Intelligent cleaning and drilling cooperative system for glass processing

Abstract

The invention relates to the technical field of glass processing, and discloses an intelligent cleaning and drilling cooperative system for glass processing. The system comprises a multi-modal information acquisition module, a strategy generation module and an equipment linkage execution module. The multi-modal information acquisition module forms a multi-dimensional signal through the data integration node unit; in the strategy generation module, an estimation unit outputs a stain residue signal and a drilling deviation risk signal, an optimization unit obtains an optimal cleaning signal, and a regulation and control unit outputs a collaborative operation threshold value; in the equipment linkage execution module, a self-adaptive cleaning execution unit adjusts parameters according to the optimal cleaning signal, a drilling equipment group control unit adjusts the parameters according to a collaborative operation threshold value and a multi-dimensional signal and collects a monitoring signal, and a man-machine interaction and prompt module outputs a comprehensive abnormal index. According to the system, intelligent cooperation of the cleaning process and the drilling process is achieved, the operation continuity and accuracy are improved through multi-module linkage, and the defects of traditional separated operation are overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of glass processing, in particular to a glass processing intelligent cleaning and drilling collaborative system. BACKGROUND

[0002] In the processing of glass products, cleaning and drilling are two closely related processes, and their operation quality directly affects the performance and pass rate of the final product. In the traditional glass processing mode, the cleaning and drilling processes are often independent of each other, using their own control logic and equipment operating parameters, and lacking effective collaboration mechanisms. This separated operation mode makes it difficult to form a positive interaction between cleaning effect and drilling precision, and can easily cause a series of problems.

[0003] From the cleaning process, the existing technology relies on fixed cleaning parameters such as spray pressure, cleaning agent concentration, and cleaning time, which are usually set according to experience and cannot be dynamically adjusted according to the actual situation of the glass surface stains. The composition of stains on the glass surface is complex and may contain different types of pollutants such as oil, dust, and metal debris, and the distribution density and adhesion strength of stains also vary. When faced with different types and degrees of stains, fixed cleaning parameters often cannot achieve the desired cleaning effect, and some areas may have residual stains. These residual stains not only affect the appearance quality of the glass, but also have a negative impact on the subsequent drilling process. For example, the presence of stains can cause optical recognition errors during drilling positioning, resulting in drilling position deviation and affecting the assembly precision of the glass product. In addition, residual stains can also form small particles during drilling, which can scratch the glass surface with the high-speed rotation of the drill bit, reducing the yield of the product.

[0004] From the drilling process, the traditional drilling equipment usually operates according to the preset program, and its drilling speed, feed rate, and other parameters will not be adjusted according to the real-time state of the glass. In actual operation, the uniformity of the glass material, the surface flatness, and the residual state after cleaning can all affect the drilling process. For example, if there are local stress concentration areas on the glass surface, or the surface friction coefficient changes due to incomplete cleaning, drilling according to fixed parameters may cause drilling deviation or even glass breakage. At the same time, the debris generated during drilling cannot be cleaned in time and will also adhere to the glass surface, forming new stains and increasing the difficulty of subsequent processing.

[0005] In the prior art, although some production lines attempt to introduce automatic control technology, most of them are limited to the optimization of a single process and fail to realize the linkage of cleaning and drilling equipment. There is a lack of information sharing between the cleaning equipment and the drilling equipment, and it is impossible to adjust the parameters of the self according to the working state of the other party. For example, when the drilling equipment detects drilling deviation, it cannot feed back to the cleaning equipment in time to optimize the subsequent cleaning strategy; the cleaning equipment also cannot transmit the stain distribution information to the drilling equipment to help it adjust the drilling parameters in advance. This information island phenomenon leads to poor adaptability and fault tolerance of the entire processing process.

[0006] With the increasing application of glass products in the fields of building, electronics, automobiles, etc., the requirements for its processing precision and surface quality are continuously increasing. The traditional separated operation mode and simple automatic control have been unable to meet the production demand of high precision and high efficiency, and how to realize the intelligent collaboration of cleaning and drilling processes has become a problem to be solved in the current glass processing field. SUMMARY

[0007] The purpose of the present application is to provide a glass processing intelligent cleaning and drilling collaboration system to solve the problems raised in the background art.

