure of the wear coefficient calculation equation is determined as follows: In this mathematical expression structure, each operand has the meaning of an engineering index. This represents the dimensionless attenuation multiplication factor, i.e., the wear coefficient, of the output. This represents the measured value of the cumulative discharge stroke data with length characteristics currently acquired by the sensor. The allowable overhaul cycle calibrated stroke constant scalar representing the fixed production configuration is substituted into the calculation using units of length with the same dimensions. This represents the conventional value of the attenuation fitting coefficient extracted from long-term statistical analysis, preset between 0.02 and 0.05. This calculation strategy strictly relies on the rule of dividing length values ​​to achieve complete dedimensionalization of state parameters at different equipment stages.

[0026] The temperature and humidity time series of the production site are acquired and environmental normalization is performed to generate environmental feature vectors.

[0027] Specifically, the system calibrates the microclimate drift patterns of the atmosphere in the production workshop to ensure that the boundary conditions for the front-end viscosity calculation have dynamic environmental tracking capabilities. The communication gateway, based on a fixed-length data truncation protocol, acquires the time series of temperature and relative humidity in the surrounding space. Due to the incompatible nature of temperature measurement and water vapor percentage in the physical realm, the system must and can only extract signal packets for environmental normalization. The microscopic data transformation algorithm is constructed based on linear constraint logic and includes two parallel computation branches, with corresponding equations as follows: as well as Based on the instruction autoequation, the corresponding process metadata is parsed, where This represents the normalized nodal component of relative temperature after filtering and compression. The relative humidity node components obtained from scaling are collectively used as the subsequent processing array. and These are respectively the discrete floating-point probe information streams received from the industrial front end. The constraint calibration is typically set to the baseline extreme value below the 15-degree Celsius lower limit. Define the upper limit of the thermal limit warning value at 35 degrees Celsius. Define the lower baseline relative gaseous water limit at 30%. The boundary was locked at 80% of the heavy dew warning standard. The normalization step first stripped the intercept by subtraction, and then completed the amplitude ratio conversion by division, successfully eliminating the messy measurement background and generating safe spatial distribution points with values ​​between 0 and 1. Finally, it was assembled into an environmental feature vector according to the time stack.

[0028] It should also be noted that the sealing process is sensitive to temperature and humidity. Low temperatures can cause the adhesive to become viscous, while high temperatures and humidity can affect the curing of the adhesive. 30%-80% humidity and 15-35℃ are the industry's standard safe range for sealing environments.

[0029] By integrating wear coefficient, environmental feature vector, and dimensional parameters, multi-source operating condition parameters are generated.

[0030] This parameter set is directly fed back to a higher-level fluid dynamics simulation center, enabling the simulation network to receive a high-quality dynamic source that simultaneously includes the actual working wear and tear of the equipment, subtle changes in regional thermal climate, and is compatible with various non-standard profile dimensions of glass to be sealed, thereby improving the underlying fidelity of twin scenario pre-simulation calculations.

[0031] The digital twin simulation module is used to simulate and calculate the evolution of multi-source operating parameters using a physical field coupling algorithm, generating a digital colloid representing the colloid forming state and a predicted map of sealing weak points indicating the spatial location of defects.

[0032] Reference Figure 3 As shown, in a specific embodiment of the present invention, the generation of a digital colloid characterizing the colloid molding state and a predicted map of sealing weak points indicating the spatial location of defects includes: extracting spatial geometric constraints, flow constitutive parameters obtained from the analysis of colloid material property parameters, wear coefficients, and environmental feature vectors from the multi-source working condition parameters, and generating an initial vector for the simulation boundary.

[0033] Specifically, the physical groove depth geometric data (i.e., spatial geometric constraints) of the glass plate along the sealing path within the multi-source operating parameters are read, and the rheological properties data of the colloidal material at the current ambient temperature (i.e., flow constitutive parameters) are analyzed to extract the consistency coefficient characterizing the viscosity variation of the non-Newtonian fluid. Together with the shear behavior index, geometric mesh data and original material rheological constitutive parameters are formed.

[0034] The wear coefficient and environmental feature vector from multiple operating parameters are integrated synchronously. The wear coefficient is introduced as a system drift correction factor into the initial residual setting of the iterative calculation to compensate for the influence of pump and valve body deformation on the fluid pressure boundary. At the same time, the normalized components of temperature and humidity in the environmental feature vector are used as environmental adjustment coefficients to correct the consistency coefficient in real time. The value of is determined to eliminate the interference of temperature and humidity changes on the rheological properties of the fluid.

[0035] It should be noted that the original consistency coefficient is adjusted using a linear correction relationship. Real-time dynamic correction is performed, and the correction formula is as follows: ,in This is the environmental regulation coefficient calibrated experimentally. For the temperature and humidity normalized components in the environmental feature vector, The actual consistency coefficient after taking into account the dynamic effects of temperature and humidity.

[0036] The geometric mesh data, the environmentally corrected material rheological constitutive parameters, the equipment attenuation correction determined by the wear coefficient, and the environmental adjustment coefficient are arrayed, encapsulated, and normalized to generate the final simulation boundary initial vector.

[0037] For example, the simulation boundary initial vector It consists of four parts: .in, For geometric mesh matrix, This is the material rheological constitutive matrix after environmental correction. For the environmental feature matrix, The initial value of the pressure boundary after incorporating the equipment attenuation correction is obtained after compensation by the system drift correction factor.

[0038] The initial vector of the simulation boundary is input into the preset non-Newtonian fluid dynamics equation for iterative solution to generate a digital colloid.

[0039] Specifically, the computational fluid dynamics solver is invoked to load the initial vector of the simulation boundary into a preset set of non-Newtonian fluid dynamics governing equations. Based on the typical shear-thinning characteristics of sealant, a power-law constitutive model is used to drive the iterative calculation of velocity and pressure in the flow field. The formula for calculating the apparent viscosity used to update the rheological state of the fluid elements in real time is set as follows: In the formula This represents the effective apparent viscosity physical quantity of the current mesh cell within the iteration step. This represents a scalar quantity representing the consistency coefficient derived from material property parameter inputs. This represents the shear rate modulus derived from the differential operation of the velocity gradient tensor of the flow field. The shear behavior index, representing the dimensionless flow behavior of non-Newtonian fluids, is typically provided by the material supplier and is usually between 0.3 and 0.7. The fluid dynamics solver continuously performs residual convergence calculations for momentum and mass conservation until stable full flow field data is obtained, thereby generating a digital colloid containing velocity vectors, pressure scalars, and viscosity distributions at all discrete points within the flow channel.

