Intelligent control dispensing method and system based on automatic optical detection

Abstract

The invention relates to the field of automatic control, and discloses an intelligent control dispensing method and system based on automatic optical detection, and the method comprises the steps: obtaining multi-source real-time data of a dispensing production line, carrying out the feature recognition processing through a preset feature extraction model, and obtaining the dynamically updated production state information; then interference filtering processing is carried out to obtain an effective data set, when it is detected that quality fluctuation exceeds a preset range, an alarm is triggered, deviation identification information is generated, self-adaptive parameter optimization processing is carried out in combination with equipment characteristics and historical operation records, and a control instruction is generated; a control instruction is issued to an execution unit through a double-channel transmission mechanism, transmission delay is monitored in real time, and channels are automatically switched; when the output deviation exceeds the tolerance range, a compensation instruction is generated, parameters of the feature extraction model are updated based on the novel quality fluctuation features, and the dispensing control process is iteratively optimized through closed-loop control. According to the method, the dispensing precision and the production efficiency can be remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automation control, and in particular to an intelligent control dispensing method and system based on automated optical inspection. BACKGROUND

[0002] Currently, in the field of precision dispensing processing, automated optical inspection (AOI) and control technology collect production line data through preset algorithms and generate control instructions.

[0003] In the prior art, such as patent document CN115591742B entitled "Dispensing machine automatic control method and system for dispensing quality recognition" and patent document CN116152245B entitled "Glue line defect detection system based on artificial intelligence", both belong to this type of technology. The system determines whether a quality problem has occurred through optical detection and issues a unified control instruction to adjust the production process parameters. However, the above-mentioned technologies have deficiencies and cannot meet the current production control requirements. On the one hand, the control instruction is not optimized in combination with the equipment characteristics and historical operation records, making it difficult to meet individual needs in the face of fine operations in different positions and different environments, exacerbating the lack of operation accuracy. On the other hand, the feature extraction model cannot identify new quality fluctuation characteristics, and lacks a closed-loop iterative optimization mechanism, leading to continuous deterioration of quality fluctuations under changing production conditions. Moreover, they all transmit data in real time through the main communication channel, often resulting in transmission delays. In high-speed automated production processes, this can cause execution delays, and production control cannot be executed and adjusted in a timely manner, affecting the yield and production efficiency of the entire production line.

[0004] The existing technology has the problems of lack of linkage optimization in detection and control, signal transmission delay affecting production line efficiency, and difficulty in adapting to complex and variable production conditions. SUMMARY

[0005] The present application provides an intelligent control dispensing method and system based on automated optical inspection to solve the problems of lack of linkage optimization in detection and control, signal transmission delay affecting production line efficiency, and difficulty in adapting to complex and variable production conditions in the prior art.

[0006] In a first aspect, to solve the above technical problems, the present application provides an intelligent control dispensing method based on automated optical inspection, comprising: acquiring multi-source real-time data of a dispensing production line, wherein the multi-source real-time data includes dispensing position coordinates, glue amount distribution data, and glue uniformity parameters collected by an optical sensor, and dispensing head temperature, air cylinder pressure, and equipment vibration frequency collected by a physical sensor; According to the multi-source real-time data, feature recognition processing is performed through a preset feature extraction model to obtain dynamically updated production state information; According to the production state information, interference filtering processing is performed to obtain an effective data set, and when it is detected that quality fluctuation exceeds a preset range, an alarm is triggered and deviation identification information is generated; Based on the deviation identification information, adaptive parameter optimization processing is performed in combination with device characteristics and historical operation records to generate control instructions; The control instructions are issued to an execution unit through a dual-channel transmission mechanism, transmission delay is monitored in real time, and the channel is automatically switched; According to execution feedback data, device parameters are dynamically regulated and controlled, when output deviation exceeds a tolerance range, compensation instructions are generated, and parameters of the feature extraction model are updated based on new quality fluctuation characteristics, and the dispensing control process is iteratively optimized through closed-loop control.

[0007] Preferably, the multi-source real-time data is processed through a preset feature extraction model to obtain dynamically updated production state information, including: The multi-source real-time data is subjected to structured storage processing, and a standardized data set is generated according to device number and time stamp; A decision tree algorithm is used to dynamically extract features from the standardized data set, and when the feature value exceeds a preset parameter threshold, it is marked as an abnormal feature; Based on the marked results, real-time abnormal state detection is performed, and if an abnormal state is detected, production state updating is triggered; The updated abnormal features are associated and matched with historical operation records to generate dynamic production state information containing device state labels.

[0008] Preferably, the interference filtering processing according to the production state information to obtain an effective data set includes: The heterogeneous data in the production state information is subjected to format standardization, and a spatiotemporally consistent standard data set is generated through time stamp alignment and unit unification; Based on the spatiotemporally consistent standard data set, mean filtering processing is performed, and if the deviation of a data point from the mean value of a sliding window exceeds a preset noise threshold, it is marked as a noise point and removed to generate a denoised data set; A decision tree classifier is used to perform outlier determination processing on the denoised data set, and when the feature matching degree is lower than a preset confidence threshold, it is determined as an outlier and removed to generate an effective data set.

[0009] Preferably, when it is detected that quality fluctuation exceeds a preset range, an alarm is triggered and deviation identification information is generated, including: Based on the effective data set, dynamic change characteristics are extracted by a configurable sliding window, the size of which is self-adaptively adjusted according to the device type and the production stage. The feature change rate of the dynamic change characteristics is calculated. If the feature change rate exceeds a primary threshold value, a primary alarm is triggered and a deviation identifier containing the device number is generated. If the feature change rate exceeds a secondary threshold value, a serious alarm is triggered and a deviation identifier containing the spatial position index is generated.

[0010] Preferably, based on the deviation identifier information, adaptive parameter optimization processing is performed in combination with the device characteristics and the historical operation records to generate control instructions, including: The abnormal type and the device position in the deviation identifier information are analyzed to match a pre-stored operation instruction library to generate a preliminary control parameter set. The preliminary control parameter set is subjected to amplitude screening according to the dynamic response curve and the dispensing accuracy parameter in the device characteristics. In combination with similar working condition optimization data in the historical operation records, the screened parameters are subjected to secondary fine-tuning by a logistic regression model. When the fine-tuned control parameters meet the individualized dispensing control requirements, individualized control instructions containing the device position identifier are generated.

[0011] Preferably, the control instructions are issued to the execution unit through a dual-channel transmission mechanism, the transmission delay is monitored in real time, and the channel is automatically switched, including: The control instructions are subjected to data packet processing and are attached with the device position identifier and the integrity check code to obtain data packets. The data packets are transmitted in real time through a main communication channel, and the transmission delay and the packet loss rate are continuously monitored. When the transmission delay exceeds a dynamic response threshold value or the packet loss rate continuously exceeds the standard, the transmission of the remaining control instructions is completed by automatically switching to a backup communication channel.