[0008] To achieve the above purpose, the present application provides a glass processing intelligent cleaning and drilling collaboration system, which comprises:

[0009] A multi-modal information acquisition module, which comprises a data integration node unit, and the data integration node unit forms a multi-dimensional signal;

[0010] A strategy generation module, which receives the multi-dimensional signal, and the strategy generation module comprises a prediction unit, an optimization unit and a control unit, the prediction unit processes the multi-dimensional signal to output a stain residue signal and a drilling deviation risk signal, the optimization unit processes the multi-dimensional signal and the stain residue signal to obtain an optimal cleaning signal, and the control unit defines an accuracy constraint condition according to the multi-dimensional signal to output a collaborative operation threshold;

[0011] An equipment linkage execution module, which comprises an adaptive cleaning execution unit, a drilling equipment group control unit and a man-machine interaction and prompting module, the adaptive cleaning execution unit adjusts the cleaning parameters according to the optimal cleaning signal, the drilling equipment group control unit adjusts the drilling equipment parameters according to the collaborative operation threshold and the multi-dimensional signal, and collects real-time monitoring signals, and the man-machine interaction and prompting module processes the drilling deviation risk signal and the stain residue signal and outputs a comprehensive abnormality index.

[0012] Preferably, the multi-modal information acquisition module includes a distributed sensing array unit and a visual recognition detection unit, the multi-modal information acquisition module is arranged in the cleaning tank, the drilling machine table and the conveying track, and collects sensing signals, the visual recognition detection unit obtains glass surface feature data through image acquisition technology, and forms a surface signal, the data integration node unit receives the sensing signals and the surface signal to realize heterologous data fusion, and forms the multi-dimensional signal, the multi-dimensional signal includes an estimated multi-dimensional signal, an optimal multi-dimensional signal and a regulation multi-dimensional signal.

[0013] Preferably, the estimation unit includes a convolutional neural network-based image analysis model and a fuzzy inference fusion model, the estimated multi-dimensional signal includes glass surface gray data, equipment vibration frequency data, visual recognition feature gradient data, historical operation parameters and historical stain residual thickness data, the glass surface gray data, the historical operation parameters, the equipment vibration frequency data, the visual recognition feature gradient data and the historical stain residual thickness data are input into the image analysis model and output the stain residual signal, and the glass surface gray data, the equipment vibration frequency data and the visual recognition feature gradient data are fused through the improved fuzzy inference fusion model and output the drilling deviation risk signal.

[0014] Preferably, the image analysis model includes an image feature extraction submodule, a multi-signal fusion submodule, an abnormality quantization submodule and an estimation result display module, the image feature extraction submodule performs time-frequency domain decomposition on the equipment vibration frequency data, extracts the energy change characteristics of the frequency band meeting the stain adhesion precursor frequency band range as the stain adhesion precursor signal, the multi-signal fusion submodule adopts a bidirectional convolutional network with attention mechanism weighting to process the glass surface gray data, the equipment vibration frequency data and the visual recognition feature gradient data and outputs the data to the abnormality quantization submodule, the abnormality quantization submodule defines an abnormality index algorithm, and the estimation result display module generates a dynamic abnormality distribution map of the abnormality index based on a gradual color block, and uses color blocks of different colors to represent different abnormality indexes at different positions.

[0015] Preferably, the optimization unit processes the optimal multi-dimensional signal and the stain residual signal through a cleaning-stain nonlinear relationship model, outputs a cleaning signal, and processes the cleaning coverage signal through a particle swarm algorithm to obtain an optimal cleaning signal; the optimal multi-dimensional signal includes glass surface roughness and cleaning liquid flow signal, and the cleaning-stain nonlinear relationship model is constructed through the glass surface roughness, the stain residual signal and the cleaning liquid flow signal.

[0016] Preferably, the preferred unit further comprises a cleaning liquid characteristic matching library, a drilling path connectivity analysis submodule and a cleaning path planning submodule, the cleaning liquid characteristic matching library stores historical flow signals of historical cleaning liquids, the drilling path connectivity analysis submodule calculates a through coefficient of a drilling path according to the cleaning pressure and the coverage radius, and the cleaning path planning submodule obtains the optimal cleaning signal through a particle swarm algorithm according to the through coefficient, the cleaning-stain non-linear relationship model and the cleaning liquid flow signal.

[0017] Preferably, the regulation unit defines an accuracy constraint condition according to the regulation multi-dimensional signal, outputs a collaborative operation threshold value, and dynamically adjusts the collaborative operation threshold value to be not lower than a collaborative operation minimum value through PID control, the regulation multi-dimensional signal includes a real-time positioning deviation signal and a glass thickness detection signal, the real-time positioning deviation signal and the glass thickness detection signal define the accuracy constraint condition, and the PID control makes the collaborative operation threshold value not lower than the collaborative operation minimum value by dynamically adjusting the cleaning nozzle moving speed and the drilling equipment feeding step distance.