[0040] It should be noted that the types of computational fluid dynamics solvers, such as Fluent and COMSOL, are commercial software choices that can be adapted by those skilled in the art. The specific threshold for residual convergence is a routine parameter for simulation debugging and can be set according to actual process requirements. The mesh type for flow field discretization is a simulation detail optimization and does not affect the implementation of simulation logic. The specific compensation for iterative calculations can be automatically adapted by the solver according to the flow field state, and will not be detailed here.

[0041] The stress distribution of the digital colloid is scanned and abnormal nodes exceeding the allowable threshold stored in the system are identified, generating a prediction map of weak points in the seal.

[0042] Specifically, the stress distribution scanning program is initiated to traverse the shear stress values ​​of all grid nodes inside the digital colloid. Through logical comparison, abnormal nodes whose calculated shear stress values ​​exceed the material allowable threshold stored in the system are identified. The material allowable threshold is, for example, 0.5 MPa. The spatial three-dimensional coordinates, i.e., risk coordinates, of high stress gradients or flow dead zones are locked and projected onto the virtual geometric model of the glass plate.

[0043] It should be noted that excessive shear stress in the colloid can easily lead to sealing defects such as bubbles, adhesive breakage, and interlayer delamination. Therefore, for digital colloids that characterize the colloid forming state, shear stress is selected as the judgment index to match the actual risks of the insulating glass sealing process.

[0044] It should be noted that the flow dead zone refers to the area where the fluid hardly flows in the flow channel. It is a typical defect area in the sealing process that is prone to generating air bubbles and uneven filling.

[0045] The final output consists of a virtual fluid object, namely a digital colloid, composed of high-fidelity physical field data, and a three-dimensional topological mapping set, namely a sealing weak point prediction map, which intuitively marks the coordinates of potential air bubble voids, adhesive breakage risks, or bonding failure points. The sealing weak point prediction map provides an accurate spatial navigation reference for subsequent targeted process compensation.

[0046] This invention constructs a digital colloidal model using a physical field coupling algorithm, enabling the prediction and simulation of the colloidal forming state within complex and confined spaces. This allows for the identification of weak points in the seal caused by geometric deformation or fluctuations in material flow resistance before sealing is performed. This prediction mechanism based on digital evolution overcomes the lag of traditional processes that rely on post-process detection, enabling the production system to implement targeted pressure and heat pre-compensation for potential failure points, thereby improving the reliability of the initial seal of insulating glass from a mechanistic perspective.

[0047] The process target planning module is used to extract failure assessment factors from the sealing weak point prediction map and perform calculations in combination with the ideal sealing index stored in the system to generate a dynamic target curve.

[0048] Reference Figure 4 As shown, in a specific embodiment of the present invention, the generation of the dynamic target curve includes: identifying the set of risk coordinates in the prediction map of the sealing weak point and obtaining the material flow channel resistance characteristics corresponding to the risk coordinates, and generating a local compensation demand matrix.

[0049] Specifically, the process begins by analyzing the embedded mesh data structure within the seal weakness prediction map, extracting the spatial three-dimensional coordinates of all abnormal state nodes, and forming a risk coordinate set. The end effectors of the equipment along the adhesive application path typically face pressure losses caused by abrupt changes in cross-section or corner turning. Based on this, the controller retrieves coordinate sectional information, calculates the material flow resistance characteristics caused by wall friction resistance, and converts these characteristics into local pressure drop gradient values. The computation unit then concatenates the coordinate sequence array with the corresponding resistance characteristic scalar in memory to generate a two-dimensional data mapping dimension local compensation demand matrix.

[0050] It should be noted that the coordinate section information of abnormal state nodes can be retrieved according to the mesh design of the actual simulation model. The specific calculation method of material flow channel resistance characteristics and the calculation of wall friction resistance are common algorithms in the fluid mechanics industry. Those skilled in the art can adapt them independently according to the pressure drop gradient and flow channel geometry characteristics, which will not be described in detail here.

[0051] The ideal physical quantities required to maintain interfacial adhesion are calculated based on the local compensation demand matrix, and the desired apparent viscosity value and desired interfacial wetting energy value are generated.

[0052] Specifically, for each independent element node traversed within the local compensation demand matrix, the ideal physical quantities required to maintain interfacial adhesion are calculated. The first stage constructs an inversion calculation link based on the physical properties of the material's viscosity characteristics, and uses dynamic formulas... Customized directional solution for flow regime indices, where This represents the desired apparent viscosity value to be output, with its basic physical unit set to Pascal. Second, The material flow channel resistance characteristic represented by the buffer extracted in the preceding steps has been characterized as a scalar equivalent of the effective shear force in the mainstream upward direction, in Pascals. The constant calibration shear rate variable, derived by referring to the geometric cross-sectional constant of the discharge chamber of the feed pump, has a basic dimension of per second. This algebraic expression directly uses division to eliminate dimensional inconsistencies caused by irrelevant empirical constants, thus ensuring an absolutely consistent conversion of dynamic characteristics.

[0053] After obtaining the apparent flow properties, the second calculation stage begins. This stage estimates the interfacial bonding requirements based on the degree of molecular layer spreading and bonding at the gas-liquid-solid three-phase interface at the thermodynamic level. The computational layer utilizes the dimensional conversion equation of the material surface spreading energy. Complete the quantitative analysis of energy density values. The output represents the desired interfacial wetting energy value, used to describe the energy required for wetting and adsorption of the surface and interface per unit contact grid. Its equivalent unit is joules per square meter. This represents the reference surface tension constant of the sealing silicone, retrieved and distributed from the central control database within the higher-level digital system to the current task batch. This physical property value is typically maintained within the range of 0.03-0.06 N / m. This represents the pure proportional dimensionless wetting increase correction coefficient specified by the process platform for the micro-roughness compensation requirements of the original edge interface of the glass, which is usually set in the range of 1.2-1.5.