[0012] Preferably, the device parameters are dynamically regulated according to the execution feedback data, and compensation instructions are generated when the output deviation exceeds the tolerance range, including: The device operation state data and the dispensing quality parameters fed back by the execution unit are collected, formatted, and processed to generate a standardized feedback data set. The standardized feedback data set is subjected to multi-dimensional deviation analysis with a pre-set target parameter, and when the dispensing position accuracy or the glue amount uniformity exceeds the dynamic tolerance range, the deviation type and the device position are marked. Based on the marking result, a mechanical response model in the device characteristics library is called, and initial compensation parameters are generated in combination with the historical compensation records. The initial compensation parameters are fine-tuned and verified using an adaptive control algorithm. When the fine-tuned parameters meet the stability requirements of the dispensing process, a compensation command containing the equipment position identifier is output.

[0013] Preferably, the step of updating the parameters of the feature extraction model based on novel quality fluctuation characteristics and iteratively optimizing the dispensing control process through closed-loop control includes: The dispensing quality fluctuation feedback data is obtained from the sensors of the production equipment, and novel fluctuation characteristics are extracted through time series analysis. When the novel fluctuation feature exceeds the preset recognition threshold, the weight parameters of the feature extraction model are adjusted using the gradient descent algorithm. Based on the updated feature extraction model, a new set of operation instructions is generated using the logistic regression method. The new set of operating instructions is sent to the dispensing equipment to adjust the operating parameters, and the adjusted dispensing quality data is collected in real time. When the adjusted dispensing quality parameters consistently meet the preset stability requirements, the closed-loop optimization is considered complete.

[0014] Secondly, the present invention provides an intelligent control dispensing system based on automatic optical detection, comprising: The data acquisition module is used to acquire multi-source real-time data of the dispensing production line. The multi-source real-time data includes dispensing position coordinates, glue volume distribution data and glue uniformity parameters collected by optical sensors, and dispensing head temperature, pneumatic cylinder pressure and equipment vibration frequency collected by physical sensors. The production status determination module is used to perform feature recognition processing based on the multi-source real-time data through a preset feature extraction model to obtain dynamically updated production status information. The alarm and deviation identification generation module is used to perform interference filtering processing based on the production status information to obtain a valid dataset. When a quality fluctuation is detected to exceed a preset range, an alarm is triggered and deviation identification information is generated. The control command generation module is used to generate control commands by performing adaptive parameter optimization processing based on the deviation identification information, combined with equipment characteristics and historical operation records. The communication channel switching module is used to send the control commands to the execution unit through a dual-channel transmission mechanism, monitor the transmission delay in real time, and automatically switch channels. The dispensing control optimization module is used to dynamically adjust equipment parameters based on execution feedback data. When the output deviation exceeds the tolerance range, a compensation command is generated, and the parameters of the feature extraction model are updated based on the new quality fluctuation characteristics. The dispensing control process is iteratively optimized through closed-loop control.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) Obtain multi-source real-time data from the dispensing production line, with optical detection as the main method and other detections, including environmental parameters, as a supplement. This breaks through the limitations of traditional single data sources and provides comprehensive input for subsequent processing. It ensures the completeness of quality fluctuation detection from the source and avoids misjudgment caused by incomplete data coverage.

[0016] (2) By processing the dynamically updated production status information through feature recognition, the original multi-source real-time data is transformed into operable production status labels, which solves the problem that the data cannot be converted into instructions and improves the efficiency of information conversion. Combined with the steps of interference filtering and deviation label generation, the real quality fluctuations and instantaneous interference are effectively distinguished, the problem of frequent production line shutdowns caused by false alarms is overcome, and the false alarm rate is significantly reduced.

[0017] (3) Adaptive optimization combining equipment characteristics and historical records can meet personalized needs and improve the accuracy of dispensing production; and dynamic control and closed-loop iterative optimization can generate compensation instructions based on feedback data and continuously update the feature extraction model to break through the bottleneck of being unable to adapt to information quality fluctuations.

[0018] (4) The dual-channel transmission mechanism automatically switches, with redundant transmission of the main and backup channels. When the delay exceeds the standard, the switching is done at the millisecond level. This can effectively avoid the risk of deviation expansion caused by instruction transmission delay, and ensure that control instructions can reach the execution components in time, thereby realizing timely execution and adjustment of production control and greatly improving production accuracy and efficiency.

[0019] In summary, this invention addresses the problems of lack of linkage optimization in detection and control, signal transmission delay affecting production line efficiency, and difficulty in adapting to complex and changing production conditions by constructing a full-process intelligent control system that integrates data perception, dynamic decision-making, reliable execution, and closed-loop optimization. This results in a significant improvement in dispensing accuracy and production efficiency. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of an intelligent control dispensing method based on automatic optical detection provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent control dispensing system based on automatic optical detection provided in the second embodiment of the present invention. 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] ReferenceFigure 1 The first embodiment of the present invention provides an intelligent control dispensing method based on automatic optical detection, comprising the following steps: S11, acquire multi-source real-time data of the dispensing production line, wherein the multi-source real-time data includes dispensing position coordinates, glue volume distribution data and glue uniformity parameters collected by optical sensors, and dispensing head temperature, pneumatic cylinder pressure and equipment vibration frequency collected by physical sensors. S12, Based on multi-source real-time data, feature recognition processing is performed through a preset feature extraction model to obtain dynamically updated production status information; S13, perform interference filtering based on production status information to obtain a valid dataset, and trigger an alarm and generate deviation identification information when a quality fluctuation is detected to exceed the preset range. S14, based on deviation identification information, combined with equipment characteristics and historical operation records, performs adaptive parameter optimization processing to generate control commands; S15 sends control commands to the execution unit through a dual-channel transmission mechanism, monitors transmission delay in real time, and automatically switches channels; S16 dynamically adjusts equipment parameters based on execution feedback data. When the output deviation exceeds the tolerance range, a compensation command is generated, and the parameters of the feature extraction model are updated based on the new quality fluctuation characteristics. The dispensing control process is iteratively optimized through closed-loop control.

[0023] This embodiment provides an intelligent control dispensing method based on automated optical inspection (AOI). Through multi-source data sensing, dynamic decision-making, reliable execution, and closed-loop optimization, it solves the problems of poor environmental adaptability, response delay, and lack of personalized adjustment in precision dispensing.

[0024] In some embodiments, multi-source real-time data acquisition is achieved by using optical and physical sensors to collaboratively collect data on the size, position coordinates, temperature, pressure, and speed of adhesive dots in the dispensing production line. The raw data is then organized into a structured dataset by timestamp and device number. For example, a thermistor sensor collects temperature data every second, a pressure sensor collects pressure data every 30 seconds, and an AOI camera captures the morphology of the adhesive strip in real time.