[0018] Preferably, the human-computer interaction and prompting module includes a two-dimensional monitoring platform and an abnormality prompting engine, the two-dimensional monitoring platform receives the multi-dimensional signal and the real-time monitoring signal and dynamically displays a stain cleaning process, and the abnormality prompting engine processes the drilling deviation risk signal and the stain residual signal according to an improved analytic hierarchy process and outputs a comprehensive abnormality index, when the comprehensive abnormality index exceeds an abnormality index threshold value, the human-computer interaction and prompting module controls the adaptive cleaning execution unit to perform emergency cleaning, the two-dimensional monitoring platform supports arbitrary angle area viewing, can display the spatial relationship between the drilling position and the cleaning area in real time, and simulates glass deformation and residual changes under different operation schemes based on the finite element method.

[0019] Preferably, the abnormality prompting engine includes a color state of a multi-level prompting mechanism, which are a yellow prompting state, an orange prompting state and a red prompting state, when the stain residual thickness in the stain residual signal rises to a first thickness threshold value, the yellow prompting state is started, and the operation speed of the operation equipment is reduced to a prompt operation speed, when the stain residual thickness in the stain residual signal rises to a second thickness threshold value, the operation of the operation equipment is paused and the adaptive cleaning execution unit is controlled to perform emergency cleaning, and when the stain residual thickness in the stain residual signal rises to a third thickness threshold value, the abnormality prompting engine sends an emergency fault signal and controls the glass processing intelligent cleaning and drilling collaborative system to stop urgently.

[0020] Preferably, the optimal cleaning signal includes but is not limited to cleaning area coordinates, cleaning fluid type priority, cleaning pressure range and coverage path planning, the adaptive cleaning execution unit includes an intelligent cleaning nozzle and an integrated cleaning fluid concentration online detection device, the intelligent cleaning nozzle can identify the optimal cleaning signal and adjust the cleaning flow and cleaning pressure as needed, the integrated cleaning fluid concentration online detection device receives the optimal cleaning signal and adjusts the mixing concentration of the cleaning fluid in real time based on the conductivity detection method, and the real-time monitoring signal includes but is not limited to device temperature, device humidity, device current, cleaning depth, drilling pressure and coverage radius.

[0021] Compared with the prior art, the beneficial effects of the present application are:

[0022] Through the organic combination of multiple modules, the cleaning and drilling processes are deeply linked, breaking the situation of mutual separation in the traditional processing mode. The multi-modal information acquisition module can integrate multi-dimensional signals to provide comprehensive raw data for subsequent strategy generation, allowing the system to grasp the real-time state of the glass from multiple angles, including surface stain distribution, material characteristics, device operating parameters, etc., avoiding the judgment bias that may be caused by a single information source.

[0023] The estimation unit in the strategy generation module outputs stain residue signals and drilling deviation risk signals in advance through analysis of multi-dimensional signals, allowing the system to make a pre-judgment before the problem occurs, and gaining time for subsequent parameter adjustment. The optimal unit determines the optimal cleaning signal based on multi-dimensional signals and stain residue signals to ensure that the cleaning operation can be adjusted according to the actual stain situation, avoiding the problems of insufficient cleaning or excessive cleaning that may occur in traditional fixed parameter cleaning. The precision constraints defined by the regulation unit and the output of the collaborative operation threshold provide a clear operating boundary for the linkage of cleaning and drilling equipment, allowing both to maintain consistent precision standards during the operation process, reducing quality problems caused by parameter mismatch.

[0024] The adaptive cleaning execution unit in the device linkage execution module dynamically adjusts the cleaning parameters according to the optimal cleaning signal, which can optimize the cleaning process for different stain types and distribution states, improve the uniformity and thoroughness of cleaning, and reduce the impact of stain residue on subsequent drilling processes. The drilling equipment group control unit adjusts the drilling parameters according to the collaborative operation threshold and multi-dimensional signals, and collects real-time monitoring signals, allowing the drilling equipment to flexibly change the operating parameters according to the actual state of the glass and the cleaning effect, reducing the possibility of drilling deviation, and dynamically controlling the drilling process through real-time monitoring. The human-computer interaction and prompting module converts the drilling deviation risk signal and the stain residue signal into a comprehensive abnormality index, providing intuitive quantitative information of abnormal conditions for the operator, facilitating his quick understanding of the severity and comprehensive impact of the problem, and reducing the subjectivity and error rate of manual judgment.

[0025] The system realizes intelligent management of the whole process from information collection, strategy generation to device execution through the cooperation of each module, so that the cleaning and drilling processes are no longer isolated links. The optimization of the cleaning effect can create more ideal surface conditions for drilling operation, and the real-time feedback of the drilling process can in turn guide the adjustment of the cleaning parameters, forming a virtuous cycle. This cooperative mechanism can effectively reduce the drilling deviation caused by incomplete cleaning and the glass damage caused by improper drilling parameters, while reducing the frequency and intensity of manual intervention, making the whole processing process more stable and efficient.

