Glass wine bottle direct heating alumite quality control method based on Internet of Things

Abstract

The invention discloses a glass wine bottle direct heating alumite quality control method based on the Internet of Things, and relates to the technical field of quality control, and the method comprises the steps: synchronously collecting real-time temperature field data in a contact area of an alumite foil and a heating roller, and collecting visual image data at an online detection station after a wine bottle is separated from hot pressing; processing the temperature field data to obtain a first quality result representing heating uniformity, and processing the visual image data to obtain a second quality result representing appearance qualification; inputting the two into a rule-based state decision maker for fusion analysis, and outputting a comprehensive judgment result and a control instruction; according to the system and the method, multi-dimensional data fusion analysis and closed-loop control of the technological process and the product quality are realized, a defect source is accurately positioned, the conversion from post-event detection to beforehand early warning is realized, and the system and the method have the advantages that the system and the method are simple and convenient to operate, and the production efficiency is improved. And the accuracy and the real-time performance of quality control are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quality control, more particularly, the present application relates to a glass wine bottle direct aluminum quality control method based on the Internet of Things. BACKGROUND

[0002] In the direct aluminum production process of glass wine bottles, the quality of hot stamping directly affects the appearance level and brand value of the product. At present, the quality control methods commonly used in the industry mainly include two kinds: one is to rely on manual periodic sampling inspection, and the operating personnel assess the surface quality of the hot stamped wine bottle through visual observation or simple tools; the second is to set an independent visual detection system at the end of the production line, collect images of the aluminum pattern, and use image processing algorithms to identify whether there are defects such as incomplete adhesion, bubbles, scratches, etc. However, in actual use, it still has some shortcomings, such as: on the one hand, the existing technology only focuses on the surface quality of the final product, ignoring the key process parameters in the hot stamping process, such as heating uniformity, process monitoring and quality detection are independent of each other, forming a data island, which makes it impossible to establish a correlation between process parameters and product quality. When defects occur, it is difficult to accurately trace the cause of the defects to uneven heating, material defects or abnormal pressure, making quality control lack of pertinence.

[0003] On the other hand, the existing method is essentially a "post-detection", that is, identification and rejection are carried out after the defect has been formed. This mode cannot provide real-time early warning and advance intervention for process abnormalities that may cause defects during the production process, resulting in waste of materials and time, and also easily leading to batch quality accidents. At the same time, the investigation of abnormalities and the adjustment of parameters are highly dependent on manual experience, and the response is slow, with low automation and intelligence. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a glass wine bottle direct aluminum quality control method based on the Internet of Things, which solves the problems of single quality judgment dimension and lagging and passive quality control in the background art through the following scheme.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a glass wine bottle direct aluminum quality control method based on the Internet of Things, comprising:

[0006] S1: synchronously collecting data at two key positions with process time sequence correlation, wherein the first position is the contact area of the aluminum foil and the heating roller, used to collect real-time temperature field data in the hot pressing stage; the second position is the online quality detection station after the wine bottle is separated from the hot pressing, used to collect visual image data of the aluminum pattern after forming;

[0007] S2: processing the temperature field data of the first position to obtain a first quality result for characterizing heating uniformity, and processing the visual image data of the second position to obtain a second quality result for characterizing appearance qualification;

[0008] S3: inputting the first quality result and the second quality result into a rule-based state decision maker for fusion analysis, outputting a comprehensive determination result, and generating a control instruction based on the comprehensive determination result;

[0009] S4: driving the audible and visual alarm, the human-machine interface or the production line control system to perform corresponding quality control actions according to the control instruction.

[0010] Preferably, the real-time temperature field data includes a sequence of continuous temperature readings with spatial coordinate information and time stamp collected by a distributed temperature sensor array deployed circumferentially and axially around the contact area of the heating roller, for reconstructing a three-dimensional temperature distribution map of the heating roller surface during the hot pressing stage.

[0011] Preferably, the visual image data includes one or more high-resolution RGB images of the complete electrochemical aluminum pattern from one or more viewing angles collected under illumination of a specific angle light source.