[0025] In some embodiments, dynamic production status information is generated in the following way: specifically, a decision tree algorithm is used to extract features from structured data: calculate the fluctuation range of parameters, such as the hourly fluctuation value of temperature; if the feature value exceeds a preset threshold, such as temperature > 30°C, it is marked as an abnormal feature; then abnormal status detection is performed and the status information is updated. For example, when the temperature fluctuation range of a certain device exceeds the preset threshold, the system marks "temperature abnormal" and updates the status information to "device A1 - temperature abnormal".

[0026] As a feasible implementation, the generation of effective datasets and deviation identification are achieved in the following ways: Specifically, the format of multi-source data is standardized, and the timestamp / unit is unified; mean filtering is used to remove noisy data points, such as deviation > 3℃ is considered noise; data points are classified using a decision tree algorithm to remove outliers with low feature matching; when the quality fluctuation parameter (such as glue volume deviation) exceeds the preset range, a deviation identification with the equipment location is generated. For example, when the glue volume deviation is detected to exceed 0.3 ml, the "Equipment B2 - Glue Volume Exceeds Standard" identification is generated.

[0027] In some embodiments, the steps of generating adaptive control instructions specifically include: parsing the anomaly type and equipment location in the deviation identifier; querying the operation instruction library to match preliminary control parameters, such as reducing the dispensing pressure by 5% for "excessive glue content"; filtering parameter ranges based on the equipment's dynamic response curve; and making secondary fine adjustments based on historical records of similar operating conditions, for example, for equipment B2 with excessive glue content, optimizing the pressure adjustment range from 5% to 7% based on its historical response data.

[0028] In some embodiments, dual-channel command transmission is implemented in the following ways: specifically, control commands are attached with device location identifiers and check codes; latency and packet loss rate are continuously monitored during main channel transmission; when latency exceeds a threshold or there is continuous packet loss, the backup channel is switched; the receiving end performs verification and reassembly and generates a transmission integrity report. For example, when the main channel transmission latency exceeds 2 seconds, the 5G backup channel is automatically switched to ensure that the command arrives completely at the execution unit.

[0029] As a feasible implementation example, the specific implementation process of dynamic control of equipment parameters is as follows: collect execution feedback data, such as actual glue volume and position accuracy; perform multi-dimensional deviation analysis with preset targets; when the glue dispensing position accuracy or glue volume uniformity exceeds the dynamic tolerance range: call the equipment mechanical response model to generate initial compensation parameters, verify the validity of the parameters by combining historical compensation records, and output compensation instructions containing equipment position identifiers. For example, when a glue strip position deviation of 0.1mm is detected, the instruction "Equipment M1 - Position Compensation + 0.08mm" is generated.

[0030] As a feasible implementation, closed-loop iterative optimization is achieved in the following way: Specifically, novel quality fluctuation features, such as the frequency of glue volume mutations, are extracted through time series analysis; when the features exceed the recognition threshold, the gradient descent algorithm is used to adjust the weights of the feature extraction model; a new set of operation instructions is generated based on the updated model; after verifying stability in a virtual environment, it is deployed to the real-time system. For example, when a new type of glue is found to cause abnormal glue volume fluctuation cycles, the model recognition parameters are updated and targeted pressure control instructions are generated.

[0031] In some embodiments, the control process optimization verification specifically includes: issuing new instructions to the dispensing equipment to adjust parameters; collecting data on the adjusted adhesive strip width and adhesive dot diameter in real time; and determining that the optimization is complete when the parameters continuously meet the process stability requirements (such as adhesive width tolerance ±0.1mm). For example, if the adhesive volume deviation is ≤0.1 ml for 10 consecutive dispensing cycles after the new instructions are deployed, the closed-loop optimization is determined to be successful.

[0032] This application fundamentally solves the problem of insufficient dynamic linkage in detection and control, which leads to inadequate personalized adjustment capabilities and difficulty in adapting to complex environments, through a technical chain of structured data stream processing, dynamic feature extraction, dual-channel reliable transmission, and closed-loop iterative optimization, thereby achieving a breakthrough improvement in dispensing accuracy and production efficiency.

[0033] In step S11, multi-source real-time data of the dispensing production line is acquired. The multi-source real-time data includes dispensing position coordinates, glue volume distribution data and glue uniformity parameters collected by optical sensors, and dispensing head temperature, pneumatic cylinder pressure and equipment vibration frequency collected by physical sensors.

[0034] In some embodiments, multi-source real-time data acquisition refers to the simultaneous acquisition of key parameters of the dispensing process by optical and physical sensors. The specific implementation process is as follows: AOI optical sensors acquire dispensing position coordinates, adhesive distribution data, and adhesive uniformity parameters; physical sensors acquire dispensing head temperature, pneumatic cylinder pressure, and equipment vibration frequency in parallel. The data from both types of sensors are aligned by timestamps, forming a complete foundation for production line status monitoring.

[0035] For example, on a certain dispensing production line, AOI sensors capture glue dot diameter data at a frequency of 5 frames per second; temperature sensors record the dispensing head temperature value every 0.5 seconds; and vibration sensors monitor the vibration frequency of the equipment base in real time.

[0036] The multi-source real-time data acquisition method of the above dispensing production line breaks through the limitations of traditional single data sources, provides comprehensive input for subsequent processing, ensures the completeness of quality fluctuation detection from the source, and avoids the problem of misjudgment caused by incomplete data coverage.

[0037] In step S12, based on multi-source real-time data, feature recognition processing is performed using a preset feature extraction model to obtain dynamically updated production status information. This includes: performing structured storage processing on the multi-source real-time data to generate a standardized data set according to equipment number and timestamp; using a decision tree algorithm to dynamically extract features from the standardized data set, marking features as abnormal features when the feature values ​​exceed preset parameter thresholds; performing real-time abnormal status detection based on the marking results, triggering production status updates if an abnormal status is detected; and associating and matching the updated abnormal features with historical operation records to generate dynamic production status information containing equipment status labels.

[0038] In some embodiments, structured data storage organizes raw data according to the spatiotemporal dimensions of the device. The specific implementation process is as follows: multi-source data is generated into a standardized dataset based on device number and timestamp; data fields include: device ID, timestamp, optical parameters (location / adhesive quantity / uniformity), and physical parameters (temperature / pressure / vibration). For example, the data record for device A1 at 10:00:00 on 2023-08-01 is: {Device ID: "A1", Timestamp: "20230801100000", Location coordinates: [x1, y1], Adhesive quantity: 0.25ml, Uniformity: 92%, Temperature: 28℃, Pressure: 2.3Bar, Vibration: 15Hz}.