[0026] The output of the comprehensive anomaly index enables the operator to timely grasp the overall operation state of the system, quickly identify the source of the anomaly and take corresponding measures, shorten the fault handling time, and reduce the loss caused by downtime. The continuous analysis and processing of multi-dimensional signals by the system can continuously accumulate operation data, provide reference for subsequent process optimization, and make the glass processing process gradually develop towards more accurate and intelligent direction, adapting to the diversified needs of glass product quality in different scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a working principle diagram of the glass processing intelligent cleaning and drilling collaborative system described in the present application;

[0028] Figure 2 is a flowchart of data fusion of the multi-modal information collection module;

[0029] Figure 3 is a flowchart of the image analysis model working;

[0030] Figure 4 is a flowchart of the preferred unit cleaning signal generation;

[0031] Figure 5 is a flowchart of the regulation unit dynamic adjustment. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0033] Please refer to Figures 1-5 The present application provides a glass processing intelligent cleaning and drilling collaborative system, which comprises a multi-modal information collection module, a strategy generation module and a device linkage execution module. The specific implementation is as follows:

[0034] The multi-modal information acquisition module includes a data integration node unit that processes various types of signals collected to form multi-dimensional signals. The strategy generation module receives the multi-dimensional signals, and an estimation unit inside the module processes the multi-dimensional signals to output a stain residue signal and a drilling deviation risk signal. An optimization unit processes the multi-dimensional signals and the stain residue signal to obtain an optimal cleaning signal. A regulation unit defines precision constraints based on the multi-dimensional signals to output a collaborative operation threshold. An adaptive cleaning execution unit in the equipment linkage execution module adjusts cleaning parameters based on the optimal cleaning signal. A drilling equipment group control unit adjusts drilling equipment parameters based on the collaborative operation threshold and the multi-dimensional signals, and collects real-time monitoring signals. A human-machine interaction and prompting module processes the drilling deviation risk signal and the stain residue signal to output a comprehensive abnormality index.

[0035] Embodiment 1:

[0036] The multi-modal information acquisition module is composed of a distributed sensing array unit and a visual recognition detection unit, which covers key positions of the cleaning tank, drilling machine, and conveying track, forming a comprehensive signal acquisition network. The distributed sensing array unit includes temperature sensors, pressure sensors, vibration sensors, and liquid level sensors, among other types of sensing elements. The temperature sensors are evenly distributed on the inner wall and bottom of the cleaning tank to collect real-time temperature changes of the cleaning liquid. The pressure sensors are installed on the clamping device of the drilling machine and the support structure of the conveying track to record pressure fluctuations experienced by the glass during processing. The vibration sensors are fixed on the main shaft of the drilling machine and the side wall of the cleaning tank to capture vibration signals generated during equipment operation. The liquid level sensor is placed inside the cleaning tank to monitor changes in the liquid level of the cleaning liquid. These sensing elements convert the collected physical quantities into electrical signals, forming a continuous sensing signal stream.

[0037] The visual recognition detection unit is composed of high-definition industrial cameras, light source systems, and image acquisition cards. The high-definition industrial cameras are installed above the cleaning tank, on both sides of the drilling machine, and at the entrance and exit of the conveying track. Each camera is equipped with a corresponding ring light or strip light to ensure clear imaging of the glass surface at different processing stages. The cameras above the cleaning tank focus on the surface state of the glass before it enters the cleaning stage. The cameras on both sides of the drilling machine capture surface features of the drilling area from horizontal and 45-degree angles, respectively. The cameras at the entrance and exit of the conveying track record the overall surface conditions of the glass before and after it enters the processing flow. The image acquisition cards convert the analog image signals captured by the cameras into digital signals and perform preliminary processing on the images through edge computing technology. The characteristic data of the glass surface, such as stain distribution, scratches, and bubbles, are extracted to form surface signals.

[0038] The data integration node unit adopts an industrial-grade data processing chip and a special heterogeneous data fusion algorithm. The receiving end thereof is connected with the distributed sensing array unit and the visual recognition detection unit through a multi-channel signal interface, so as to realize the synchronous reception of the sensing signals and the surface signals. For the sensing signals, the data integration node unit first performs filtering processing to remove the noise generated by the environmental interference, and then aligns the signals of different sensors according to the time stamp to form a structured sensing data set. For the surface signals, the data integration node unit first performs compression and format conversion on the digital image data to retain the key feature information, and extracts the shooting time, position coordinates and other metadata of the image, and stores the metadata in association with the image feature data.