[0012] The specific angle light source is a low angle light source and a coaxial light source, for highlighting the profile, adhesion uniformity and surface defects of the electrochemical aluminum pattern through shadow effect or elimination of reflection.

[0013] Preferably, processing the temperature field data of the first position includes:

[0014] Calculating the standard deviation and range of the temperature field data, and comparing the standard deviation and range with corresponding preset threshold values respectively, to quantify the heating uniformity, and outputting the first quality result as a binary determination state of uniform or non-uniform.

[0015] Preferably, processing the visual image data of the second position includes:

[0016] Defect recognition and classification based on a convolutional neural network model on the preprocessed visual image to identify partial adhesion, bubble or scratch defects, and outputting the second quality result as a binary determination state of no defect or defect.

[0017] Preferably, the rule-based state decision maker has a decision logic configured to perform the following mapping:

[0018] If the first quality result is uniform and the second quality result is no defect, output a process excellent comprehensive determination result;

[0019] If the first quality result is non-uniform and the second quality result is no defect, output a process critical comprehensive determination result.

[0020] If the first quality result is uniform and the second quality result is defective, output a comprehensive determination result of material or pressure abnormality;

[0021] If the first quality result is non-uniform and the second quality result is defective, output a comprehensive determination result of process failure.

[0022] Preferably, the control instruction is generated according to the comprehensive determination result, wherein:

[0023] The process optimal result corresponds to an instruction to maintain the current process parameters; the process critical result corresponds to an instruction to generate a warning log; the material or pressure abnormality result corresponds to an alarm instruction to trigger inspection of the material quality and pressure parameters; and the process failure result corresponds to an instruction to trigger automatic shutdown.

[0024] Preferably, the quality control action comprises:

[0025] One or more of the following: displaying diagnostic information on a human-machine interface, activating an audible and visual alarm, uploading a warning log to a cloud platform, automatically adjusting heating roller temperature or mechanical pressure parameters, and executing production line emergency stop.

[0026] Technical effects and advantages of the present application:

[0027] 1. The present application synchronously collects temperature field data and visual image data at key positions associated with two process time sequences, correlates and analyzes the process parameter heating uniformity in the production process with the appearance defects of the final product, overcomes the limitations of traditional methods of single or post-detection, and realizes full-chain quality monitoring from process to result.

[0028] 2. The present application performs fusion analysis of the heating uniformity and appearance qualification results by a rule-based state decision maker, can output comprehensive determination results not limited to "qualified / unqualified", accurately locates the root cause of quality problems, and distinguishes different reasons such as process, material or equipment pressure, thereby providing a clear direction for rapid troubleshooting and problem solving.

[0029] 3. The present application generates and executes corresponding control instructions according to the comprehensive determination result, from maintaining parameters, warning, alarm to automatic shutdown, realizes a closed-loop automatic quality control, significantly reduces manual intervention, reduces the risk of human error, and can respond quickly to abnormal working conditions, effectively preventing the occurrence of batch quality accidents.

[0030] 4.The method can discover abnormal trends of process parameters in time before visual defects appear, and issue early warning, so as to move the quality control forward, realize the change from post-repair to pre-prevention, reduce the scrap rate and improve production efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 It is a whole quality control flowchart of the application;

[0032] Figure 2 It is a temperature field data processing flowchart of the application;

[0033] Figure 3 It is a visual image data processing flowchart of the application;

[0034] Figure 4 It is a state decision maker decision logic flowchart of the application;

[0035] Figure 5 It is a quality control action execution flowchart of the application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the application.