[0039] In some embodiments, dynamic feature extraction and anomaly labeling are performed based on real-time feature analysis using decision trees. The specific implementation is as follows: key features are extracted using a decision tree algorithm, such as hourly temperature fluctuations and glue volume standard deviation. When a feature value exceeds a preset threshold, it is labeled as an anomalous feature. For example, if the hourly temperature fluctuation is >5℃, it is labeled "Temperature Anomaly"; if the glue volume standard deviation is >0.1ml, it is labeled "Glue Volume Fluctuation Anomaly". For instance, if a device detects a glue volume standard deviation of 0.15ml three times consecutively (the preset threshold is 0.1ml), the system labels it "Device B2 - Glue Volume Fluctuation Anomaly".

[0040] In some embodiments, a multi-condition triggered state management mechanism is used to achieve real-time anomaly detection and state updates. The specific implementation process is as follows: the marking result is verified in multiple dimensions, such as duration and associated parameters; if the verification passes, a state update is triggered. In one feasible embodiment, the update triggering condition needs to be met simultaneously: the abnormal feature persists for more than 3 detection cycles, and the associated parameter (such as pressure) deviates synchronously from the normal range. For example, when the temperature anomaly marker for device A1 persists for 5 minutes and the pressure rises synchronously by 10%, the system updates the state to "Double anomaly of device A1 - temperature and pressure".

[0041] In some embodiments, historically associated status information is generated by combining device-level status tags based on historical data. Current anomaly characteristics are matched with historical operating records, such as historical failure cases of the same equipment, to generate dynamic production status information containing device status tags. The tag generation logic combines current anomaly characteristics with similar historical cases to generate device status tags. For example, if the current glue quantity fluctuation characteristics of device B2 match the "glue valve blockage" case in the historical record, the status tag "Device B2 - Suspected glue valve blockage" is generated.

[0042] The method described above for obtaining dynamically updated production status information through feature recognition processing transforms raw multi-source real-time data into operable status tags, solving the problem of data being unable to be converted into commands and improving information conversion efficiency.

[0043] In step S13, interference filtering is performed on the production status information to obtain a valid dataset, including: standardizing the format of heterogeneous data in the production status information, and generating a spatiotemporally consistent standard dataset by aligning with timestamps and unifying with units; performing mean filtering on the spatiotemporally consistent standard dataset, and marking data points as noise points and removing them if the deviation between the data points and the mean of the sliding window exceeds a preset noise threshold, thereby generating a denoised dataset; and using a decision tree classifier to perform outlier detection processing on the denoised dataset, and determining outliers as outliers and removing them when the feature matching degree is lower than a preset confidence threshold, thereby generating a valid dataset.

[0044] In some embodiments, data format standardization processing includes spatiotemporal alignment and unit unification of heterogeneous data. The specific implementation process includes: extracting multi-source heterogeneous data from production status information, such as temperature, pressure, and vibration; unifying data from different acquisition frequencies to a reference time unit through timestamp alignment; and converting all data units to the International System of Units (SI). For example, temperature sensor data per minute and vibration sensor data per 10 seconds are unified to a per-minute reference; pressure units are uniformly converted to Pascals (Pa), and vibration units are uniformly converted to Hertz (Hz).

[0045] In some embodiments, noise point identification and removal based on a sliding window achieves mean filtering denoising. Specifically, a sliding window with a time dimension is set, such as a 5-minute window; the arithmetic mean of the data points within the window is calculated; if the deviation between the data point and the window mean exceeds a preset noise threshold, it is marked as a noise point; and noise points are removed from the dataset to generate a denoised dataset.

[0046] As a feasible embodiment, the logic for setting the preset noise threshold is as follows: the temperature noise threshold is set to ±3℃, and the pressure noise threshold is set to ±200Pa. For example, if the average temperature within a 5-minute window is 25℃, and a data point is 32℃ (deviation 7℃ > 3℃), it is marked as a noise point and removed.

[0047] In some embodiments, outlier identification is performed based on multi-feature joint judgment using a decision tree. A feature matching model is constructed using a decision tree classifier; input parameters include denoised data such as temperature, pressure, and vibration; when the feature matching degree is lower than a preset confidence threshold (e.g., 85%), it is identified as an outlier; outliers are removed to generate a valid dataset.

[0048] As a specific implementation, the decision rule is as follows: for example, if the temperature is >30℃ and the vibration is >20Hz, the feature matching degree is 92%; for another example, if the temperature is >30℃ and the vibration is ≤15Hz, the feature matching degree is 75%. For instance, a data point has a temperature of 28℃ (normal), a pressure of 2.2 Bar (normal), and a vibration of 25Hz (abnormal), with a comprehensive matching degree of 78% < 85%, and is therefore judged as an outlier and removed.

[0049] In some embodiments, data validity verification is achieved through state consistency verification via multi-dimensional data fusion. Data validity verification is implemented by: weighted fusion of temperature, pressure, and vibration parameters from the valid dataset; calculation of a comprehensive state index value; and triggering a state update if the index value meets the preset production state condition range. For example, under normal state conditions, temperature [20-30℃] ∩ pressure [2.0-2.5 Bar] ∩ vibration [10-15Hz], the state is updated to "Equipment operating normally" when the fused data meets all conditions.

[0050] The above-mentioned method for obtaining effective datasets by filtering interference based on production status information solves the problem of multi-source heterogeneity through data standardization, improves processing efficiency, eliminates instantaneous interference by mean filtering, reduces noise false alarm rate, and improves the accuracy of anomaly identification by decision tree multi-feature judgment, thereby increasing the retention rate of effective data.

[0051] In step S13, when a quality fluctuation exceeding a preset range is detected, an alarm is triggered and deviation identification information is generated, including: extracting dynamic change features based on a valid dataset through a configurable sliding window, the size of which can be adaptively adjusted according to the equipment type and production stage; calculating the feature change rate of the dynamic change features; if the feature change rate exceeds a first-level threshold, triggering a primary alarm and generating a deviation identification with the equipment number; if the feature change rate exceeds a second-level threshold, triggering a critical alarm and generating a deviation identification with a spatial location index.

[0052] In some embodiments, dynamic feature extraction is achieved through an adaptive sliding window mechanism. Specifically, a configurable sliding window is set based on the effective dataset; the window size is dynamically adjusted according to the device type and production stage: a small window, such as 3 minutes, is used for high-precision devices, while a large window, such as 10 minutes, is used for stable production stages.

[0053] For example, the feature extraction logic specifically includes: capturing data points within a window every minute; calculating the temporal variation characteristics of parameters such as temperature and pressure. For instance, in the precision dispensing stage, a 3-minute sliding window is enabled for device C3, and the changing trend of adhesive distribution within the window is analyzed every minute.

[0054] In some embodiments, the calculation of feature change is a multi-parameter change quantification analysis. Specifically, the time-series features extracted by the sliding window are differentiated; the feature change rate per unit time is calculated by first calculating the difference between the current value and the previous value, and then dividing the difference by the time interval to obtain the change. For example, if the temperature rises from 26.5℃ to 27.8℃ within a 3-minute window, the change rate is calculated as (27.8-26.5) / 3 = 0.43℃ / minute.