[0039] In the heterogeneous data fusion process, the data integration node unit first performs data consistency verification on the sensing signals and the surface signals, removes abnormal data by comparing whether there is a logical contradiction between different types of signals at the same time point and at the same position. Subsequently, a time series-based correlation analysis method is adopted to correlate the cleaning liquid temperature change with the dissolution state of the glass surface stains, correlate the vibration signal of the drilling machine table with the vibration trace on the glass surface, and correlate the pressure signal of the conveying track with the position deviation of the glass. Through such correlation analysis, a mapping relationship between different types of signals is established, and then a feature extraction algorithm is used to extract key features related to the glass processing state from the correlated data set, such as the rate of change of the cleaning liquid temperature with time, the frequency distribution of the drilling vibration, the area ratio of the glass surface stains, etc.

[0040] After the above processing, the data integration node unit forms multi-dimensional signals, wherein the estimated multi-dimensional signals include basic data related to subsequent stain residue and drilling deviation estimation, such as the initial surface features of the glass before entering the cleaning, the real-time temperature and vibration frequency of the cleaning liquid, etc.; the preferred multi-dimensional signals focus on data related to cleaning parameter optimization, such as the roughness distribution of the glass surface, the flow speed and direction of the cleaning liquid in the tank, etc.; and the regulated multi-dimensional signals cover information related to the processing precision control, such as the real-time position coordinates of the glass on the conveying track, the positioning error of the drilling machine table, etc.

[0041] The cooperative work of the distributed sensing array unit and the visual recognition detection unit realizes the comprehensive perception of the glass processing environment, the equipment state and the glass surface features, and the heterogeneous data fusion of the data integration node unit breaks the barriers between different types of signals, so that the multi-dimensional signals can fully reflect the complex state of the glass processing.

[0042] Example 2:

[0043] The estimation unit is composed of a convolutional neural network-based image analysis model and a fuzzy inference fusion model, and is used to process the estimated multidimensional signal in the multidimensional signal, and then output the stain residue signal and the drilling deviation risk signal. The estimated multidimensional signal includes glass surface gray data, equipment vibration frequency data, visual recognition feature gradient data, historical operation parameters and historical stain residue thickness data. After these data are acquired by the distributed sensing array unit and the visual recognition detection unit of the multi-modal information acquisition module, they are processed by the data integration node unit to form structured input data.

[0044] The image analysis model adopts a multi-layer convolutional neural network architecture, and its input layer receives glass surface gray data, historical operation parameters, equipment vibration frequency data, visual recognition feature gradient data and historical stain residue thickness data. The glass surface gray data comes from the gray-scale image taken by the visual recognition detection unit, and contains the light and dark information of different areas of the glass surface, which can reflect the distribution range and adhesion state of the stain; the historical operation parameters cover the setting parameters such as cleaning time, cleaning pressure and drilling speed in the past processing process; the equipment vibration frequency data is collected by vibration sensors installed on the drilling machine and cleaning equipment, and records the vibration intensity and frequency change of the equipment during operation; the visual recognition feature gradient data is obtained by edge detection on the glass surface image, which reflects the change rate of surface features; the historical stain residue thickness data is the actual thickness value of the stain residue measured after the past processing.

[0045] In the processing process of the image analysis model, first, the input data is standardized to adjust the numerical range of different types of data to the same interval, so as to avoid the influence of data magnitude difference on the model processing result. Subsequently, the glass surface gray data and the visual recognition feature gradient data are extracted by convolution layer, and the convolution kernel slides on the image to capture the gray change and gradient change features of the local area, and generates a feature map; for the equipment vibration frequency data and the historical operation parameters, the dimension conversion and feature mapping are performed by the fully connected layer, so that the image features are consistent in dimension; the historical stain residue thickness data is used as label information to participate in the training process of the model. After the alternating processing of multiple convolution layers and pooling layers, the model gradually extracts more abstract and representative deep features, and finally outputs the stain residue signal through the output layer. The signal represents the possible stain residue amount at different positions on the glass surface in numerical form.

[0046] The fuzzy inference fusion model is an improved structure, mainly used for processing glass surface gray data, equipment vibration frequency data and visual recognition feature gradient data, and outputting a drilling deviation risk signal. The input of the model is the preprocessed three kinds of data, among which the glass surface gray data is divided into multiple fuzzy subsets, such as "low gray", "medium gray" and "high gray", which correspond to different membership functions respectively; the equipment vibration frequency data is divided into "weak vibration", "medium vibration" and "strong vibration" according to the frequency range; and the visual recognition feature gradient data is divided into "small gradient", "medium gradient" and "large gradient" according to the gradient value.

