UV adhesive viscosity real-time regulation and control system based on edge calculation

Abstract

The invention discloses a UV adhesive viscosity real-time regulation and control system based on edge calculation, and particularly relates to the technical field of UV adhesive production and processing intelligent regulation and control, and the UV adhesive viscosity real-time regulation and control system comprises a viscosity sensing unit, an edge calculation node, an execution unit and a far-end cloud platform in a production field. The viscosity sensing unit collects seven types of parameters such as UV glue temperature, shearing rate and solid content, and the self-cleaning assembly is matched to guarantee the detection precision; a self-adaptive fuzzy PID algorithm is built in the edge computing node, a viscosity value and deviation are calculated in real time, PID parameters are dynamically adjusted, disturbance compensation is introduced, and a control instruction is generated; the execution unit preferentially starts component regulation and control or environment regulation and control according to the viscosity deviation and cooperates with an instruction to realize linkage; the cloud platform stores data, performs remote monitoring, optimizes algorithm parameters through federal learning, and pre-warns equipment faults through a built-in predictive maintenance module. According to the system, the real-time performance and precision of regulating and controlling the viscosity of the UV adhesive are improved, the failure outage rate is reduced, and the large-scale production requirement is met.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology in UV adhesive production and processing, and more specifically, to a real-time viscosity control system for UV adhesives based on edge computing. Background Technology

[0002] UV adhesives are widely used in electronic component packaging, optical device bonding, and precision instrument assembly due to their advantages such as fast curing speed, high bonding strength, and good environmental performance. Their viscosity directly determines the product molding quality. In electronic packaging, excessively high viscosity can easily lead to uneven packaging gaps, while excessively low viscosity can easily cause adhesive overflow. Therefore, the viscosity of UV adhesives needs to be precisely controlled during the production process.

[0003] Currently, there are two main technical solutions for UV adhesive viscosity control. One is offline sampling and detection control, where adhesive samples are periodically extracted from the production vessel manually, and the viscosity is measured using a rotational viscometer. Based on the measurement results, diluents or thickeners are manually added to adjust the composition, or the reactor temperature is adjusted. The other is a preliminary online control solution, where some companies attempt to use a single temperature sensor or shear rate sensor to collect parameters and control the heating device or stirring speed using a conventional PID algorithm. However, in practical use, this still has some drawbacks, such as: on the one hand, existing technologies only use temperature as a single parameter for control, which cannot cope with viscosity changes caused by the coupling of multiple factors, resulting in limited control accuracy.

[0004] On the other hand, existing technologies use offline sampling methods, which have serious lag and cannot meet the needs of real-time production. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a real-time viscosity control system for UV adhesives based on edge computing, which solves the problems mentioned in the background art through the following solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time viscosity control system for UV adhesive based on edge computing, comprising: a viscosity sensing unit, an edge computing node, an execution unit, and a cloud platform located at a remote location, all installed at the production site. Viscosity sensing unit: used to collect viscosity-related parameters of UV adhesive in real time and send the parameters to the edge computing node; Edge computing node: It has a built-in control algorithm module for receiving viscosity-related parameters, processing and calculating them in real time to obtain the current viscosity value, and calling the control algorithm module to generate real-time control instructions for adjusting the viscosity based on the comparison result between the current viscosity value and the preset target viscosity threshold. The instructions are then sent to the execution unit, and the processed data is uploaded to the cloud platform. Execution unit: used to receive and execute the real-time control command for regulating viscosity, and to regulate the viscosity of UV adhesive in real time by adjusting the composition or environmental conditions; Cloud platform: Used to receive and store data uploaded by edge computing nodes, perform remote monitoring and parameter setting, analyze historical data, perform nonlinear optimization of the parameters of the control algorithm module in the edge computing node, and send the optimized parameter model to the edge computing node to update its control algorithm module.

[0007] Preferably, the viscosity-related parameters of the UV adhesive include temperature, shear rate, shear stress, solid content, pre-curing degree, ambient humidity, and stirring speed.

[0008] Preferably, the viscosity sensing unit includes a non-contact ultrasonic sensing module, a temperature sensing module, an ambient humidity sensing module, a shear rate sensing module, a shear stress sensing module, a near-infrared solids content sensing module, an ultraviolet transmittance sensing module, a stirring speed sensing module, and a self-cleaning component.

[0009] Preferably, the control algorithm module of the edge computing node is an adaptive fuzzy PID algorithm, and the execution logic of this algorithm includes: The deviation between the current viscosity value and the target viscosity threshold, as well as the rate of change of the deviation, are calculated in real time. Through the fuzzy inference module, the core parameters of the PID are dynamically adjusted based on the preset fuzzy rule base, and a disturbance compensation term is introduced to correct the final control command output.