[0055] In some embodiments, tiered alarm triggering is achieved through a dual-threshold determination mechanism. Specifically, a primary threshold (warning line) and a secondary threshold (danger line) are preset, and the triggering conditions and response strategies are as follows: when the rate of change is greater than the primary threshold, a primary alarm is triggered, and the corresponding identifier is the device number and the anomaly type; when the rate of change is greater than the secondary threshold, a critical alarm is triggered, and the corresponding identifier is the space number, the device number, and the anomaly level.

[0056] For example, the first threshold for temperature change rate is 0.5℃ / minute, and the second threshold for temperature change rate is 0.7℃ / minute. When the temperature change rate of device D4 is detected to reach 0.6℃ / minute, a primary alarm is triggered and a "Device D4 - Temperature Abnormal" flag is generated.

[0057] In some embodiments, deviation identifier generation is achieved through spatial location index binding. In the event of a critical alarm, the spatial coordinates detected by the AOI are associated; a deviation identifier containing three-dimensional location information is generated: {Device ID: "E5", Coordinates: [X120, Y80, Z5], Anomaly Type: "Sudden Increase in Adhesive Volume", Level: "Critical"}.

[0058] For example, when the rate of change of adhesive amount exceeds the secondary threshold, a "location [X50,Y30] - adhesive amount abnormal" label is generated by combining the AOI visual coordinates.

[0059] In some embodiments, real-time database linkage enables status-related updates. Specifically, deviation identifiers are bound to equipment operating status parameters; production status database fields are updated: timestamp of the anomaly occurrence; associated real-time temperature / pressure values; and processing status flags. For example, after equipment F6 triggers a critical alarm, the database record is updated as follows: {Time: "2023-08-01 14:05", Equipment: "F6", Status: "Critical Anomaly", Temperature: 75℃, Pressure: 3.2 Bar}.

[0060] The above-mentioned method for triggering alarms and generating deviation identification information uses a sliding window for adaptive adjustment to reduce the false alarm rate, dual-threshold hierarchical alarms to optimize resource allocation efficiency, and spatial location indexing to improve anomaly location accuracy to the millimeter level.

[0061] In step S14, based on the deviation identification information, adaptive parameter optimization is performed in combination with equipment characteristics and historical operation records to generate control instructions. This includes: parsing the anomaly type and equipment position in the deviation identification information, matching the pre-stored operation instruction library to generate a preliminary control parameter set; filtering the preliminary control parameter set based on the dynamic response curve and dispensing accuracy parameters in the equipment characteristics; fine-tuning the filtered parameters a second time using a logistic regression model, combined with similar working condition optimization data in historical operation records; and generating a personalized control instruction containing the equipment position identifier when the fine-tuned control parameters meet the personalized dispensing control requirements.

[0062] In some embodiments, environmental data standardization is achieved through cleaning and integrating multi-source environmental variables. Specifically, environmental variables (humidity / vibration frequency) and device status (runtime / power) are collected from distributed sensor nodes; threshold filtering is used to remove outliers caused by signal interference, such as humidity fluctuations exceeding 15%; and a standardized dataset {timestamp, device location identifier, humidity value, vibration value, runtime, power value} is generated. For example, the data record for device B2 is {time: 09:00, device location identifier: "B2", humidity: 75%, vibration: 15Hz, runtime: 8h, power: 95%}.

[0063] In some embodiments, the parameter mapping matching mechanism is achieved through rapid matching of environmental variables with preset thresholds. Specifically, a preset parameter mapping table is used for threshold matching, such as a humidity threshold of 70% and a vibration threshold of 12Hz; when environmental variables exceed the limits, an adaptive control algorithm is triggered. For example, if the humidity exceeds the 75% threshold, the system marks "humidity exceeds the limit" and activates the control parameter adjustment process.

[0064] In some embodiments, the generation of preliminary control parameters is achieved through a matching mechanism of the operation instruction library. Specifically, the environmental anomaly type and equipment location identifier are parsed; a preliminary control parameter set is generated by matching the pre-stored operation instruction library. For example, for the "humidity exceeds the standard" identifier, the preliminary parameter "reduce ventilation power by 5%" is obtained.

[0065] As an optional implementation method, parameter amplitude selection refers to optimization based on the device's dynamic response curve. The specific implementation process is as follows: based on the dynamic response curve of the device characteristics, such as the power adjustment-humidity change relationship, and combined with dispensing accuracy parameters (such as glue width tolerance ±0.1mm), the parameter amplitude is selected. For example, the response curve of the old device B2 shows that the adjustment amplitude needs to be increased by 20%.

[0066] As an optional implementation method, historical data fine-tuning refers to secondary optimization under similar operating conditions. The specific implementation process is as follows: Similar operating condition data from historical operation records is retrieved, for example, ambient humidity > 70% and operating time > 6 hours; the optimal adjustment range is calculated using a logistic regression model. For example, historical records show that reducing humidity to 68% under the same operating conditions requires a 7% power reduction; therefore, the initial parameter is fine-tuned from 5% to 7%.

[0067] As an optional implementation, instruction generation and issuance are achieved through a closed-loop verification mechanism. When the parameters meet the personalized dispensing control requirements, the final instruction is generated. The instruction format is {equipment location identifier: "B2", instruction: "reduce power by 7%", target value: 68%}. After issuance, the humidity change curve is monitored in real time. For example, if the humidity drops to 70% after 10 minutes, and the target is not reached, compensation adjustment is triggered.

[0068] The above-mentioned method for generating personalized control commands ensures input reliability through environmental data standardization, improves parameter adaptation accuracy through equipment dynamic response curves, and enhances control efficiency through fine-tuning based on historical operating conditions.

[0069] In step S15, control commands are sent to the execution unit through a dual-channel transmission mechanism, and the transmission delay is monitored in real time and the channel is automatically switched. This includes: processing the control commands into data packets and attaching device location identifiers and integrity check codes to obtain data packets; transmitting data packets in real time through the main communication channel and continuously monitoring the transmission delay and packet loss rate; and automatically switching to the backup communication channel to complete the transmission of the remaining control commands when the transmission delay exceeds the dynamic response threshold or the packet loss rate exceeds the standard continuously.

[0070] In some embodiments, control command preprocessing is specifically implemented through the following steps: Data packet processing is performed on the control commands, and a device location identifier and integrity check code are appended to generate data packets. The device location identifier is derived from the location code bound in the adaptive parameter optimization stage, and the integrity check code is generated using a hash algorithm. For example, a speed adjustment command (target value 75%) for device M1 is split into multiple data packets, each carrying the location identifier M1 and a CRC32 check code.