[0047] The inference process of the improved fuzzy inference fusion model is based on a preset fuzzy rule base. The rules in the rule base are summarized from processing experience and historical data, such as "if the glass surface gray is high and the equipment vibration is strong, then the drilling deviation risk is high", "if the visual recognition feature gradient is small and the equipment vibration is weak, then the drilling deviation risk is low", and the like. After the input data is processed by the model, inference is performed according to the fuzzy rules, the triggering strength of each rule is calculated, and then the inference results of all rules are combined by a fuzzy synthesis algorithm to obtain a comprehensive fuzzy output. Finally, the fuzzy output is converted into a specific numerical value through defuzzification processing to form a drilling deviation risk signal, which reflects the possibility of drilling position deviation under the current processing conditions.

[0048] Through the above process, the pre-estimation unit simultaneously outputs a stain residue signal and a drilling deviation risk signal. The former provides a basis for subsequent cleaning parameter optimization, and the latter provides a reference for parameter adjustment and risk warning of the drilling equipment. Both signals are transmitted to other units of the strategy generation module in the form of real-time data stream to participate in the formulation of the collaborative operation strategy.

[0049] Embodiment 3:

[0050] The image analysis model is composed of an image feature extraction submodule, a multi-signal fusion submodule, an abnormality quantification submodule and a pre-estimation result display submodule. The image feature extraction submodule processes the equipment vibration frequency data, which is decomposed into signal components of different frequency bands by using a time-frequency domain decomposition method. In the decomposition process, the continuous vibration frequency data is first divided into multiple data segments according to a fixed time interval, each data segment corresponding to a time window, and then frequency analysis is performed on the data in each time window to obtain the frequency distribution in the window. Through a preset stain adhesion precursor frequency band range, the frequency bands that meet the range are selected from the decomposed signal components, the energy values of these frequency bands in different time windows are calculated, and then the energy change features are extracted and defined as stain adhesion precursor signals. The energy change features include the rising rate of energy value, the time interval of peak value occurrence and the concentration degree of energy distribution, etc., which can reflect the correlation between equipment vibration and stain adhesion.

[0051] The multi-signal fusion sub-module adopts an attention mechanism weighted bidirectional convolutional network to fuse the glass surface grayscale data, equipment vibration frequency data, and visual recognition feature gradient data. The input layer of the network receives the three types of data after standardization processing. The glass surface grayscale data is input in the form of a two-dimensional matrix, with each element representing the grayscale value at the corresponding position. The equipment vibration frequency data is input in the form of a one-dimensional time series, containing vibration frequency values at different time points. The visual recognition feature gradient data is also a two-dimensional matrix, with the element value representing the gradient size at the corresponding position. The bidirectional convolutional network is composed of a forward convolutional layer and a backward convolutional layer. The forward convolutional layer extracts features from the start to the end of the data, and the backward convolutional layer extracts features from the end to the start. Through this bidirectional processing method, the features associated with the front and back of the data can be captured. The attention mechanism calculates the weights of different data features in the network. Features that have a greater impact on the fusion result are given higher weights, and vice versa. The calculation of the weights is based on the correlation analysis between the features and the target output. After processing by the bidirectional convolutional layer and the attention mechanism, the three types of data are fused into a comprehensive feature vector, which is output to the anomaly quantification sub-module.

[0052] The anomaly quantification sub-module defines an anomaly index algorithm to convert the comprehensive feature vector into a specific anomaly index value. The anomaly index algorithm determines the anomaly index by calculating the deviation between the comprehensive feature vector and a preset normal feature vector, which is generated based on historical normal processing data. The calculation formula of the anomaly index is as follows:

[0053]

[0054] wherein, represents the anomaly index, represents the dimension of the feature vector, represents the weight of the th feature, represents the value of the th feature in the comprehensive feature vector, represents the value of the th feature in the normal feature vector. The larger the anomaly index value calculated by the formula, the higher the degree of abnormality of the current glass surface.

[0055] The estimated result display module generates a dynamic abnormality distribution map based on the abnormality index output by the abnormality quantification submodule. This module first divides the glass surface into multiple uniform grid areas, each of which corresponds to an abnormality index value. The dynamic abnormality distribution map uses gradient color blocks for visual display, with the color gradually transitioning from blue to red. Blue color blocks represent areas with lower abnormality index, green represents areas with medium abnormality index, and yellow and red represent areas with higher and extremely high abnormality index, respectively. The distribution map can be updated in real time, and as the processing process progresses, when new abnormality index data is generated, the corresponding color block color will change accordingly to reflect the real-time changes in the abnormality state of the glass surface. The operator can intuitively understand the abnormality of different positions on the glass surface through this distribution map, including the severity and distribution range of the abnormality.