[0054] It should be noted that the ideal physical quantity is decomposed into the expected apparent viscosity value in the fluid dynamics dimension and the expected interfacial wetting energy value in the surface thermodynamics dimension. The former solves the problem of colloid flow and filling, while the latter solves the problem of adhesion between colloid and glass interface. Together, they constitute the physical quantity that maintains the sealing performance of the sealant interface. Based on the inversion logic of local compensation demand matrix - flow channel resistance characteristics - expected apparent viscosity, it corresponds to the technical pain point of weak seal caused by abnormal flow channel resistance mentioned earlier. The resistance characteristics are transformed into an adjustable viscosity index through kinetic formulas, realizing a logical connection and seamless connection between problem diagnosis and solution quantification.

[0055] It should also be noted that, and The range of values ​​all conforms to the material properties of the sealing silicone and the process compensation requirements of the glass edge interface.

[0056] The desired apparent viscosity and desired interfacial wetting energy values ​​are arranged along the production time axis to generate a dynamic target curve.

[0057] After the microprocessor calculates the compensation energy index of the high-frequency scattered spatial points on the path of the edge to be processed, it sends a command to the bus to call the trajectory speed feedforward parameters solidified in the underlying path control module of the insulating glass sealing robot. The linear velocity reference of the servo execution is usually locked between 200-600 mm / s. The controller divides the absolute displacement scalar of the coordinate axis mileage of the defect dispersion by the above-extracted stable running linear velocity, thereby completing the mapping action of the spatial displacement length vector to the clock trigger sequence with time span as the scale variable. The system's time series serialization and reconstruction algorithm tool then begins to read the array of expected apparent viscosity values ​​and expected interface wetting energy values ​​at scattered discrete time points, and according to the reset and calibrated power-on clock tick sequence, inserts and aligns them one by one into the production time axis queue channel, and starts the Bezier spline interpolation filtering algorithm to smooth the data step differential stitching discontinuity region of the clock gap.

[0058] The final output data is presented as a continuous instruction set, i.e., a dynamic target curve, that carries the process feedforward execution basis across multiple physical domain dimensions and conforms to the mechanical clock rhythm. This control variable trend graph not only closely aligns with the actual spindle movement at a microsecond-level low-level refresh rate, but also prepares a high-fidelity control blueprint for the next stage of pressure dynamic intervention and synchronous regional local temperature enhancement adjustment to be implemented by the downstream servo system, enabling direct communication and execution.

[0059] The composite parameter modulation module is used to analyze the dynamic target curve to obtain the drive response command, and generate a composite parameter cooperative modulation strategy by performing pressure and speed control and local heat compensation actions on the sealing equipment.

[0060] In a specific embodiment of the present invention, the generated composite parameter collaborative modulation strategy includes: analyzing the dynamic target curve to extract the main pump pressure regulation amount and generating the main shaft control signal.

[0061] The controller's digital signal processing engine extracts numerical slices of preset step size from the dynamic target curve in real time and converts them into spindle control signals that control the operation of the extrusion pump unit. To counteract the interference of material internal viscosity fluctuations on flow rate stability, the system utilizes a built-in dynamic conversion algorithm. Calculate the output pumping frequency command, where This represents the speed and frequency control voltage or frequency setpoint output to the servo driver, with a commonly used engineering range of 0-50 Hz. The pressure conversion sensitivity coefficient represents the system calibration, with physical dimensions in volt-seconds / pascal-seconds. It is used to accurately map viscosity characteristics into electrical control quantities for the servo drive. It is the scalar value of the expected apparent viscosity generated by the aforementioned steps.

[0062] It should be noted that by using a kinetic conversion algorithm to map the desired apparent viscosity value, i.e., the physical quantity of the process, into the speed and frequency control quantity, i.e., the electrical control quantity, the core problem of unstable flow rate caused by fluctuations in material viscosity force is accurately solved. This achieves active compensation for viscosity changes by real-time pressure / frequency adjustment, which is highly compatible with the pain points of the process.

[0063] It should also be noted that the preset step size can be set according to the control accuracy and dispensing speed of the sealing equipment. Those skilled in the art can adapt it independently. The pressure conversion sensitivity coefficient is strongly related to the model of the extrusion pump and the type of adhesive. It can be obtained through experimental calibration or equipment manual and does not need to be uniformly limited.

[0064] Calculate the required power drive value for the micro-focused heating unit based on the desired interface wetting energy value, and generate auxiliary thermal compensation commands.

[0065] While regulating the main pump pressure, the system simultaneously analyzes the thermal parameters in the dynamic target curve and calculates the power drive value required for the micro-focused heating unit to enhance local wetting activity. To ensure that the thermal response gradient of the glass surface meets the requirements of colloidal wetting kinetics, the system uses a power compensation calculation formula. Output auxiliary thermal compensation command, in the formula, This indicates the output electrical power to the heating element, measured in watts, and typically adjustable from 50 to 300 watts. This represents the desired interface wetting energy value, expressed in joules per square meter. This represents the measured linear velocity of the glue-applying robotic arm along its path, expressed in meters per second. This indicates the effective width of the characteristic coating on the sealing head, in meters. This is a dimensionless correction coefficient used to characterize electrothermal conversion efficiency and spatial radiation loss, obtained through experiments or equipment manuals. This calculation, based on the product of energy area density and sweep rate, ensures the consistency of power consumption levels on both sides of the equation per unit time.

[0066] It should be noted that the effective width of the sealing head's characteristic coating is an inherent physical constant of the sealing equipment, which is matched with the equipment model and retrieved from the equipment parameters.

[0067] Simultaneously send spindle control signals and auxiliary thermal compensation commands to drive hardware actions and generate a composite parameter collaborative modulation strategy.

[0068] Subsequently, the communication management module uses a deterministic fieldbus to synchronously send spindle control signals and auxiliary thermal compensation commands to the lower-level variable frequency pump control module and high-frequency thermal compensation controller with a synchronization jitter error of less than two milliseconds. This directly drives the hardware actuator to complete the coordinated action of dynamic pressure adjustment and spatial focused heating.