[0037] As shown in the accompanying drawings, a glass wine bottle direct lamination electrochemical aluminum quality control method based on the Internet of Things comprises: Figure 1 S1: synchronously collecting data at two key positions with process time sequence association, wherein the first position is the contact area of electrochemical aluminum foil and heating roller, and is used for collecting real-time temperature field data in the hot pressing stage; the second position is an online quality detection station after the wine bottle is separated from the hot pressing, and is used for collecting visual image data of the electrochemical aluminum pattern after forming;

[0038]

[0039] ​It needs to be further explained that the application constructs a "cloud-edge-end" collaborative Internet of Things architecture to realize whole-chain quality control. The end side is composed of a distributed temperature sensor array, an industrial camera, an audible and visual alarm, a human-machine interface and a PLC controller deployed on the production line, responsible for raw data collection and final execution of control instructions. The edge side is composed of an industrial control computer beside the production line as an edge computing node, responsible for running temperature field processing, visual image processing, state decision maker and other core algorithms, realizing low-latency real-time control and response. The cloud platform receives the processing results, alarm logs and key image data uploaded by the edge side through the Internet of Things gateway installed on the production line, and performs big data storage, analysis and model optimization.

[0040] It needs to be specifically explained that the real-time temperature field data of the first position is composed of a distributed sensor array of PT100 platinum resistance temperature sensors, with a sensor accuracy of ±0.5℃ and a response time of ≤1s. Along the heating roller, 1 sensor is deployed every 30° of the circumference of the roller body with a length of 1.2m and a diameter of 300mm, a total of 12 sensors; 1 sensor is deployed every 10cm along the axial direction of the heating roller, a total of 12 sensors, forming a three-dimensional sensing network of 144 acquisition points. All sensors are connected to a data acquisition card through an RS485 bus. The sampling frequency is set to 1Hz, and each acquisition point generates 1 set of data containing "spatial coordinates-time stamp-temperature value" each time. The spatial coordinates are predefined by the sensor deployment position, the circumferential coordinate is an angle value, and the axial coordinate is the distance from the left end of the heating roller, both in mm. The time stamp is accurate to the millisecond level, and the format is "YYYY-MM-DD HH:MM:SS.ms".

[0041] It needs to be further explained that the collected continuous temperature reading sequence is used to reconstruct the hot pressing stage, and the hot pressing time is set to 2s. The three-dimensional temperature distribution map of the heating roller surface, for example, the temperature reading of the heating roller at 30° of the circumference and 500mm of the axial position at a certain time is 118.5℃. This data will be used as the basis for subsequent heating uniformity judgment. This data acquisition method meets the key position data acquisition requirements of process timing association with temperature field data usage.

[0042] It needs to be specifically pointed out that the visual image data of the second position is obtained by arranging an inspection station 1.5 m downstream of the hot pressing station, ensuring that the temperature of the wine bottle is reduced to below 50°C after the wine bottle is separated from the hot pressing, avoiding the influence of high temperature on image acquisition, deploying two industrial cameras with a resolution of 2048*1536 pixels and a frame rate of 5 fps, respectively collecting the front and 45° side view angles of the wine bottle anodized aluminum pattern, ensuring that the complete anodized aluminum hot stamping area is covered, which is usually a circular area with a diameter of 80 mm; low-angle light source with a 20° angle with the detection table, power 30W, wavelength 550nm, coaxial light source, power 50W, wavelength 550nm combination lighting; the low-angle light source highlights the edge defects of the anodized aluminum pattern through the shadow effect, and the coaxial light source clearly presents the internal defects such as bubbles and scratches by eliminating metal reflection, and this light source combination method can effectively highlight the features of the anodized aluminum pattern to meet the needs of visual inspection.

[0043] It needs to be further explained that the camera collects high-resolution RGB images, and each wine bottle corresponds to 2 groups of images of front view + side view, the image format is BMP, the file size is about 9MB per image, and the image naming rule is "wine bottle number-view angle-time stamp.bmp", such as "JP20240501001-front-20240501100000.123.bmp", after collection, it is transmitted to the image processing server in real time through Ethernet, and the image processing server is configured with a multi-core processor and 32GB memory.