[0056] The processing flow of the preferred unit works with the image analysis model, which processes the preferred multi-dimensional signal and the stain residue signal through the cleaning-stain nonlinear relationship model. The glass surface roughness data in the preferred multi-dimensional signal is obtained by a surface profiler, reflecting the concave-convex degree of the glass surface; the cleaning liquid flow signal is obtained by a flow sensor installed at the cleaning nozzle, containing flow rate and flow change information. The cleaning-stain nonlinear relationship model constructs a nonlinear mapping relationship between the stain residue and the cleaning parameters through these data, and outputs the preliminary cleaning signal. The particle swarm algorithm optimizes the cleaning path, cleaning pressure and other parameters in the cleaning signal, finds the best parameter combination that makes the cleaning effect best through the simulation of the search process of particles in the solution space, and forms the best cleaning signal. The historical flow signal stored in the cleaning liquid characteristic matching library contains the flow characteristics of different types of cleaning liquids at different temperatures and pressures. When calculating the penetration coefficient according to the cleaning pressure and the coverage radius, the drilling path connectivity analysis submodule considers the penetration ability and coverage range of the cleaning liquid in the drilling path, and the cleaning path planning submodule combines the penetration coefficient, the output of the cleaning-stain nonlinear relationship model and the cleaning liquid flow signal to plan the cleaning path that can cover all abnormal areas and has the highest efficiency through the particle swarm algorithm, ensuring the rationality and effectiveness of the best cleaning signal.

[0057] Example 4:

[0058] The core function of the regulation unit is to define precision constraints based on the regulation multidimensional signals, and then output the collaborative operation threshold. The regulation multidimensional signals mainly include real-time positioning offset signals and glass thickness detection signals, which are obtained through the distributed sensing array unit and the visual recognition detection unit in the multi-modal information acquisition module. The real-time positioning offset signals are collected by laser positioning sensors installed on both sides of the conveying track. The laser beams emitted by the sensors irradiate the glass edges. By calculating the position change of the reflected light beams, the offset of the glass relative to the preset path during the conveying process is determined. The offset is in millimeters and includes horizontal and vertical offset data. The glass thickness detection signals are generated by an ultrasonic thickness detector deployed at the entrance of the drilling machine. The detector emits ultrasonic waves to the glass surface. The actual thickness of the glass is calculated based on the reflection time of the sound waves. The thickness data is accurate to the micron level and covers the thickness values of different regions of the glass, including the thickness difference between the edge and the center region.

[0059] The definition of the precision constraint condition is based on the comprehensive analysis of the real-time positioning offset signals and the glass thickness detection signals. When the horizontal offset in the real-time positioning offset signals exceeds a certain range, it will directly affect the accuracy of the drilling position. Therefore, the maximum allowed value of the horizontal offset is included in the precision constraint condition. The vertical offset will cause the relative height of the glass and the drilling equipment to deviate if it is too large, so it is also listed as a constraint indicator. In the glass thickness detection signals, if the thickness difference between different regions is too large, the uneven stress during drilling may cause deviation, so the allowed range of the thickness difference also becomes part of the precision constraint condition. These constraint conditions collectively form the basis of the collaborative operation threshold, which is represented in numerical form and comprehensively reflects the minimum standard that can ensure the cleaning and drilling accuracy under the current processing environment.

[0060] The PID control is used in the system to dynamically adjust the collaborative working threshold, and the core is to correct the cleaning nozzle moving speed and the drilling equipment feeding step distance in real time through the feedback mechanism. The input of the PID control is the deviation between the current collaborative working threshold and the collaborative working minimum value, and the output is the specific adjustment amount. When it is monitored in real time that the collaborative working threshold is lower than the collaborative working minimum value, the PID control starts the adjustment process. For example, when the real-time positioning deviation signal shows that the glass increases in horizontal deviation, causing the collaborative working threshold to decrease, the control system will first adjust the moving speed of the cleaning nozzle. If the deviation is small, the moving speed is appropriately reduced, so that the cleaning process has more sufficient time to adapt to the glass position change; if the deviation continues to increase, the speed is further slowed down, and a signal is sent to the drilling equipment group control unit to adjust the feeding step distance of the drilling equipment. The feeding step distance of the drilling equipment refers to the distance of the drilling spindle each time, and the step distance adjustment is carried out according to the glass thickness detection signal. When it is detected that the thickness of a certain area of the glass is large, the feeding step distance is reduced, so that the drilling process is more stable; when the thickness is small, the step distance is appropriately increased to ensure the processing efficiency.