[0069] The output process control entity is a set of command stream sequences with spatiotemporal synchronization characteristics, namely a composite parameter cooperative modulation strategy. This strategy deeply couples the mechanical pumping intensity with the physicochemical interface activity enhancement measures. By precisely aligning the millisecond-level fine-tuning of the pump output pressure with temperature rise intervention measures at specific weak sealing points on the time axis, a cooperative control logic capable of dynamically adapting to complex non-standard operating conditions is formed. This action solves the problem of uneven filling caused by the non-Newtonian properties of materials at the hardware execution level, providing a controlled benchmark excitation source for subsequent acquisition of production feedback signals and evaluation of operating condition robustness.

[0070] The collaborative modulation strategy execution monitoring module collects multi-physical dimension feedback signals from the sealing equipment during the execution of the composite parameter collaborative modulation strategy, and performs time-frequency domain transformation processing on the multi-physical dimension feedback signals to generate process oscillation spectrum characteristics.

[0071] In a specific embodiment of the present invention, the generation process oscillation spectrum characteristics include: acquiring the stator current waveform signal of the drive motor of the sealing equipment and extracting the vibration acceleration signal at the end of the robotic arm to generate the original physical feedback signal.

[0072] The monitoring center's bottom-level analog-to-digital conversion acquisition board, synchronized with the encoder of the glue-applying feed axis, acquires and senses external physical data. It reads the electrical signal from a non-contact electromagnetic current-measuring element wrapped in the power supply phase harness of the servo glue pump, reconstructing the stator current waveform. Within the aligned time stamp, it extracts the vibration acceleration signal from a piezoelectric vibration sensor mounted on the glue outlet at the end of the robotic arm. The transient fluctuations captured by the current signal reflect the microscopic drag force changes generated by the motor drive shaft resisting the thickening of the colloidal medium. Furthermore, its real-time recording frame rate is typically constrained within a dense sampling bandwidth of 3000-15000 Hz. The rapid increase in front-end acceleration data often corresponds to the bursting of air bubbles within the material through narrow gaps, resulting in an unsmooth spraying vibration effect. For the current and acceleration physical sequences extracted by the system at the same glue application time scale, which possess time-frequency non-smooth characteristics, the data synchronization combiner uses a dual-line array mechanism to stitch and encapsulate the data, generating unmodified electrical-mechanical dual-dimensional raw physical feedback signals.

[0073] A short-time fast Fourier transform is performed on the original physical feedback signal to generate a frequency domain distribution map of the signal power.

[0074] To overcome the computational challenge of determining the severity of local transient harmonic oscillations solely based on time series data, and to track the dynamic transitions within the excitation force's frequency band, the logic control module employs a window function truncation strategy to perform a short-time fast Fourier transform (SFT) operation on the extracted sequence. This analysis unit utilizes a glide time segment sequence to perform high-resolution analytical transformation on the time-domain signal segments, revealing a two-dimensional matrix structure array of frequency intensity characteristics and durations—essentially a signal power frequency domain distribution map. When stripping noise energy and calculating the absolute physical intensity of the discrete array's frequency points, the system invokes the transformation equation. Calculate the exact spectral amplitude of the frequency points corresponding to the discrete sequence. In this expression, the target result variable is... The output signal sequence is represented by the index frequency scale point of the current time window segment. The true energy scalar at that point, in the equation This represents the total count of clock sample data points truncated for a given discrete segment of data. It is typically fixed at an operator scale of 256 or 512 and is directly used as a scalar divisor to prevent nonlinear energy overflow caused by summation. The operation factor within the summation sign... , defined as the first label in the sensor time series The actual captured sampling calibration number corresponding to each location, while retaining the original metric basis generated by mechanical vibration measurement, or the physical measurement properties of fluid-driven load current, calculates the multiplier. To eliminate the frequency domain energy leakage and overflow caused by forced segmentation of signals, the dimensionless Hamming smoothing window coefficient is used in the tail operator, which utilizes the natural constant base. and the imaginary part of the complex field As a basis, the coordinate angular velocity projection rotation from the real space time domain axis to the frequency domain axis is realized through exponentiation, and the coefficients of the entire frequency band are... Time sequence number Total number of sampling points All of these are dimensionless physical quantities, which together form the pure phase angle offset rotation factor.

[0075] It should be noted that the specific window length and sliding step size of the window function (i.e., the window length and sliding step size of the Hamming window) can be set according to the acquisition frame rate of the original signal. Those skilled in the art can adjust and match the characteristics of the process signal without uniform limitation. Based on MATLAB / Python function calls and the low-level implementation of short-time fast Fourier transform in hardware FPGA, the value of N is 256 / 512, which is a conventional operator scale for digital signal processing, and is adapted to the acquisition and processing efficiency of industrial signals.

[0076] Energy amplitude characteristics of a specific frequency band are extracted from the signal power frequency domain distribution map to generate process oscillation spectrum characteristics.

[0077] After the algorithm process is completed, the digital terminal will skip the low-frequency background noise position dominated by the steady-state reciprocating operation of the servo motor. Based on the experience of the high-frequency response distribution of the weak vibration of the equipment caused by abnormal pulsation of colloidal flow, it will set a typical abnormal vibration monitoring band of 50-450 Hz and accurately define the boundary area within the above spectrum. Then, it will calculate the total integral average energy value of the intensity of all discrete spectral segments within the specific filter frequency band and generate the energy amplitude feature used to quantify the degree of rheological instability.

[0078] The output vector set of dynamic quality dimensions of the evaluation conditions serves as the quantitative benchmark for characterizing the smoothness of equipment processes and the uniformity of fluid shear, namely, the process oscillation spectrum characteristics. This set of process oscillation spectrum characteristics, obtained through precise high-frequency spatiotemporal transformation, effectively filters out redundant noise from the field while retaining the essential dynamic characteristics of glue line breakage and molding resistance fluctuations. It can be directly delivered to the lower-level model, serving as a source for judging and verifying the similarity difference scores of matching simulation reference templates.