[0044] S2: processing the temperature field data of the first position to obtain a first quality result for characterizing heating uniformity, and processing the visual image data of the second position to obtain a second quality result for characterizing appearance qualification;

[0045] As shown in Figure 2 , processing the temperature field data of the first position includes:

[0046] The standard deviation and range of the temperature field data are calculated, and the standard deviation and range are compared with the corresponding preset threshold, respectively, to quantify the heating uniformity, and the first quality result is output as a binary judgment state of uniform or non-uniform.

[0047] It needs to be further explained that processing the temperature field data of the first position first eliminates outliers, adopts the 3σ criterion, that is, eliminates temperature readings exceeding the range of "average value ± 3*standard deviation", if the proportion of outliers in a single group of collected data is >5%, the group of data is re-collected, and the temperature average value x, standard deviation σ and range R of all valid collection points are calculated, that is, the difference between the maximum value and the minimum value; the calculation formula of the standard deviation is where x i is a single valid temperature reading, The average value of all valid readings, n is the number of valid readings, n = 144 under normal circumstances; the calculation formula of range is R = Max(x i )-Min(x i ), wherein Max(x i ) is the maximum value in the valid readings, and Min(x i ) is the minimum value in the valid readings.

[0048] It should be further explained that the temperature field preset threshold value determines the qualified threshold value of heating uniformity through orthogonal test. Taking "temperature 110℃, 120℃, 130℃, pressure 0.2MPa, 0.3MPa, 0.4MPa, vehicle speed 0.4m / s, 0.5m / s, 0.6m / s" as test factors, 3 levels are set for each factor, and a total of 27 tests are selected. The maximum value 2℃ of the temperature field standard deviation and the maximum value 5℃ of the range of all hot stamping quality qualified groups are selected as the judgment threshold value, that is, the standard deviation is ≤2℃ and the range is ≤5℃. If the processed temperature field standard deviation is ≤2℃ and the range is ≤5℃, the binary judgment state of "uniform" is output. If any index exceeds the threshold value, such as standard deviation 2.5℃, range 4℃, or standard deviation 1.8℃, range 5.5℃, the binary judgment state of "non-uniform" is output. This processing method meets the quantitative requirements of temperature field data processing according to the binary judgment rule.

[0049] As shown in Figure 3 , the processing of the visual image data at the second position includes:

[0050] Based on the convolutional neural network model, the preprocessed visual image is recognized and classified for defects, to identify incomplete attachment, bubbles or scratch defects, and output the second quality result as a binary judgment state of no defect or defect.

[0051] It should be further explained that the image preprocessing first uses Gaussian filtering to denoise the collected RGB image, with a convolution kernel size of 3*3 and a standard deviation of 0.8. Then, the electrochemical aluminum pattern area is separated from the wine bottle glass background using adaptive threshold segmentation, with a threshold range of 120-180, and the segmented area is grayed.

[0052] A lightweight convolutional neural network model, such as MobileNetV3-Small, is employed for deep feature extraction and defect identification of preprocessed images. This network model is pre-trained and optimized on the server side using thousands of labeled defect sample images, including incomplete attachment, bubbles, scratches, and acceptable samples. The preprocessed images are then input into the CNN model, which directly outputs the image's defect classification results, such as "no defect," "incomplete attachment," "bubble," and "scratch," along with their corresponding confidence scores. To improve the system's interpretability and robustness, the following geometric and optical features are computed in parallel to assist in validating the deep learning model's output:

[0053] Contour integrity is determined by matching the processed pattern contour with the standard anodized aluminum pattern contour and calculating the degree of overlap. The overlap ratio is ≥95% to be considered qualified. The standard electroplated aluminum pattern is a CAD file pre-stored on the server, and the contour error is allowed to be ±0.1mm.

[0054] Brightness consistency is achieved by calculating the average brightness of the pattern area. With the standard deviation of brightness σ y , σ y A value ≤10 is considered acceptable, with a brightness range of 0-255. The standard deviation of brightness is expressed as follows: Among them, y j y is the brightness value of a single pixel within the pattern area, ranging from 0 to 255; y is the average brightness value of all pixels in the pattern area; and m is the total number of pixels in the pattern area.