[0010] Preferably, the real-time control commands for regulating viscosity include component adjustment commands, environmental parameter adjustment commands, and regulation coordination commands; The component adjustment instructions include diluent injection instructions and thickener injection instructions. The instruction parameters include injection volume, injection rate, and injection duration. When the UV adhesive viscosity is detected to be higher than the target threshold, a diluent injection instruction is issued. When the viscosity is lower than the target threshold, a thickener injection instruction is issued. The environmental parameter adjustment instructions include temperature control instructions and humidity control instructions. The temperature control instruction parameters are the target temperature value and the rate of temperature change; the humidity control instruction parameters are the target humidity value, and the humidity adjustment lags behind the temperature adjustment by 0.3-0.5 seconds. The control and coordination instructions include pause, restart, and weight allocation instructions.

[0011] Preferably, the execution unit includes a component control subunit and an environmental control subunit. When the absolute value of the viscosity deviation is >10%, the component control subunit is activated first; when the absolute value of the viscosity deviation is <10%, the environmental control subunit is activated first. The component regulation subunit uses a servo-driven micro-injection pump to precisely inject diluent or thickener into the UV collagen solution, and an electromagnetic flow sensor is installed at the pump outlet to provide real-time feedback on the actual injection volume. The environmental control subunit adopts a temperature-humidity dual closed-loop controller. Temperature control is achieved through a semiconductor cooling / heating element, and humidity control is achieved through a dehumidification module and a humidification module.

[0012] Preferably, the nonlinear optimization of the parameters of the control algorithm module in the edge computing node adopts a federated learning framework.

[0013] Preferably, the cloud platform also has a built-in predictive maintenance module, which includes a data preprocessing unit, a fault prediction model, and an early warning push unit.

[0014] The technical effects and advantages of this invention are as follows: 1. This invention uses a viscosity sensing unit that collects 7 types of parameters, combined with a self-cleaning component consisting of inert gas purging and a flexible scraper, to avoid interference from colloid residue, ensure data acquisition accuracy, provide an accurate data foundation for subsequent regulation, and solve the problems of limited perception and large data deviation in traditional single-parameter sensing. 2. This invention utilizes an adaptive fuzzy PID algorithm built into the edge computing node to complete data processing, deviation analysis, and instruction generation within 0.5 seconds, dynamically adjusting PID parameters and introducing disturbance compensation; the execution unit prioritizes component or environmental control based on viscosity deviation, achieving "rapid correction of large deviations and fine adjustment of small deviations"; 3. This invention is based on a federated learning framework, which integrates historical data from multiple nodes to optimize the algorithm parameters of edge computing, breaking through the limitation of "data silos" and improving the control response speed by 30%. At the same time, through continuous iterative training, the system adapts to complex working conditions such as raw material batch differences and environmental fluctuations, and the control stability is continuously optimized. 4. This invention uses fault feature extraction and gradient boosting decision tree model to provide early warning of faults such as sensor contamination and injection pump blockage, minimizing the risk of sudden downtime caused by traditional passive maintenance, significantly reducing downtime due to faults, and improving the continuity and efficiency of UV adhesive production. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is a schematic diagram of the self-cleaning judgment process of the present invention; Figure 3 This is a flowchart illustrating the execution unit control process of the present invention. Figure 4 This is a flowchart of the core workflow of the cloud platform of the present invention. Detailed Implementation

[0016] 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.

[0017] As attached Figure 1-4 The UV adhesive viscosity real-time control system based on edge computing shown includes: a viscosity sensing unit, an edge computing node, an execution unit, and a cloud platform located at a remote location, all installed at the production site. Viscosity sensing unit: used to collect viscosity-related parameters of UV adhesive in real time and send the parameters to the edge computing node; It should be specifically noted that the viscosity sensing unit includes a non-contact ultrasonic sensing module, a temperature sensing module, an ambient humidity sensing module, a shear rate sensing module, a shear stress sensing module, a near-infrared solid content sensing module, an ultraviolet light transmittance sensing module, a stirring speed sensing module, and a self-cleaning component. It should be further noted that the viscosity sensing unit is fixed in the critical fluid path of the UV adhesive production site. The specific deployment location, determined through flow field simulation and actual testing, is in the lower middle part of the side wall of the mixing tank, 1 / 3 of the tank height from the bottom, avoiding mixing dead zones and horizontal sections of the delivery pipeline. The upstream pipeline length is ≥5 times the pipe diameter, and the downstream length is ≥3 times the pipe diameter, away from valves and elbows. The installation status of each submodule is as follows: Non-contact ultrasonic sensing module: Employs an industrial-grade non-contact ultrasonic sensor with a working frequency of 5MHz; detection distance of 10-30mm; acoustic impedance measurement range of 1*10. 6 -3*10 6 Pa・s / m, measurement accuracy ±1%FS, output 4-20mA analog quantity, sensor detection surface perpendicular to the colloid surface, spacing 15mm, installation position ≥300mm from the stirring shaft.