[0071] In some embodiments, dual-channel transmission monitoring is specifically implemented through the following steps: Data packets are transmitted in real time through the main communication channel, and transmission latency and packet loss rate are continuously monitored. Transmission latency detection includes recording the time difference between the instruction sending time and the execution unit receiving the acknowledgment time. Packet loss rate calculation includes statistically analyzing the proportion of data packets for which no acknowledgment response has been received.

[0072] As an optional implementation, the system automatically switches to the backup communication channel when the following occurs: the transmission delay exceeds the dynamic response threshold (set according to the device location sensitivity); or the packet loss rate exceeds the standard for three consecutive data packets. For example, when the main channel transmits instructions from device M1, if the delay exceeds 200ms (dynamic response threshold) or the consecutive packet loss rate is >5%, the backup wireless channel is immediately activated to transmit the remaining data packets.

[0073] In some embodiments, instruction integrity verification is performed by the execution unit after receiving the data packet, followed by reassembly verification. Specifically, the device location identifier is checked for matching to confirm the device to which the instruction belongs; data integrity is verified using an integrity check code; if the verification fails, retransmission is requested. For example, if device M1 detects an error in the check code of a data packet, it sends a NACK signal to the system to trigger the retransmission of that packet until all packets are successfully reassembled.

[0074] In some embodiments, after the execution unit feeds back the actual operating parameters, it performs the following operations: calculates the deviation between the actual parameters and the expected values, such as comparing an actual speed of 74% with a target of 75%, and performs a matching degree analysis. It then executes a deviation compensation decision. If the deviation is within the tolerance range, such as ±1%, it records the execution verification data; if the deviation exceeds the tolerance range, such as >2%, it generates a compensation instruction. For example, if the actual speed of device M1 is 74% and the target is 75%, a deviation of 1% is within the tolerance range, and the verification is marked as passed; if the actual speed is 73%, a compensation instruction of speed +2% is generated. Finally, it performs adaptive tolerance adjustment, dynamically adjusting the tolerance threshold based on the continuous operating time of the device. For example, after device M1 has run continuously for 12 hours, the speed tolerance threshold is relaxed from ±1% to ±1.5%.

[0075] As an optional implementation, the execution result verification data is used for iterative optimization. Specifically, the execution effect of compensation instructions is recorded, and the historical operation record library is updated; over-tolerance events trigger the update of feature extraction model parameters. For example, if device M1 experiences multiple speed deviations, the decision tree splitting threshold in the feature extraction model is updated after analyzing new fluctuation characteristics.

[0076] The aforementioned method for monitoring transmission delay and automatically switching channels addresses command transmission delay and loss issues through device location identifier binding, automatic dual-channel switching, and a complete retransmission mechanism. Dynamic tolerance adjustment and compensation command generation in the execution feedback process form a closed-loop control, ensuring precise execution of control commands. All aspects are linked through device location identifiers (e.g., M1), guaranteeing the continuity of personalized control.

[0077] In step S16, the equipment parameters are dynamically adjusted based on the execution feedback data. When the output deviation exceeds the tolerance range, a compensation instruction is generated. This includes: collecting equipment operating status data and dispensing quality parameters from the execution unit, formatting them to generate a standardized feedback dataset; performing multi-dimensional deviation analysis between the standardized feedback dataset and preset target parameters, and marking the deviation type and equipment position when the dispensing position accuracy or glue volume uniformity exceeds the dynamic tolerance range; calling the mechanical response model in the equipment characteristic library based on the marking results, and generating initial compensation parameters by combining historical compensation records; fine-tuning and verifying the initial compensation parameters through an adaptive control algorithm; and outputting a compensation instruction containing the equipment position identifier when the fine-tuned parameters meet the dispensing process stability requirements.

[0078] In some embodiments, the acquisition and standardization of operational status data specifically includes collecting real-time operational status data from production line equipment and generating a standardized feedback dataset through formatting processing. The real-time operational status data includes equipment operating speed, temperature, vibration, and dispensing quality parameters (adhesive uniformity, dispensing position accuracy). Generating the standardized feedback dataset through formatting processing is achieved through the following steps: reorganizing the data according to three dimensions—time stamp, equipment number, and parameter type—to achieve formatting processing; and capturing the adhesive strip width and position coordinates in real time using an AOI detection system to obtain dispensing quality parameters. For example, welding robot R1 collects data at 10:00, showing an operating speed of 90%, a temperature of 75℃, and a dispensing position deviation of 0.05mm, which is then formatted and stored as {Time: 10:00, Equipment: R1, Speed: 90%, Temperature: 75℃, Position Deviation: 0.05mm}.

[0079] As an optional implementation method, multi-dimensional deviation analysis refers to matching and analyzing standardized feedback datasets with preset target parameters. Specifically, the preset target parameters are sourced from: speed range, temperature limit, glue volume tolerance, etc. in the process specification database; dynamic tolerance range setting: the threshold is automatically adjusted according to the continuous working time of the equipment; deviation marking rules: when the glue dispensing position accuracy is >0.1mm or the glue volume uniformity is >15%, an anomaly is marked. For example, if the speed of conveyor belt T1 is 2.2m / s, while the target is ≤2m / s, and the glue volume uniformity is 18%, while the target is ≤15%, the system marks the deviation type as "speed exceeds the standard" and "glue volume is uneven", and associates it with the equipment position T1.

[0080] As an optional implementation, compensation parameters are generated by performing compensation decisions based on deviation marking results. Specifically, the mechanical response model is invoked: the response model corresponding to the position number is extracted from the equipment characteristic library, such as the conveyor belt speed-load relationship curve. Initial compensation parameters are generated: successful cases are screened based on historical compensation records; the basic compensation value is calculated through linear regression. For example, for T1 speed exceeding the standard, its mechanical response model is invoked, and the initial compensation parameter "speed -0.3m / s" is generated based on historical records. Parameter fine-tuning verification: the feasibility of the parameters is verified using an adaptive control algorithm, which must meet the following requirements: the uniformity of adhesive content after adjustment ≤12%, and the positional accuracy error ≤0.08mm. For example, after fine-tuning the initial parameter "speed -0.3m / s" to "speed -0.25m / s", the simulation results meet the process stability requirements.

[0081] In some embodiments, the compensation command output refers to generating a compensation command carrying the device location identifier. Specifically, the command structure is {target device: T1, adjustment parameter: speed, adjustment value: -0.25m / s, validity period: 2 hours}. Process binding mechanism: When the compensation value touches the process boundary, such as the lower limit of speed 1.5m / s, it is automatically replaced with a safety value. For example, if the current speed of T1 is 2.2m / s, after the compensation command -0.25m / s, it becomes 1.95m / s, which is higher than the lower limit of process speed 1.5m / s, and the command is valid.