[0055] It should be further explained that if the CNN model identifies any defect (confidence level higher than a preset threshold, such as 90%), or any auxiliary feature is unqualified, the second quality result output is "defective"; otherwise, it is "no defect".

[0056] S3: Input the first quality result and the second quality result into a rule-based state decision-maker for fusion analysis, output a comprehensive judgment result, and generate control instructions based on the comprehensive judgment result;

[0057] like Figure 4 As shown, the core of the rule-based state decision-maker lies in an adaptively updatable rule base. The initial rules of this rule base are set based on the experience of process experts and include the following mapping logic:

[0058] If the first quality result is uniform and the second quality result is defect-free, then the output is a comprehensive judgment result indicating that the process is excellent.

[0059] If the first quality result is non-uniform and the second quality result is defect-free, then output the comprehensive judgment result of the process criticality.

[0060] If the first quality result is uniform and the second quality result is defective, output a comprehensive determination result of material or pressure abnormality;

[0061] If the first quality result is non-uniform and the second quality result is defective, output a comprehensive determination result of process failure.

[0062] It should be further explained that, in order to further improve the accuracy and adaptability of the decision, the present application introduces a rule optimization mechanism based on historical data: the industrial control computer periodically packages and uploads the collected temperature field data, visual image data, artificially confirmed defect sources, and final comprehensive determination results to the cloud platform, and the data analysis service of the cloud platform uses these historical data to mine and verify the initial rules through decision tree algorithm, thereby optimizing the decision logic. For example, the platform may find that when the temperature standard deviation is between 2.0°C and 2.5°C and the range is between 5°C and 6°C, if the vehicle speed is less than 0.45 m / s, there is still an 85% probability of producing qualified products. At this time, this combination condition can be added as a "process observation" rule, rather than simply determining as "process critical", and the updated rule library can be periodically or according to instructions issued to the state decision maker on the edge side of the production line, thereby realizing the continuous evolution of decision-making ability.

[0063] It should be further explained that when the comprehensive determination result is process optimal, the heating uniformity in the hot pressing stage meets the requirements, and there is no appearance defect in the hot stamping pattern, so the corresponding control instruction is to maintain the current process parameters;

[0064] When the comprehensive determination result is process critical, the heating uniformity exceeds the threshold value, but does not cause appearance defects, which may be due to the current slow vehicle speed, sufficient hot pressing time to compensate for the uneven temperature, so the corresponding control instruction is to generate a warning log and upload it to the cloud platform;

[0065] When the comprehensive determination result is material or pressure abnormality, the heating uniformity is qualified, and the appearance defect is not related to heating, and it is highly probable that it is caused by the anodized aluminum foil material, such as uneven base film thickness, insufficient adhesive force of the glue layer, or abnormal hot pressing pressure (too high to cause pattern rupture, too low to cause poor adhesion), so the corresponding control instruction is to trigger an audible and light alarm and prompt to check the material and pressure;

[0066] When the comprehensive determination result is process failure, the heating uniformity and appearance quality are both unqualified, and the process as a whole fails, so the corresponding control instruction is to trigger the automatic shutdown of the production line and record the fault data.

[0067] It needs to be further explained that the control instruction is written in JSON format, containing four fields of "instruction type, execution object, parameter value, timestamp", for example, the instruction corresponding to "process optimization" is: {"instruction type": "parameter maintenance", "execution object": "heating roller temperature controller, pressure regulating valve", "parameter value": {"heating temperature": 120℃, "pressure": 0.3MPa}, "timestamp": "20240501100001.234"}; The instruction corresponding to "process failure" is: {"instruction type": "emergency stop", "execution object": "production line main motor, hot pressing mechanism", "parameter value": {"stop delay": 0s}, "timestamp": "20240501100002.567"}; The instruction is transmitted to each execution mechanism through the industrial bus, and the instruction generation logic meets the generation requirements of the corresponding results.

[0068] S4: According to the control instruction, drive the sound and light alarm, human-computer interface or production line control system to execute corresponding quality control action.