[0018] Temperature sensing module: adopts a platinum resistance temperature sensor with an accuracy class A, a measurement range of -20-80℃, and an error of ±0.1℃. It is equipped with a temperature transmitter and outputs a 4-20mA signal. It is an insertion type with an insertion depth of 1 / 3 of the total height of the colloid.

[0019] Ambient humidity sensing module: adopts an industrial-grade digital humidity sensor with a measurement range of 0-100%RH, an accuracy of ±2%RH (20-60%RH range), an I2C interface, and a response time of ≤8s; it is fixed near the ventilation opening above the reactor, ≥500mm from the surface of the colloid, and equipped with a dust cover with a 0.5mm aperture.

[0020] Shear rate / shear stress sensing module: Uses an industrial-grade rotational viscometer module, with a shear rate measurement range of 0-1000s. -1 Accuracy ±2%; Shear stress measurement range 0-100Pa, accuracy ±3%; Connect the sensor rotor to the stirring shaft of the reactor, with coaxiality error ≤0.1mm, the rotor immersion depth is 2 / 3 of the colloid height, and it rotates synchronously with the stirring shaft, with the speed consistent with the stirring speed.

[0021] Near-infrared solids content sensing module: adopts an industrial-grade near-infrared sensor with a wavelength range of 900-1700nm, covering the characteristic absorption peaks of resin and diluent in UV adhesive; resolution of 5nm, measurement accuracy of ±0.5% (solids content range of 30%-70%); fixed on the outside of the transparent observation window of the reactor, with the emitting and receiving ends arranged symmetrically, and the path length of light passing through the colloid is 50mm (to ensure optical path stability).

[0022] Ultraviolet transmittance sensing module: adopts industrial-grade ultraviolet sensor, detection wavelength 365nm; transmittance measurement range 0-100%T, accuracy ±1%T, output 0-5V analog signal; the transmitter and receiver are respectively installed in symmetrical mounting holes on both sides of the reaction vessel (100mm from the colloidal surface).

[0023] Stirring speed sensing module: adopts photoelectric speed sensor with a resolution of 1000 pulses / revolution and a measurement range of 0-300 rpm; outputs NPN open collector signal; fixed at the output shaft end of the stirring motor, 5mm away from the shaft end.

[0024] Self-cleaning components: 99.99% pure nitrogen is used (to prevent oxidation of the UV adhesive), and the pressure is stabilized at 0.3MPa via a pressure reducing valve; a 6mm diameter 316L stainless steel tube is used, one end connected to the nitrogen cylinder, and the other end extends 10mm above the ultrasonic sensing module's detection surface, with a 2mm diameter angled purge port at a 45° angle to the detection surface; the scraper is made of flexible polytetrafluoroethylene (PTFE) with a Shore D50 hardness, a thickness of 2mm, and a width matching the detection surface diameter (20mm), and is controlled by a micro-stepping motor. The machine is driven by a step angle of 1.8° and a scraper stroke of 25mm (covering the edge of the detection surface). Cleaning is triggered every 8 minutes, within the range of 5-10 minutes, adjusted according to the viscosity of the UV adhesive: 5 minutes when the viscosity is >6000cP, and 10 minutes when the viscosity is <4000cP. First, nitrogen purging is started for 30 seconds to remove the adhesive on the surface of the detection surface. The stepper motor drives the scraper to scrape back and forth twice, 10 seconds each time, at a speed of 5mm / s, to remove stubborn residues. Purge again for 10 seconds to ensure that there are no residues on the detection surface.

[0025] It should be further noted that, in order to ensure the accuracy of the collected data, each sensor was calibrated and the raw data was preprocessed.

[0026] Near-infrared solid content calibration: Prepare 100 mL of UV adhesive standard samples with solid contents of 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, and 70% using the weighing method, with an accuracy of ±0.1 g. Pour the standard samples sequentially into transparent containers, place them on the near-infrared sensor test stage, and collect the absorption intensity of each sample at a wavelength of 1200 nm, identifying the characteristic peak of UV adhesive solid content. Use the least squares method to fit a linear regression model between absorption intensity and solid content. Where C is the solid content, I is the absorption intensity, and the model determination coefficient is... .