[0082] In some embodiments, closed-loop optimization is achieved by collecting new feedback data after the compensation command is executed. Specifically, matching degree reassessment: calculating the deviation between the actual parameters and the compensation target. Adaptive tolerance adjustment: dynamically relaxing the threshold based on the aging coefficient of the equipment's runtime. For example, after T1 is executed, the actual speed is 1.93 m / s, the target is 1.95 m / s, and the deviation is 0.02 m / s. If the equipment runs continuously for 10 hours, the tolerance threshold is relaxed from 0.05 m / s to 0.07 m / s, and the compensation is considered successful.

[0083] The above-mentioned method for generating compensation instructions simultaneously monitors equipment operating parameters (speed / temperature) and dispensing quality parameters (position accuracy / dispensing quantity uniformity), breaking through the limitations of single-dimensional monitoring. It optimizes decision-making efficiency through a two-level compensation parameter generation mechanism, and the dynamic tolerance mechanism can ensure the safety of regulation.

[0084] In step S16, the parameters of the feature extraction model are updated based on the novel quality fluctuation characteristics, and the dispensing control process is iteratively optimized through closed-loop control. This includes: acquiring dispensing quality fluctuation feedback data from the production equipment sensors and extracting novel fluctuation characteristics using time series analysis; when the novel fluctuation characteristics exceed a preset recognition threshold, adjusting the weight parameters of the feature extraction model using a gradient descent algorithm; generating a new set of operation instructions based on the updated feature extraction model using a logistic regression method; sending the new set of operation instructions to the dispensing equipment to adjust the operating parameters and collecting the adjusted dispensing quality data in real time; and determining that the closed-loop optimization is complete when the adjusted dispensing quality parameters continuously meet the preset stability requirements.

[0085] In some embodiments, dispensing quality data acquisition and feature extraction are achieved by obtaining dispensing quality feedback data from sensors in the production equipment. The dispensing quality feedback data includes parameters such as glue quantity, glue strip width, and uniformity.

[0086] As a feasible implementation method, novel fluctuation characteristics are extracted through time series analysis. Specifically, the time series analysis is implemented by calculating the fluctuation frequency, amplitude, and trend change rate. The feature set is composed of three-dimensional indicators: peak deviation, period length, and duration. For example, if a sensor collects adhesive data for a certain period of time, the analysis shows that the fluctuation frequency is 2 times per minute, the peak deviation is 0.5 ml, and the trend is upward, a feature set {frequency: 2 times / min, peak deviation: 0.5 ml, trend: upward} is generated.

[0087] In some embodiments, when a novel fluctuation feature exceeds a preset recognition threshold, a gradient descent algorithm is used to adjust the feature extraction model parameters. The specific implementation process is as follows: Threshold setting is based on: the upper limit of the historical normal fluctuation range, such as a peak deviation ≤ 0.3 ml; parameter optimization constraints: binding to process safety boundaries, such as a minimum adhesive volume threshold; gradient descent execution: dynamically setting the learning rate based on the feature deviation, for example, if a peak deviation of 0.5 ml > the threshold of 0.3 ml is detected, gradient descent is initiated to optimize the model weight parameters. If the current adhesive volume is close to the minimum process limit, the parameter adjustment range is limited.

[0088] As one feasible implementation, the operation instructions are generated using a logistic regression method based on the updated feature extraction model, resulting in a new set of operation instructions. Specifically, the input-output mapping maps fluctuation features to equipment parameter adjustment amounts; the regression verification mechanism simulates adhesive volume changes before instruction execution. For example, for peak deviation features, the logistic regression outputs a dispensing pressure adjustment instruction: from 3.5 bar to 3.2 bar. The generated instruction set is {Equipment: D1, Parameter: Pressure, Adjustment Value: -0.3 bar}.

[0089] As a feasible implementation method, the process of instruction execution and effect monitoring is achieved by sending a new set of operation instructions to the dispensing equipment to adjust the operating parameters and collecting the adjusted quality data in real time. Specifically, key monitoring indicators include: glue volume deviation, positional accuracy, and uniformity; data acquisition frequency: at least once per second. For example, after the equipment D1 performs pressure adjustment, the sensor provides real-time feedback of glue volume data, and the system records that the peak deviation after adjustment has decreased to 0.2 ml.

[0090] In some embodiments, closed-loop optimization is determined when the adjusted dispensing quality parameters continuously meet preset stability requirements. The stability requirement is: dispensing volume deviation < tolerance threshold for N consecutive cycles. The dynamic determination mechanism adjusts the number of cycles based on the colloid characteristics; for example, in a high-temperature environment, optimization is considered successful if the dispensing volume deviation is <0.15ml for 5 consecutive sampling cycles. If the deviation exceeds the threshold, a new round of model update is triggered.

[0091] The above-mentioned closed-loop control iteratively optimizes the dispensing control process, generates compensation instructions based on feedback data, and continuously updates the feature extraction model, thus overcoming the bottleneck of being unable to adapt to fluctuations in information quality.

[0092] Reference Figure 2 The second embodiment of the present invention provides an intelligent control dispensing system based on automatic optical detection, comprising: The data acquisition module is used to acquire multi-source real-time data of the dispensing production line. The multi-source real-time data includes dispensing position coordinates, glue volume distribution data and glue uniformity parameters collected by optical sensors, and dispensing head temperature, pneumatic cylinder pressure and equipment vibration frequency collected by physical sensors. The production status determination module is used to perform feature recognition processing based on multi-source real-time data through a preset feature extraction model to obtain dynamically updated production status information. The alarm and deviation identification generation module is used to perform interference filtering based on production status information to obtain a valid dataset. When a quality fluctuation exceeding the preset range is detected, an alarm is triggered and deviation identification information is generated. The control command generation module is used to generate control commands by performing adaptive parameter optimization based on deviation identification information, combined with equipment characteristics and historical operation records. The communication channel switching module is used to send control commands to the execution unit through a dual-channel transmission mechanism, monitor the transmission delay in real time, and automatically switch channels. The dispensing control optimization module is used to dynamically adjust equipment parameters based on execution feedback data. When the output deviation exceeds the tolerance range, a compensation command is generated, and the parameters of the feature extraction model are updated based on the new quality fluctuation characteristics. The dispensing control process is iteratively optimized through closed-loop control.

[0093] It should be noted that the intelligent control dispensing system based on automatic optical detection provided in this embodiment of the invention is used to execute all the process steps of the intelligent control dispensing method based on automatic optical detection in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0094] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0095] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A smart control dispensing method based on automatic optical detection, characterized in that, include: Acquire multi-source real-time data from the dispensing production line, wherein the multi-source real-time data includes dispensing position coordinates, glue volume distribution data and glue uniformity parameters collected by optical sensors, and dispensing head temperature, pneumatic cylinder pressure and equipment vibration frequency collected by physical sensors. Based on the multi-source real-time data, feature recognition processing is performed through a preset feature extraction model to obtain dynamically updated production status information; Based on the production status information, interference filtering is performed to obtain a valid dataset. When a quality fluctuation exceeding a preset range is detected, an alarm is triggered and deviation identification information is generated. Based on the deviation identification information, adaptive parameter optimization is performed in combination with equipment characteristics and historical operation records to generate control commands. The control commands are sent to the execution unit through a dual-channel transmission mechanism, and the transmission delay is monitored in real time and the channel is automatically switched. The equipment parameters are dynamically adjusted based on the execution feedback data. When the output deviation exceeds the tolerance range, a compensation command is generated. The parameters of the feature extraction model are updated based on the new quality fluctuation characteristics. The dispensing control process is iteratively optimized through closed-loop control.