[0069] As Figure 5 shown, the quality control action includes:

[0070] Displaying diagnostic information on the human-computer interface, activating the sound and light alarm, uploading alarm logs to the cloud platform, automatically adjusting the heating roller temperature or mechanical pressure parameters, and executing one or more of the production line emergency stop.

[0071] It needs to be further explained that when the control instruction is to maintain the current process parameters, the heating roller temperature controller controls the actual temperature fluctuation within ±1℃ by PID algorithm, maintains the set temperature such as 120℃; The pressure regulating valve maintains the set pressure, such as 0.3MPa, and the pressure fluctuation is controlled within ±0.02MPa; The human-computer interface displays "current process is stable, parameters are normal", without alarm prompt;

[0072] When the control instruction is to generate a warning log and upload it to the cloud platform, the industrial control computer automatically generates a warning log, and the log content includes "timestamp, first quality result data (temperature average value, standard deviation, range), second quality result data (defect type, characteristic parameter), judgment result", for example, "20240501100003.789, temperature x=119℃, σ=2.2℃, R=4.7℃ (non-uniform), visual defect-free, judge process critical"; The log is uploaded to the cloud platform server through HTTPS protocol, the cloud platform triggers "process critical" message push, and pushes to the mobile application of the production line administrator, and at the same time marks the data of this period as "to be concerned" in the platform background;

[0073] When the control instruction is to trigger the sound and light alarm and prompt to check the material and pressure, the sound and light alarm installed at a height of 1.5 m beside the production line emits a red constant light with a brightness of ≥500 cd / m after receiving the instruction 2 with a buzzer sound of 1 kHz and a volume of ≥85 dB, and the alarm continues until manual confirmation; the man-machine interface displays "material or pressure abnormality, please check: whether the material is a PET-based qualified product, and whether the thickness is 12 μm ± 1 μm; the hot-pressing pressure parameter: the current pressure is 0.25 MPa, and the standard value is 0.3 MPa ± 0.02 MPa, and the historical pressure curve (the last 10 minutes of data) and the material batch information are displayed simultaneously;

[0074] When the corresponding control instruction is to trigger the automatic shutdown of the production line and record the fault data, the main motor controller of the production line receives the instruction and immediately cuts off the power supply of the main motor, realizes emergency stop, and the shutdown time is ≤0.5 s, and at the same time, the heating power supply and the pressure gas source of the hot-pressing mechanism are cut off; the industrial control computer records the fault data, including "shutdown time, 10 groups of temperature field data before the fault, 5 groups of visual images before the fault, judgment result", and automatically generates a fault report in PDF format, which is stored in the local and cloud platform; the man-machine interface displays "process failure, the production line has been shut down, please check the heating roller temperature control fault and visual detection abnormality", and lists the possible causes of the fault, such as temperature sensor fault, camera lens contamination, and heating tube damage.

[0075] It should be further pointed out that after all the actuators execute the actions, they need to feedback the "action execution status" to the industrial control computer: success / failure, for example, the sound and light alarm feedbacks "the alarm has been activated", and the main motor controller feedbacks "it has been shut down"; if the feedback is "failure", such as the alarm fault is not activated, the industrial control computer triggers the standby control action, such as the man-machine interface pops up a pop-up window "alarm fault, please manually patrol", to ensure that there is no omission in control, in addition, when the "material or pressure abnormality" instruction is executed, the parameters are adjusted manually, such as replacing the qualified electrochemical aluminum foil and adjusting the pressure to 0.3 MPa, the system can execute the S1-S4 process again by clicking "parameter reset" on the man-machine interface to verify the adjustment effect.