[0027] UV transmittance-pre-curing degree calibration: Add uncured UV collagen solution to the reactor, pre-curing degree 0%, and measure UV transmittance T. 0， In this embodiment, T0 = 95%T; a UV curing lamp with a power of 100W was used to perform gradient curing of the collagen solution, and the pre-curing degree α (0%, 10%, 20%, ..., 100%) of each sample was measured by differential scanning calorimetry; the transmittance T of each gradient sample was also measured. t Fitting T t The relationship between / T0 and α: .

[0028] Edge computing node: It has a built-in control algorithm module for receiving viscosity-related parameters, processing and calculating them in real time to obtain the current viscosity value, and calling the control algorithm module to generate real-time control instructions for adjusting the viscosity based on the comparison result between the current viscosity value and the preset target viscosity threshold. The instructions are then sent to the execution unit, and the processed data is uploaded to the cloud platform.

[0029] It should be specifically noted that the control algorithm module of the edge computing node is an adaptive fuzzy PID algorithm, and the execution logic of this algorithm includes: The deviation between the current viscosity value and the target viscosity threshold, as well as the rate of change of the deviation, are calculated in real time. Through the fuzzy inference module, the core parameters of the PID are dynamically adjusted based on the preset fuzzy rule base, and a disturbance compensation term is introduced to correct the final control command output.

[0030] It should be further noted that the edge computing node uses an industrial-grade edge gateway, with a processor of 4 cores and 8 threads, a base frequency of 1.0GHz, and a turbo frequency of 3.6GHz; 8GB of DDR4-2666 memory, which can be expanded to 32GB; 128GB of SSD storage, which can be expanded with an external 1TB HDD; and interfaces including 2 Profinet ports, 4 RS485 ports, 2 Gigabit Ethernet ports, and 1 4G module slot to meet the needs of multiple device access.

[0031] It should be further explained that the adaptive fuzzy PID algorithm achieves precise viscosity control through four steps: deviation analysis, fuzzy inference, PID parameter adjustment, and disturbance compensation. The specific logic and parameters are as follows: Define core parameters: Target viscosity threshold η0, set to 5000 cP (allowable fluctuation range ±250 cP) based on the requirements of electronic packaging UV adhesive; Current viscosity value η(t), calculated using a multiple linear regression model; Deviation Design range [-2000, 2000] cP, alarm triggered if exceeded; deviation change rate Where Δt is the sampling period of 0.5s, and the design range is [-1000, 1000] cP / s; the initial PID parameters are determined by trial and error, with the initial proportional coefficient Kp0=2.5, the initial integral coefficient Ki0=0.1, and the initial derivative coefficient Kd0=0.05. Without fuzzy adjustment, the viscosity can be stabilized within the range of η0±10%.

[0032] Viscosity calculation model: Based on the preprocessed parameters of 7 types of sensors, a multiple linear regression model is used for calculation. Where T is temperature, γ is shear rate, τ is shear stress, C is solid content, α is pre-curing degree, RH is ambient humidity, and n is stirring speed; the model was calibrated using 100 sets of experimental data, with a coefficient of determination R² = 0.985 and a calculation error of ±200 cP.

[0033] Fuzzy Inference and PID Parameter Tuning: Fuzzy Subset Definition: Deviation e(t) and Deviation Change Rate e c The fuzzy subsets of (t) are all {NB, NM, NS, ZO, PS, PM, PB} (negative large, negative medium, negative small, zero, positive small, positive medium, positive large); the membership function adopts the triangular membership function.

[0034] Fuzzy rule base establishment: based on deviation e(t) (the difference between the current viscosity and the target value) and the rate of change of deviation e c The magnitude of (t) (the rate of change of deviation) determines the adjustment principles for PID parameters (Kp proportional coefficient, Ki integral coefficient, Kd derivative coefficient) in four scenarios: When the deviation e(t) is large (large negative NB / large positive PB): the core objective is to quickly reduce the deviation and avoid overshoot; Kp takes a large value (large positive PB / medium positive PM) to enhance the proportional action and speed up the response; Ki takes a small value (large negative NB / small negative NS / zero ZO) to weaken the integral action and avoid cumulative overshoot; Kd is based on e c (t) Adjustment: If e c (t) is negative (the deviation is widening), and Kd takes a smaller value (large negative NB / medium negative NM) to suppress divergence; if e c(t) is positive (the deviation is decreasing), and Kd takes a small positive value (positive small PS / zero ZO) to avoid oscillation.