[0079] The signal feature diagnosis module is used to compare the process oscillation spectrum characteristics with the simulation expected characteristics derived from the digital colloid, and output process health diagnosis results for evaluating production robustness.

[0080] In a specific embodiment of the present invention, the output is a process health diagnostic result used to evaluate production robustness, which includes: mapping the frequency energy centroid of the process oscillation spectrum characteristics and generating a measured feature vector.

[0081] The monitoring system maps the frequency energy centroid of the process oscillation spectrum characteristics using a frequency-weighted averaging algorithm, extracting a frequency scalar that represents the core location of the signal energy distribution, and using this scalar as a sensitive indicator of the equipment status. The formula for calculating this frequency energy centroid is as follows: In this mathematical description logic, the output term Defined as the extracted center location of the energy distribution with frequency dimension, i.e., the centroid of frequency energy, its engineering calculation results are typically distributed within the 80-500 Hz frequency response range, and the computational components... The first characteristic of the process oscillation spectrum The discrete frequency values ​​of each sampling frequency band, whose physical dimension is Hertz, and the corresponding multiplication weight term The signal energy amplitude data extracted at that specific frequency point is received. Its physical representation is a power intensity scalar. By summing the frequency power products and dividing the total power by the total power, the interference of absolute power fluctuations on frequency characteristics is eliminated, and the center of gravity of the frequency energy is determined. Encapsulate it as a measured feature vector representing the robustness of the current job.

[0082] Calculate the Euclidean distance between the measured feature vector and the simulated expected feature, and generate a state deviation score.

[0083] Next, the system comparison unit retrieves the simulation-expected features derived from the digital colloid from the storage medium. These features include the ideal oscillation amplitude-frequency response distribution curve obtained from the digital twin system simulation under the same operating conditions. The calculation module uses a geometric distance metric model to solve for the Euclidean distance between the measured feature vector and the simulation-expected features, thereby quantifying the degree of dispersion of the actual colloid molding effect from the ideal value of the physical field coupling simulation. The function performing the calculation is defined as... . in the formula This represents the final generated state deviation score, used to measure the severity of operational deviations. It is expressed as a non-negative scalar value as a relative deviation distance. Represents the measured feature vector in the th... Real-time physical response component data in multiple dimensions This corresponds to the expected digital simulation value in the same dimension in the simulation expected features. At the same time, both components have undergone feature scale unification processing before calculation, ensuring that the subtraction operation has the rationality of subtraction of the same dimension in physical logic.

[0084] In this insulated glass monitoring system, the measured feature vector is encapsulated by the frequency energy centroid, and is a one-dimensional feature vector. Therefore, the feature vector here... In practice, only 1 is taken, and the formula can be simplified to: , which is the absolute difference between the measured energy centroid and the simulated energy centroid.

[0085] It should be noted that, based on the digital colloid, virtual sensing technology is used to extract virtual pressure pulsations or virtual flow channel wall shear force fluctuation signals generated by the colloid flow by corresponding to the installation positions of the actual physical sensors in the digital model. The virtual signals are then processed by the same short-time Fourier transform as the measured signals to generate simulation expected features that can be compared with the oscillation spectrum characteristics of the measured process.

[0086] The state deviation score is compared with the production safety boundary defined within the system, and the process health diagnosis result is output.

[0087] After obtaining the state deviation score, the diagnostic logic compares the score with the production safety boundary defined within the system. This boundary is the allowable limit of dispersion calculated from historical high-quality product production data across multiple batches. Typically, its fluctuation threshold is set between 5% and 12% of the original characteristic mean. The production line is deemed fault-free as it does not exceed production safety boundaries, and the process health diagnosis result is marked as normal operation; if If the production safety boundary is exceeded, the production line is determined to be faulty, and the fault type is determined based on the specific frequency value measured: when the measured frequency is in the low-to-mid frequency range of 80-200Hz, the sealing equipment is determined to have a risk of intermittent glue extrusion; when the measured frequency is in the high-frequency range of 300-450Hz, the sealing mechanism is determined to have a risk of mechanical resonance exceeding the physical elastic limit.

[0088] The final output is a report containing quantitative indicators of the current production line's operational risk level and process deviation trends—a process health diagnosis result. This diagnosis not only accurately identifies whether the sealing mechanism has mechanical resonance risks exceeding physical elastic limits or risks of intermittent adhesive extrusion, but also provides numerical basis for back-end closed-loop adjustments. Through this logical judgment based on frequency domain centroid drift and Euclidean space distance, the production monitoring system can overcome the limitations of false alarms caused by simple time domain amplitude fluctuations, achieving a scientific assessment of the stability of the entire insulating glass production process, the degree of fault, and early warning of faults. The final output serves as a key decision label guiding production scheduling and quality backtracking.

[0089] This invention integrates stator current waveforms with robotic arm vibration characteristics and implements short-time fast Fourier transform to establish a time-frequency domain-based process health diagnostic system. By comparing the measured physical response centroid with the simulated baseline, this scheme can capture metastable oscillations caused by microbubbles in the flow channel or decreased spindle transmission accuracy. This multi-dimensional real-time feedback not only improves the accuracy of robustness evaluation of the production process but also provides non-contact fault early-stage inducing factors location capabilities for the processing site, reducing the downtime risk caused by sudden equipment failures.

[0090] The closed-loop update module of the process knowledge graph is used to aggregate multi-source operating parameters, process health diagnosis results, and final product sealing test results fed back by back-end testing equipment to generate and update a process knowledge graph to guide subsequent production.

[0091] In a specific embodiment of the present invention, the generation and updating of the process knowledge graph for guiding subsequent production includes: obtaining the leakage rate value measured by an offline helium mass spectrometer leak detection device, and generating the final product sealing test result.

[0092] The system acquires the leakage rate value measured by an offline helium mass spectrometer leak detection device at the back-end of the glass production line. During the detection execution phase, the leak detection chamber is evacuated to a preset vacuum level, and the tracer gas diffusion intensity sensed by the mass spectrometer is quantified as the rate of change of the product of pressure and volume over time. Its physical quantity value is typically within... Within the range of Pascals per cubic meter per second, this data is converted into a standard data packet by a range converter to generate the final product sealing test result.