[0076] Secondly, in the drawings of the disclosed embodiments, only the structures related to the disclosed embodiments are involved, and other structures can be referred to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other;

[0077] Finally, the above-mentioned is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for quality control of direct-heating electroplated aluminum foil on glass wine bottles based on the Internet of Things, characterized in that, include: S1: Data is collected synchronously at two locations with process timing correlation, namely, the first location is the contact area between the electroplated aluminum foil and the heating roller, used to collect real-time temperature field data during the hot pressing stage; the second location is the online quality inspection station after the bottle is removed from the hot pressing stage, used to collect visual image data of the electroplated aluminum pattern after molding. S2: Process the temperature field data at the first location to obtain a first quality result characterizing the heating uniformity; process the visual image data at the second location to obtain a second quality result characterizing the appearance conformity. S3: Input the first quality result and the second quality result into a rule-based state decision-maker for fusion analysis, output a comprehensive judgment result, and generate control instructions based on the comprehensive judgment result; S4: According to the control command, drive the audible and visual alarm, human-machine interface or production line control system to perform corresponding quality control actions.

2. The method for quality control of direct-heating electroplated aluminum on glass wine bottles based on the Internet of Things as described in claim 1, characterized in that: The real-time temperature field data includes a continuous temperature reading sequence with spatial coordinate information and timestamps, collected by a distributed temperature sensor array deployed in the circumferential and axial directions of the heating roller contact area, which is used to reconstruct a three-dimensional temperature distribution map of the heating roller surface during the hot pressing stage.

3. The method for quality control of direct-heating electroplated aluminum on glass wine bottles based on the Internet of Things as described in claim 1, characterized in that: The visual image data includes high-resolution RGB images from one or more perspectives that cover the entire electroplated aluminum pattern, acquired under illumination from a light source at a specific angle. The light source at the specific angle is a low-angle light source and a coaxial light source, used to highlight the outline, adhesion uniformity and surface defects of the electroplated aluminum pattern through the shadow effect or elimination of reflection.

4. The method for quality control of direct-heating electroplated aluminum on glass wine bottles based on the Internet of Things as described in claim 1, characterized in that: The processing of the temperature field data at the first location includes: The standard deviation and range of the temperature field data are calculated, and the standard deviation and range are compared with the corresponding preset thresholds to quantify the heating uniformity. The first quality result is output as a binary judgment state of uniformity or non-uniformity.

5. The method for quality control of direct-heating electroplated aluminum on glass wine bottles based on the Internet of Things as described in claim 1, characterized in that: The processing of the visual image data at the second location includes: The preprocessed visual images are used to identify and classify defects based on a convolutional neural network model, in order to identify defects such as incomplete attachment, bubbles or scratches, and the second quality result is output as a binary judgment state of no defect or defective.

6. The method for quality control of direct-heating electroplated aluminum on glass wine bottles based on the Internet of Things as described in claim 1, characterized in that: The core of the rule-based state decision-maker lies in an adaptively updatable rule base. The initial rules of this rule base are set based on the experience of process experts and include the following mapping logic: If the first quality result is uniform and the second quality result is defect-free, then the output is a comprehensive judgment result indicating that the process is excellent. If the first quality result is non-uniform and the second quality result is defect-free, then output the comprehensive judgment result of the process criticality. If the first quality result is uniform and the second quality result is defective, then output a comprehensive judgment result of material or pressure abnormality; If the first quality result is uneven and the second quality result is defective, then the output is a comprehensive judgment result of process failure.

7. The method for quality control of direct-heating electroplated aluminum on glass wine bottles based on the Internet of Things as described in claim 1, characterized in that: The control command is generated based on the comprehensive judgment result, wherein: Excellent process results correspond to instructions to maintain the current process parameters; critical process results correspond to instructions to generate early warning logs; abnormal material or pressure results correspond to alarm instructions to trigger checks on material and pressure parameters; and process failure results correspond to instructions to trigger automatic shutdown.

8. The method for quality control of direct-heating electroplated aluminum on glass wine bottles based on the Internet of Things as described in claim 1, characterized in that: The quality control actions include: Display diagnostic information on the human-machine interface, activate the audible and visual alarm, upload alarm logs to the cloud platform, automatically adjust the temperature or mechanical pressure parameters of the heating roller, and execute one or more of the following functions for emergency stop of the production line:

Images