[0035] When the deviation e(t) is moderate (negative moderate NM / positive moderate PM): the core objective is to balance response speed and stability; Kp takes a moderate value (positive moderate PM / positive small PS / negative moderate NM) to avoid excessive oscillation; Ki takes a small value (negative small NS / zero ZO / positive small PS) to slowly accumulate and eliminate residual error; Kd takes a moderate value (negative moderate NM / negative small NS / positive small PS / positive moderate PM), based on e c (t) Directional suppression of deviation change trend.

[0036] When the deviation e(t) is small (small negative NS / small positive PS): the core objective is to suppress oscillations and approach the target value; Kp takes a small value (small positive PS / zero ZO / small negative NS) to avoid over-adjustment; Ki takes a medium value (small negative NS / zero ZO / small positive PS) to gradually eliminate small deviations; Kd takes a large value (small negative NS / small positive PS / medium positive PM) to enhance the differential action to suppress deviation fluctuations.

[0037] When the deviation e(t) is close to zero (ZO): the core objective is to maintain steady-state accuracy and avoid small fluctuations; Kp, Ki, and Kd all take values ​​close to zero (ZO / smaller positive and negative values), weakening the adjustment intensity; if e c (t) is negative (the deviation tends to widen), so slightly increase Kd to suppress it; if e c (t) is positive (the deviation tends to decrease), and Ki is slightly increased to consolidate the steady state.

[0038] Fuzzy reasoning and defuzzification: Using the Mamdani reasoning method, based on the current e(t) and e c The fuzzy value of e(t) is used to match rules in the rule base, and the output weight of each rule is calculated. For example, if e(t) is NM, e c When (t) is NS, the matching rule "Kp=PM, Ki=NS, Kd=NB" has a weight equal to the NM membership degree of e(t) multiplied by e. c The NS membership degree of (t) is obtained; the centroid method is used to defuzzify the data, and the PID parameter adjustment amounts ΔKp, ΔKi, and ΔKd of each rule are weighted and averaged to obtain the final adjustment amount.

[0039] PID parameter update: The updated parameters are as follows , , And it must satisfy the constraints: Kp>0, Ki≥0, Kd≥0.

[0040] Disturbance compensation term design: To address disturbances such as batch differences in raw materials and environmental fluctuations, a disturbance compensation term is introduced. Where 0.02 is the disturbance compensation coefficient; the final control command output is .

[0041] It should be further explained that the core workflow of the edge computing node (cycle 0.5s) is as follows: receiving seven types of raw data from the viscosity sensing unit via the Profinet bus; performing filtering, outlier detection, and normalization to generate preprocessed parameters; substituting these parameters into a multiple linear regression model to obtain the current viscosity value η(t); and calculating e(t) and e... c (t); Invoke the adaptive fuzzy PID algorithm to generate control command parameters; Send the control command to the execution unit via RS485 bus; Pack the preprocessed data, η(t), e(t), PID parameters and control command (JSON format) every 5 seconds and upload them to the cloud platform via 4G module (using MQTT protocol, QoS level 2).

[0042] Execution unit: Used to receive and execute the real-time control command for adjusting viscosity, thereby adjusting the viscosity of the UV adhesive in real time by adjusting the composition or environmental conditions.

[0043] It should be specifically noted that the execution unit includes a component control subunit and an environmental control subunit. When the absolute value of the viscosity deviation is >10%, the component control subunit is activated first; when the absolute value of the viscosity deviation is <10%, the environmental control subunit is activated first. It should be specifically noted that the component control subunit uses two industrial-grade servo-driven micro-injection pumps, one for diluent and one for thickener; the injection accuracy is ±0.1μL, ensuring accurate component adjustment; the injection rate range is 0.1-100μL / min, adapting to different deviation requirements; the syringe capacity is 10mL; the drive method is a stepper motor with a step angle of 1.8° and 16 subdivisions, ensuring stable operation; an electromagnetic flow sensor is installed on the injection pump outlet pipeline, with a measurement range of 0-200μL / min and an accuracy of ±2%FS, providing real-time feedback on the actual injection volume; the diluent is ethyl acetate with a purity of 99.9%, and the addition amount is ≤5%; the thickener is fumed silica with a particle size of 20nm, and the addition amount is ≤3%.