[0093] The multi-source operating condition parameters are matched with the process health diagnosis results using a unique traceability identifier to generate a causal analysis data chain.

[0094] The monitoring unit initiates a global traceability protocol, using the unique traceability identifier printed on the glass frame as the association index to perform causal chain matching between multi-source operating condition parameters stored in the multi-channel buffer and the corresponding process health diagnosis results. This process, through spatiotemporal alignment technology, establishes a logical correspondence between dynamic mechanical losses and environmental weather fluctuations during the glue application process and the instantaneously generated oscillation characteristic deviations, integrating and generating a causal analysis data chain carrying multi-stage data. This causal analysis data chain covers the process execution trajectory and predicted health score for each processing unit.

[0095] The predicted health score is established as follows: the state deviation score corresponding to the process health diagnosis result is used as the process scoring factor, and the leakage rate value corresponding to the final product sealing test result is used as the result scoring factor. The ratio of the process scoring factor to the preset production safety boundary and the ratio of the result scoring factor to the preset standard leakage rate threshold are weighted and summed to obtain the normalized quantitative score, which is the predicted health score.

[0096] Logical verification is performed between the causal analysis data chain and the final product sealing test results to generate and update the process knowledge graph.

[0097] The central logic engine then performs logical verification between the causal analysis data chain and the final product sealing test results. The computational hub uses a consistency error metric formula. Determine the prediction accuracy. (In the formula) A scalar representing the verification deviation between the assessment process early warning model and the actual performance. This represents the measured component of the leakage rate, measured in Pascals per cubic meter per second, directly read from a helium mass spectrometer. This represents the preset critical leakage rate constant that separates the system from failure, and it has the same physical dimensions as the measured value. The ratio calculation within the parentheses eliminates the complex physical dimensions of flow rate, generating a proportionality coefficient that reflects the quality of airtightness. The symbol... Receive the dimensionless process health diagnostic results scalar from the preceding process stage. Obtain the residual value between the prediction and the actual measurement by subtracting the two dimensionless scalars.

[0098] The process early warning model is a health early warning logic model based on the spatiotemporal coupling characteristics of the entire sealing process, and it is established based on the following technical foundations:

[0099] Using the measured frequency energy center of gravity as the core input, the real-time health status and fault type of the sealing process are determined by comparing it with the simulated fault-free benchmark value and combining the quantitative relationship between the state deviation score D and the preset production safety boundary.

[0100] Using causal analysis data chains as spatiotemporal constraints, the process execution trajectory and predicted health score of a single insulating glass unit are integrated to establish a dynamic correlation between process parameters, operating condition fluctuations and product health.

[0101] Based on the dynamic association weight model of the process knowledge graph, the sensitivity coefficient of the model is adaptively updated by using nonlinear drift verification bias, so as to realize early warning and trend prediction of the sealing quality of subsequent processing monomers; the output of the process early warning model is the predicted health score, which is used to characterize the sealing quality and sealing risk level of the processed monomers.

[0102] The process knowledge graph adopts a weighted directed graph structure, where nodes represent process parameters or states, and edges represent causal relationships and their influence weights.

[0103] Specifically, the final output production decision is a structured database containing the probability of corresponding equipment state evolution characteristics and product sealing performance, i.e., a process knowledge graph. This graph establishes a dynamic correlation weight model from equipment wear and environmental fluctuations to the final leakage risk by integrating process stress feedback in the causal analysis data chain with the finished product performance in the helium testing stage. Once the verification stage finds a non-linear drift between the predicted health and the actual leakage index in the causal chain, the graph will use the verification deviation scalar to redistribute the sensitivity coefficients between process path nodes, completing the generation and updating of the process knowledge graph. This not only achieves a causal closed loop for a single production cycle but also provides an empirical logic library validated by actual production for subsequent adaptive adjustment of process parameters based on minor changes in the production environment.

[0104] In a specific embodiment of the present invention, the generation and updating of the process knowledge graph for guiding subsequent production further includes: extracting the deviation residual value between the predicted result and the actual result in the causal analysis data chain, and generating a graph learning error vector.

[0105] The control microprocessor reads the accumulated data stack from the memory bus during the logic verification process and extracts the residual value between the predicted health diagnosis conclusion in the causal analysis data chain and the actual product leak detection performance. This residual value serves as a negative metric for the accuracy of the system simulation, and its value typically fluctuates within a small range of (-0.15, +0.15). The computation engine then uses a matrix transformation module to array and rearrange the discrete residual data according to the pipeline action sequence, generating a multidimensional tensor structure graph learning error vector to guide subsequent neural network structure optimization.

[0106] The weights of associated nodes in the process knowledge graph are adjusted using a weight correction algorithm to generate an updated process decision path.

[0107] To ensure the pre-action allocation strategy accurately adapts to process execution deviations caused by sudden changes in the batch properties of colloidal materials or irreversible physical wear of the extrusion transmission mechanism, the system's underlying computation module activates an error backpropagation iteration tool. Through a pre-defined weight correction algorithm, it performs step-by-step reverse deduction and compensation adjustments to the weights of all related nodes establishing causal relationships in the process knowledge graph. The network connection coefficient fine-tuning rules for controller mapping execution are deconstructed into update relation equations: ,in The extreme values ​​are assigned to the weights of the feedforward network associated nodes after the update at the end of the current single frame data iteration. These values ​​exist in the form of pure mapping scaling coefficients and are dimensionless parameters. The reference empirical coefficients for a batch from the process knowledge graph, extracted from the processor's temporary access memory, are dimensionless parameters. The step size of the iterative learning rate is limited to a safe range of 0.005-0.05 by an internal interception module to suppress algorithm feedback divergence. This refers to the independent numerical values ​​of the interference-free deviation residuals extracted from the learning error vector of the graph. The algebraic superposition and scaling fusion processes both employ dimensionless hierarchical parameter calculations to avoid the distortion risks caused by the cascading operation of variables from different physical domains. After multiple node gain reorganizations and ratio updates, the central computing node refreshes the optimal planning topology path describing the direction from multi-source external disturbance variables to the electromechanical mechanism correction torque, generating an updated process decision path containing new mapping rules.