2. The intelligent control dispensing method based on automatic optical detection according to claim 1, characterized in that, The step of obtaining dynamically updated production status information by performing feature recognition processing based on the multi-source real-time data through a preset feature extraction model includes: The multi-source real-time data is processed for structured storage, and a standardized data set is generated according to the device number and timestamp; The decision tree algorithm is used to dynamically extract features from the standardized dataset, and features that exceed a preset parameter threshold are marked as abnormal features. Real-time abnormal status detection is performed based on the marking results. If an abnormal status is detected, a production status update is triggered. The updated anomaly features are correlated and matched with historical operation records to generate dynamic production status information containing equipment status tags.

3. The intelligent control dispensing method based on automatic optical detection according to claim 1, characterized in that, The step of obtaining a valid dataset by performing interference filtering based on the production status information includes: The heterogeneous data in the production status information is formatted and a standard dataset with spatiotemporal consistency is generated by aligning timestamps and unifying units. The mean filtering process is performed on the spatiotemporally consistent standard dataset. If the deviation between the data point and the mean of the sliding window exceeds the preset noise threshold, it is marked as a noise point and removed to generate a denoised dataset. A decision tree classifier is used to process outliers in the denoised dataset. When the feature matching degree is lower than a preset confidence threshold, it is identified as an outlier and removed to generate a valid dataset.

4. The intelligent control dispensing method based on automatic optical detection according to claim 1, characterized in that, The step of triggering an alarm and generating deviation identification information when a quality fluctuation exceeding a preset range is detected includes: Based on the effective dataset, dynamic change features are extracted through a configurable sliding window, the size of which can be adaptively adjusted according to the device type and production stage. Calculate the characteristic change rate of the dynamic change feature; If the rate of change of the feature exceeds the first-level threshold, a primary alarm is triggered and a deviation identifier containing the device number is generated; If the rate of change of the feature exceeds the secondary threshold, a critical alarm is triggered and a deviation identifier containing a spatial location index is generated.

5. The intelligent control dispensing method based on automatic optical detection according to claim 1, characterized in that, The step of generating control commands based on the deviation identification information, combined with equipment characteristics and historical operating records, and adaptive parameter optimization processing includes: Parse the anomaly type and equipment location in the deviation identification information, and match them with the pre-stored operation instruction library to generate a preliminary control parameter set; The preliminary control parameter set is filtered based on the dynamic response curve and dispensing accuracy parameters in the equipment characteristics. By combining similar operating condition optimization data from historical operation records, the selected parameters are fine-tuned a second time using a logistic regression model; When the fine-tuned control parameters meet the personalized dispensing control requirements, a personalized control command containing the equipment location identifier is generated.

6. The intelligent control dispensing method based on automatic optical detection according to claim 1, characterized in that, The method of sending control commands to the execution unit through a dual-channel transmission mechanism, monitoring transmission delay in real time and automatically switching channels includes: The control commands are processed into data packets and an additional device location identifier and integrity check code are added to obtain data packets; The data packets are transmitted in real time through the main communication channel, and the transmission latency and packet loss rate are continuously monitored. When the transmission delay exceeds the dynamic response threshold or the packet loss rate exceeds the standard continuously, the system automatically switches to the backup communication channel to complete the transmission of the remaining control commands.

7. The intelligent control dispensing method based on automatic optical detection according to claim 1, characterized in that, The step of dynamically adjusting device parameters based on execution feedback data, and generating a compensation command when the output deviation exceeds the tolerance range, includes: The equipment operation status data and dispensing quality parameters fed back by the execution unit are collected, formatted, and generated into a standardized feedback dataset. The standardized feedback dataset is subjected to multi-dimensional deviation analysis with the preset target parameters. When the dispensing position accuracy or glue volume uniformity exceeds the dynamic tolerance range, the deviation type and equipment position are marked. Based on the marking results, the mechanical response model in the equipment characteristic library is called, and the initial compensation parameters are generated by combining the historical compensation records. The initial compensation parameters are fine-tuned and verified using an adaptive control algorithm. When the fine-tuned parameters meet the stability requirements of the dispensing process, a compensation command containing the equipment position identifier is output.

8. The intelligent control dispensing method based on automatic optical detection according to claim 1, characterized in that, The process of updating the parameters of the feature extraction model based on novel quality fluctuation characteristics and iteratively optimizing the dispensing control process through closed-loop control includes: The dispensing quality fluctuation feedback data is obtained from the sensors of the production equipment, and novel fluctuation characteristics are extracted through time series analysis. When the novel fluctuation feature exceeds the preset recognition threshold, the weight parameters of the feature extraction model are adjusted using the gradient descent algorithm. Based on the updated feature extraction model, a new set of operation instructions is generated using the logistic regression method. The new set of operating instructions is sent to the dispensing equipment to adjust the operating parameters, and the adjusted dispensing quality data is collected in real time. When the adjusted dispensing quality parameters consistently meet the preset stability requirements, the closed-loop optimization is considered complete.

9. An intelligent control dispensing system based on automatic optical inspection, characterized in that, include: The data acquisition module is used to acquire multi-source real-time data of the dispensing production line. The multi-source real-time data includes dispensing position coordinates, glue volume distribution data and glue uniformity parameters collected by optical sensors, and dispensing head temperature, pneumatic cylinder pressure and equipment vibration frequency collected by physical sensors. The production status determination module is used to perform feature recognition processing based on the multi-source real-time data through a preset feature extraction model to obtain dynamically updated production status information. The alarm and deviation identification generation module is used to perform interference filtering processing based on the production status information to obtain a valid dataset. When a quality fluctuation is detected to exceed a preset range, an alarm is triggered and deviation identification information is generated. The control command generation module is used to generate control commands by performing adaptive parameter optimization processing based on the deviation identification information, combined with equipment characteristics and historical operation records. The communication channel switching module is used to send the control commands to the execution unit through a dual-channel transmission mechanism, monitor the transmission delay in real time, and automatically switch channels. The dispensing control optimization module is used to dynamically adjust equipment parameters based on execution feedback data. When the output deviation exceeds the tolerance range, a compensation command is generated, and the parameters of the feature extraction model are updated based on the new quality fluctuation characteristics. The dispensing control process is iteratively optimized through closed-loop control.

Images