[0044] It should be specifically noted that the temperature control module of the environmental control subunit uses a semiconductor cooling / heating chip with a power of 60W, a maximum temperature difference of 60℃, and supports bidirectional temperature control; an industrial-grade PID temperature controller with a control accuracy of ±0.1℃ and output PWM signal; an aluminum heat sink with an area of ​​500cm² + a 12V cooling fan with a wind speed of 3m / s; the semiconductor cooling chip is attached to the outside of the reactor jacket with thermally conductive silicone grease (thermal conductivity 3.5W / (m・K)), and the heat sink is installed on the other side of the cooling chip; The humidity control module uses a semiconductor dehumidifier with a dehumidification capacity of 50mL / day and an operating voltage of 12V; an ultrasonic humidifier with a humidification capacity of 100mL / h and an atomized particle diameter of 5μm; and a dedicated humidity controller with a control accuracy of ±2%RH. The dehumidifier and humidifier are installed in the ventilation ducts above the reactor, and the airflow speed is controlled by a wind speed sensor (measurement range 0-5m / s).

[0045] It should be specifically noted that the real-time control commands for regulating viscosity include component adjustment commands, environmental parameter adjustment commands, and regulation coordination commands; It should be specifically noted that the component adjustment instructions include diluent injection instructions and thickener injection instructions, and the instruction parameters include injection volume, injection rate and injection duration; when the UV adhesive viscosity is detected to be higher than the target threshold, a diluent injection instruction is issued; when the viscosity is lower than the target threshold, a thickener injection instruction is issued. It needs to be further explained that when At that time, priority should be given to adjusting the components and the diluent injection volume. Where 0.01 is the injection volume coefficient, and calibration shows that for every 0.01 μL of diluent injected, the viscosity decreases by 1 cP; injection rate Injection completed within 10 seconds; electromagnetic flow sensor provides real-time feedback on the actual injection volume V. act If V act If V < 0.95V, increase the stepper motor speed; if V act >1.05V, reduce speed, V act Within the range of V ± 5%.

[0046] It should be further noted that the thickener injection volume The value of 0.012 is the injection volume coefficient. The thickener has a weaker effect, so the coefficient is slightly larger. The injection rate and duration are the same as those for the diluent, and are also controlled in a closed loop by a flow sensor.

[0047] It should be specifically noted that the environmental parameter adjustment instructions include temperature control instructions and humidity control instructions. The parameters of the temperature control instructions are the target temperature value and the rate of temperature change; the parameters of the humidity control instructions are the target humidity value, and the humidity adjustment lags behind the temperature adjustment by 0.3-0.5 seconds. It needs to be further explained that when At that time, prioritize environmental adjustments and temperature control: calculate the target temperature. That is, for every 500 cP increase in deviation, the target temperature increases by 1℃, and as the temperature increases, the viscosity decreases; the temperature change rate is 0.5℃ / s; the PID temperature controller adjusts the power of the semiconductor cooling / heating element through the PWM signal.

[0048] Humidity control: The target humidity is 50%RH by default. Because the UV adhesive is less sensitive to humidity than to temperature, it is only adjusted when RH>65% or RH<35%. The humidity adjustment lags behind the temperature adjustment by 0.4s. The execution logic is to start the dehumidifier when RH>65% and start the humidifier when RH<35% until RH stabilizes at 50%±2%RH.

[0049] It should be noted that the aforementioned regulatory coordination instructions are used to define the linkage rules between component adjustment and environmental adjustment, including pause, restart instructions and weight allocation instructions.

[0050] It should be further explained that when |e(t)| is between 8% and 12%η0, both component and environmental regulation need to be activated simultaneously, and the coordinated command is achieved through weight allocation; let the component regulation weight be w1 and the environmental regulation weight be w2. ;when When w1=0.3, w2=0.7; when w2=0.3, w1=0.3, w2=0.7; when w1=0.3, w2=0.7 ...3, w2=0.7, w2=0.7, w2=0.7, w2=0. At that time, w1=0.7, w2=0.3; , When an edge computing node detects an anomaly, it issues a pause command, and the execution unit immediately stops working. After the anomaly is resolved, a restart command is issued through the cloud platform or local touchscreen, and the execution unit resumes control from the current state. The weight allocation command is used to dynamically allocate the contribution ratio of the component control subunit and the environmental control subunit in the control process according to the magnitude of the viscosity deviation.

[0051] It should be further explained that the execution unit feeds back its working status to the edge computing node every 0.5 seconds, including the working status (running / stopping), actual injection volume, and flow sensor value of the injection pump in the component control subunit; the current temperature, target temperature, current humidity, target humidity, and cooling / heating plate power of the environmental control subunit; and fault information, such as injection pump stall (motor current > 1A) and temperature exceeding the 15-45℃ range.