[0108] It should be noted that the associated nodes of the process knowledge graph are divided into four levels: multi-source operating condition parameter layer, process execution layer, fault diagnosis layer, and sealing performance layer. The multi-source operating condition parameter layer includes nodes such as wear coefficient and environmental feature vector; the process execution layer includes nodes such as pump pressure and auxiliary heating power; the fault diagnosis layer includes nodes such as discontinuous colloid extrusion and mechanical resonance; and the sealing performance layer includes nodes such as leakage rate. The nodes in each layer are connected through causal relationships and assigned initial weights.

[0109] The updated process decision path is fed back into the calculation loop that obtains multi-source operating parameters, generating a closed-loop process parameter combination with self-evolutionary characteristics.

[0110] For example, after the updated process decision path is fed back to the multi-source operating condition parameter calculation loop, it is coupled with the real-time collected multi-source operating condition parameters, including wear coefficient, environmental feature vector, and glass specification size parameters, using a time-series weighted fusion algorithm. The weight coefficient of the process decision path is 0.6, and the weight coefficient of the real-time multi-source operating condition parameters is 0.4. The fused result is used as the basis for generating the initial vector of the simulation boundary.

[0111] The main program at the central control station bypasses the workshop's field communication network gateway, reverses the updated process decision path as a benchmark template, sends it to the calculation loop hub of the underlying field control module for real-time acquisition of multi-source operating parameters. When the edge photoelectric scanning sensor detects a new insulating glass component reaching the glue-filling trigger physical scale, the field CNC equipment, combining real-time environmental measurement data and the wear status of the front-end actuators, calculates and issues feedforward pump pressure adjustment parameters and auxiliary heating setpoint parameters based on the updated process decision path.

[0112] The output digital instruction set includes servo torque setpoint parameters and thermal component control pulses, and incorporates a pre-dynamic offset variable system formed iteratively from historical helium mass spectrometry leak detection conclusions, creating a closed-loop process parameter combination with self-evolving characteristics. This parameter combination autonomously iterates and corrects errors as the system runs, overcoming the effects of accumulated mechanical fatigue and microclimate fluctuations over time, avoiding long-term yield degradation caused by static process presets, and realizing advanced production closed-loop control of the processing system from single-piece defect alarms to whole-line adaptive self-healing.

[0113] The technical principle of this invention lies in constructing a feedforward preventive online monitoring and early warning architecture that integrates digital twins, composite control, and closed-loop learning. The system first acquires multi-source operating parameters from multi-dimensional heterogeneous data sources, including glass geometric constraints, colloidal constitutive properties, and environmental thermodynamic states, using these as initial boundary conditions to drive a multi-physics coupled simulation engine. By solving transient rheological and heat conduction equations, the system evolves a digital colloid characterizing the entire material forming cycle in virtual space and quantifies and analyzes the sealing weaknesses caused by spatial stress and abnormal filling states. Guided by this risk prediction, the system derives a dynamic target curve for proactively eliminating hidden dangers by referencing an ideal physical wetting model. This decouples the intervention command space, synchronously driving the main control actuator of the sealing machinery to adjust the main fluid dynamic factors and stimulating the micro-environment control module to release local energy interference thermodynamic micro-factors, executing proactive shaping through macro-micro synergy. Simultaneously with the implementation of the aforementioned physical interference, the system extracts the oscillation spectrum of the process by capturing the weak flutter features at the bottom layer and performs real-time verification and checking with the pre-simulation reference system. Finally, the system uses cross-temporal tracing to import the complete data chain containing the pre-set causes and actual confirmed cases into the machine learning hub, continuously refines the working condition response rules with guiding significance and updates them to the process knowledge graph, giving the control model an adaptive evolutionary mechanism with dynamic extension.

[0114] The main technical effect of this invention lies in transforming the quality control paradigm of the insulating glass sealing process from passive post-event detection to proactive pre-event prevention. Through multi-source pre-simulation and pre-analysis, blind spots induced by drastic environmental changes and complex flow resistance are eliminated, enabling production equipment to predict invisible interface defects. A more significant advancement is that the system abandons the traditional, hysteresis-based repair method that relies solely on mechanical flow adjustment. Instead, it employs a composite parameter collaborative interference network combining macroscopic motion control with microscopic targeted thermal excitation. This three-dimensional interaction of multiple physical fields creates a synergistic compensation effect, effectively mitigating the risks of edge cavitation and interlayer delamination caused by irregularly shaped slits or low-temperature viscosity, thereby improving the anti-interference robustness and process stability of the flexible production line. Simultaneously, the hidden fluid dynamics disturbances are transformed into highly recognizable frequency domain feature spectra for second-level online cross-verification. This is supplemented by an evolutionary cognitive network that connects the extreme value data of the entire cycle. This allows the entire production line to not only get rid of the dependence on the manual experience of craftsmen and the remedial costs incurred by subsequent leak detection and repair, but also to acquire the intelligent evolutionary ability to face new rubber batches and extreme weather through self-verification and long-term game learning, ensuring the high uniformity of the terminal mouth barrier performance and the absolute yield guarantee.

[0115] This invention achieves a deep self-evolving closed loop in production logic by constructing a process knowledge graph that includes multi-source operating parameters and final sealing performance test results. The system can reverse-drive the correction of associated node weights based on the actual performance indicators obtained from helium mass spectrometry leak detection, dynamically eliminating the effects of mechanical wear deviations and seasonal environmental drift caused by long-term equipment service. This adaptive learning capability based on measured data chains ensures that the process decision path is continuously optimized as service time increases, guaranteeing a high degree of consistency in finished product sealing quality under large-scale continuous production conditions.

[0116] It should also be noted that the various threshold values, parameter ranges, and judgment criteria described in this invention are specifically based on historical data of the operation of insulating glass sealing equipment and processes, actual production failure cases, and a large number of process experiments and simulation test results. At the same time, they are combined with the experience of experts in the field of insulating glass production, and comprehensively consider factors such as the actual operating environment of the production site, the design life and performance indicators of key components of the sealing equipment, the rheological properties of the sealant material, and the core requirements for quality control of the sealing process. They have been determined after repeated calibration and verification, and have a solid scientific basis and practical operability. Moreover, the methods for setting the relevant threshold values ​​are relatively mature in the existing technology, and will not be elaborated here.