[0052] Cloud platform: Used to receive and store data uploaded by edge computing nodes, perform remote monitoring and parameter setting, analyze historical data, perform nonlinear optimization of the parameters of the control algorithm module in the edge computing node, and send the optimized parameter model to the edge computing node to update its control algorithm module.

[0053] It should be further noted that the cloud platform is configured with a remote cloud server, 4 cores, 8GB of memory, 500GB of SSD storage, and 10Mbps bandwidth, supporting elastic expansion; the backend adopts an industrial-grade backend development framework and uses a layered design (control layer, service layer, data access layer); the database uses a relational database and a caching database; the frontend uses a responsive frontend development framework; and the middleware uses an MQTT message broker and a distributed search engine.

[0054] It should be further explained that the data table structure during data storage includes: Device table: storage device ID, name, installation location, IP address, online status; Sensor data table: timestamp, device_id, temperature, humidity, shear rate, solid content, etc.; Viscosity data table: timestamp, device_id, η(t), e(t), e c (t); PID parameter table: timestamp, device_id, Kp, Ki, Kd, ​​Δu; Control command table: timestamp, device_id, command type, parameter, execution status; Fault table: timestamp, device_id, fault type, fault description, processing status; Real-time data (within 1 year) is stored in a relational database, and historical data (more than 1 year) is archived to cloud object storage. The data backup strategy is daily incremental backup + weekly full backup.

[0055] It should be further explained that the front-end visualization interface provides multi-dimensional monitoring functions. The core interface includes: a device overview page, displaying the online / offline status of all edge nodes (green = online, gray = offline); a real-time viscosity value dashboard (green = normal <5% deviation, yellow = warning 5%-10% deviation, red = alarm >10% deviation); an alarm information pop-up window displaying the faulty device ID, type, and time; a sensor data page displaying the real-time values, historical curves, and average, maximum, and minimum values ​​of each sensor over a certain time period; a control command page displaying the currently executed command type, parameters, and execution status; historical commands can be queried by device ID and time range; administrators can manually issue test commands to verify the execution unit response; a viscosity trend analysis page displaying the viscosity value change trend curve, overlaid with a target threshold line (5000cP); predicting viscosity changes for the next 1 hour based on a time series prediction model; and generating a viscosity control report for each batch of UV adhesive.

[0056] It should be further noted that the configurable parameters include: production parameters such as target viscosity η0 (4000-6000 cP), temperature range (15-45℃), and humidity range (30-70%RH); algorithm parameters such as PID initial parameters Kp0, Ki0, and Kd0, and fuzzy rule base weights; sensor parameters such as near-infrared solids content calibration coefficient and ultraviolet transmittance benchmark value T0; and equipment parameters such as self-cleaning interval (5-10 min) and injection pump accuracy calibration coefficient.

[0057] It should be noted that the nonlinear optimization of the parameters of the control algorithm module in the edge computing node adopts a federated learning framework.

[0058] It should be further explained that the reinforcement learning model is designed with a state space S containing 6 dimensions. (Current viscosity, deviation, rate of change of deviation, temperature, humidity, solid content); Action space A: includes 3 dimensions ; Reward function R: The smaller the deviation, the greater the reward. The more stable the parameter adjustments, the greater the reward; balance control precision and stability.

[0059] It should be further explained that the model training process includes an initialization phase: the cloud platform initializes the global model parameters θ. global The data is distributed to 10 clients; each client loads its local historical data (approximately 100,000 records from the past 7 days) and initializes its local model θ. local =θ global Local training phase (7 days per round): Each client trains a local model using a deep Q-network algorithm, randomly sampling 1000 sets of state-action-reward samples from local data; constructs a Q-network using the ReLU activation function, 2 hidden layers with 64 neurons per layer; employs the Adam optimizer with a learning rate of 0.001 and cross-entropy loss function, training for 100 rounds with 1000 iterations per round; after training, the local model parameters θ are obtained. local Model aggregation phase: Each client will aggregate θ local With local data sample size n i Uploaded to the cloud platform; the cloud platform calculates the weight of each client. .

[0060] The new parameters of the global model are calculated using a weighted average: The cloud platform will θ global_new The data is distributed to each client, and the client updates its local model θ. local =θ global_new One round of training is completed every 7 days; after 8 rounds of training, the model converges.

[0061] It should be noted that the cloud platform also has a built-in predictive maintenance module, which includes a data preprocessing unit, a fault prediction model, and an early warning push unit.

[0062] It should be further explained that the data preprocessing unit includes data cleaning, feature extraction, and feature normalization; the fault prediction model adopts a gradient boosting decision tree model, collecting 300 sets of equipment fault data and 700 sets of normal data, and dividing the training set and test set in a 7:3 ratio; 5-fold cross-validation is used to optimize the model parameters, with a learning rate of 0.05, a tree depth of 5, and 100 iterations; real-time features are input, and the model outputs the fault probability P; the early warning push unit classifies the risk level according to the fault probability P: P<0.3 (low risk, no warning required), 0.3≤P<0.7 (medium risk, SMS warning), and P≥0.7 (high risk, push warning). The push content includes the equipment ID, fault type, predicted fault time, and maintenance suggestions, and records the early warning processing results for model iteration and optimization.

[0063] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An edge computing-based UV adhesive viscosity real-time regulation system, characterized in that, The application relates to a viscosity control system for UV glue, which comprises a viscosity sensing unit, an edge computing node, an execution unit and a cloud platform arranged at a remote end. The viscosity sensing unit is used for collecting viscosity-related parameters of the UV glue in real time and sending the parameters to the edge computing node. The viscosity-related parameters of the UV glue include temperature, shear rate, shear stress, solid content, pre-curing degree, ambient humidity and stirring speed. The edge computing node is internally provided with a control algorithm module, is used for receiving the viscosity-related parameters, and performs real-time processing and calculation on the viscosity-related parameters to obtain a current viscosity value. Based on a comparison result of the current viscosity value and a preset target viscosity threshold value, the control algorithm module generates real-time control instructions for regulating the viscosity, and then the instructions are sent to the execution unit. The control algorithm module of the edge computing node is a self-adaptive fuzzy PID algorithm. The execution logic of the algorithm comprises the following steps: The deviation and the change rate of the deviation of the current viscosity value and the target viscosity threshold value are calculated in real time. The PID core parameters are dynamically adjusted based on a preset fuzzy rule base through a fuzzy inference module. A disturbance compensation term is introduced to correct the final control instruction output. The execution unit is used for receiving and executing the real-time control instructions for regulating the viscosity. The execution unit comprises a component regulating subunit and an environment regulating subunit. When the absolute value of the viscosity deviation is greater than 10%, the component regulating subunit is preferentially started. 2.The UV viscosity real-time regulation system based on edge computing according to claim 1, wherein: When the absolute value of the viscosity deviation is less than 10%, the environment regulating subunit is preferentially started. 3.The UV viscosity real-time regulation system based on edge computing according to claim 1, wherein: The component regulating subunit adopts a servo-driven micro-injection pump to accurately inject a diluent or a thickener into the UV glue stock solution. An electromagnetic flow sensor is arranged at the outlet of the injection pump to feed back the actual injection amount in real time. The environment regulating subunit adopts a temperature-humidity double closed-loop controller. Temperature control is realized through a semiconductor refrigeration / heating sheet. Humidity control is realized through a dehumidification module and a humidification module. The cloud platform is used for receiving and storing the data uploaded by the edge computing node, performing remote monitoring and parameter setting, analyzing historical data, nonlinearly optimizing the parameters of the control algorithm module in the edge computing node, and sending the optimized parameter model to the edge computing node to update the control algorithm module. The cloud platform is internally provided with a predictive maintenance module. The viscosity sensing unit comprises a non-contact ultrasonic sensing module, a temperature sensing module, an ambient humidity sensing module, a shear rate sensing module, a shear stress sensing module, a near-infrared solid content sensing module, a UV light transmittance sensing module, a stirring speed sensing module and a self-cleaning assembly. The real-time control instructions for regulating the viscosity include component adjustment instructions, environment parameter adjustment instructions and regulation coordination instructions. The component adjustment instruction includes diluent injection instruction and thickener injection instruction, and the instruction parameters include injection amount, injection rate and injection time length; when the UV glue viscosity is higher than the target threshold, the diluent injection instruction is issued; when the viscosity is lower than the target threshold, the thickener injection instruction is issued; The environmental parameter adjustment instruction includes temperature regulation instruction and humidity regulation instruction, and the temperature regulation instruction parameters are target temperature value and temperature change rate; the humidity regulation instruction parameters are target humidity value, and the humidity adjustment lags behind the temperature adjustment by 0.3-0.5s; The regulation coordination instruction includes pause, restart instruction and weight distribution instruction. 4.The UV viscosity real-time regulation system based on edge computing according to claim 1, wherein: The non-linear optimization of the parameters of the control algorithm module in the edge computing node adopts a federated learning framework.

Images