[0117] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0118] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0119] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0120] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A smart monitoring and fault early warning system for insulating glass production, characterized in that, include: The multi-source working condition sensing module is used to acquire the specifications and dimensions of the glass to be processed, the physical properties of the colloidal material, and the environmental parameters of the production site, and generate multi-source working condition parameters. The digital twin simulation module is used to simulate and calculate the evolution of multi-source working parameters using a physical field coupling algorithm, and generate a digital colloid characterizing the colloid forming state and a prediction map of sealing weak points indicating the spatial location of defects. The process target planning module is used to extract failure assessment factors from the sealing weak point prediction map and perform calculations in combination with the ideal sealing index stored in the system to generate a dynamic target curve. The composite parameter modulation module is used to analyze the dynamic target curve to obtain the drive response command, and generate a composite parameter cooperative modulation strategy by performing pressure and speed control and local heat compensation actions on the sealing equipment. The collaborative modulation strategy execution monitoring module collects multi-physical dimension feedback signals from the sealing equipment during the execution of the composite parameter collaborative modulation strategy, and performs time-frequency domain transformation processing on the multi-physical dimension feedback signals to generate process oscillation spectrum characteristics. The signal feature diagnosis module is used to compare the process oscillation spectrum features with the simulation expected features derived from the digital colloid, and output process health diagnosis results for evaluating production robustness. The closed-loop update module of the process knowledge graph is used to aggregate multi-source operating parameters, process health diagnosis results, and final product sealing test results fed back by back-end testing equipment to generate and update a process knowledge graph to guide subsequent production.

2. The intelligent monitoring and fault early warning system for insulating glass production according to claim 1, characterized in that, The generated multi-source operating condition parameters include: The cumulative discharge stroke data of the sealing equipment is obtained and the loss is quantified to generate a wear coefficient that reflects the performance of the pump body. The temperature and humidity time series of the production site are acquired and environmental normalization is performed to generate an environmental feature vector. By integrating wear coefficient, environmental feature vector, and dimensional parameters, multi-source operating condition parameters are generated.

3. The intelligent monitoring and fault early warning system for insulating glass production according to claim 2, characterized in that, The generated digital colloid characterizing the colloid molding state and the predicted map of sealing weak points indicating the spatial location of defects include: Extract the spatial geometric constraints, flow constitutive parameters obtained from the physical property parameters of colloidal materials, wear coefficient, and environmental feature vector from the multi-source working condition parameters to generate the simulation boundary initial vector; The initial vector of the simulation boundary is input into a preset non-Newtonian fluid dynamics equation for iterative solution to generate a digital colloid. The stress distribution of the digital colloid is scanned and abnormal nodes exceeding the allowable threshold stored in the system are identified, generating a prediction map of weak points in the seal.

4. The intelligent monitoring and fault early warning system for insulating glass production according to claim 1, characterized in that, The generation of the dynamic target curve includes: Identify the set of risk coordinates in the prediction map of weak sealing points and obtain the material flow channel resistance characteristics corresponding to the risk coordinates to generate a local compensation demand matrix. Based on the local compensation demand matrix, the ideal physical quantities required to maintain interfacial adhesion are calculated, and the desired apparent viscosity value and desired interfacial wetting energy value are generated. The desired apparent viscosity and desired interfacial wetting energy values ​​are arranged along the production time axis to generate a dynamic target curve.

5. The intelligent monitoring and fault early warning system for insulating glass production according to claim 1, characterized in that, The generated composite parameter cooperative modulation strategy includes: The main pump pressure regulation is extracted by analyzing the dynamic target curve, and the main shaft control signal is generated. Calculate the power drive value required for the micro-focused heating unit based on the desired interface wetting energy value, and generate auxiliary thermal compensation commands. Simultaneously send spindle control signals and auxiliary thermal compensation commands to drive hardware actions and generate a composite parameter collaborative modulation strategy.

6. The intelligent monitoring and fault early warning system for insulating glass production according to claim 1, characterized in that, The oscillation spectrum characteristics of the generation process include: The stator current waveform signal of the drive motor of the sealing equipment is acquired and the vibration acceleration signal at the end of the robotic arm is extracted to generate the original physical feedback signal. A short-time fast Fourier transform is performed on the original physical feedback signal to generate a frequency domain distribution map of the signal power. Energy amplitude characteristics of a specific frequency band are extracted from the signal power frequency domain distribution map to generate process oscillation spectrum characteristics.

7. The intelligent monitoring and fault early warning system for insulating glass production according to claim 1, characterized in that, The output is a process health diagnostic result used to evaluate production robustness, and includes: The frequency energy centroid of the oscillation spectrum features is mapped to generate the measured feature vector. Calculate the Euclidean distance between the measured feature vector and the simulated expected feature, and generate a state deviation score; The state deviation score is compared with the production safety boundary defined within the system, and the process health diagnosis result is output.

8. The intelligent monitoring and fault early warning system for insulating glass production according to claim 1, characterized in that, The generated and updated process knowledge graph, used to guide subsequent production, includes: Obtain the leakage rate value measured by the offline helium mass spectrometer leak detection equipment and generate the final product sealing test result; The multi-source operating condition parameters are matched with the process health diagnosis results according to the unique traceability identifier to generate a causal analysis data chain; Logical verification is performed between the causal analysis data chain and the final product sealing test results to generate and update the process knowledge graph.

9. The intelligent monitoring and fault early warning system for insulating glass production according to claim 1, characterized in that, The generation and updating of the process knowledge graph to guide subsequent production also includes: Extract the residual values ​​of the deviation between the predicted results and the actual results in the causal analysis data chain, and generate a graph learning error vector; The weights of associated nodes in the process knowledge graph are adjusted using a weight correction algorithm to generate an updated process decision path. The updated process decision path is fed back into the calculation loop that obtains multi-source operating parameters, generating a closed-loop process parameter combination with self-evolutionary characteristics.

Images