[0061] In actual operation, the adjustment process of the PID control is a continuous closed-loop feedback process. For example, when the glass produces a 1.5 mm horizontal deviation on the conveying track due to uneven force, the real-time positioning deviation signal transmits this data to the control unit, the control unit calculates that the current collaborative working threshold is lower than the preset minimum value, and immediately starts the PID control. At this time, the proportional link in the PID controller directly outputs the adjustment amount according to the deviation size, and reduces the moving speed of the cleaning nozzle from the original 50 mm / s to 40 mm / s; the integral link accumulates the duration of the deviation, and if the deviation state continues for more than 2 seconds, the speed is further reduced to 35 mm / s, and at the same time, the feeding step distance of the drilling equipment is adjusted from 0.1 mm to 0.08 mm; the differential link predicts the trend of the deviation, and if it is detected that the deviation continues to increase, an early warning signal is sent to the drilling equipment in advance, so that it is prepared for further adjustment. Through these adjustments, the collaborative working threshold gradually rises until it is above the collaborative working minimum value, at which time the PID control stops adjusting and maintains the current parameters.

[0062] After receiving the coordinated operation threshold and multi-dimensional signals output by the regulation unit, the drilling equipment group control unit will adjust its parameters synchronously. For example, when the coordinated operation threshold is adjusted due to changes in glass thickness, the rotation speed of the drilling equipment will also change accordingly. The rotation speed is appropriately reduced when drilling in areas with larger thickness to avoid excessive equipment load; the rotation speed is increased when drilling in areas with smaller thickness to ensure processing progress. At the same time, the pressure sensor on the drilling equipment monitors the drilling pressure in real time and feeds back the data to the regulation unit as an auxiliary basis for PID control adjustment. If the drilling pressure abnormally rises, it may indicate that there is stress unevenness inside the glass. The regulation unit will further optimize the coordinated operation threshold in combination with this signal to ensure precision control during the entire processing process.

[0063] During the entire regulation process, real-time positioning offset signals and glass thickness detection signals are continuously collected and transmitted. The regulation unit analyzes these signals every 10 milliseconds, updates the coordinated operation threshold, and starts PID control as needed. This high-frequency dynamic adjustment can respond to various subtle changes in the processing process in a timely manner, so that cleaning and drilling operations can always be carried out within the allowed precision range. Through the coordinated adjustment of cleaning nozzle moving speed and drilling equipment feed step, seamless cooperation of the two operations is achieved, and processing deviations caused by mismatched equipment parameters are reduced.

[0064] Example 5:

[0065] The human-computer interaction and prompt module is composed of a two-dimensional monitoring platform and an abnormality prompt engine, which work together to realize real-time monitoring and abnormality processing of the glass processing process. The two-dimensional monitoring platform receives multi-dimensional signals and real-time monitoring signals through a data interface. These signals cover information such as glass surface features, equipment operating parameters, real-time status of cleaning and drilling, etc. The interface of the platform adopts a layered design. The bottom layer displays the movement trajectory of the glass on the conveying track, the middle layer superimposes dynamic identifiers of the cleaning area and the drilling position, and the upper layer displays real-time values of equipment operating parameters in a semi-transparent manner. The operator can rotate the viewing angle by mouse dragging or touch operation to view the spatial relationship between cleaning and drilling from any angle, such as observing from the overhead angle whether the cleaning area completely covers the glass surface before drilling, and confirming from the side angle whether the drilling depth matches the penetration depth of the cleaning liquid.

[0066] The simulation function based on the finite element method is integrated in the two-dimensional monitoring platform. Different operation scheme parameters such as cleaning pressure, drilling speed, conveying speed, etc. can be loaded, and the deformation and residual changes of the glass under different schemes are simulated through numerical calculation. During the simulation process, the glass is divided into multiple small units, and the deformation degree of each unit is calculated according to the material properties and stress conditions. The deformation result is displayed in a color gradient manner, with red areas representing larger deformation parts and blue areas representing smaller deformation parts. At the same time, the simulation of stain residue is based on the cleaning fluid flow path and the characteristics of the glass surface. By calculating the action time and contact area of the cleaning fluid and the stain, the residual amount in different areas is predicted, and the residual amount is also displayed in color for easy comparison of the effects of different schemes by the operator.

[0067] The core of the abnormality prompt engine is the improved analytic hierarchy process, which processes the drilling deviation risk signal and the stain residue signal and outputs a comprehensive abnormality index. During the processing, the two signals are first decomposed into multiple evaluation indicators. The drilling deviation risk signal is decomposed into equipment vibration amplitude, positioning offset, drill bit wear degree, etc. The stain residue signal is decomposed into stain area ratio, residual thickness, stain type, etc. Each indicator is assigned a corresponding weight according to its impact on the processing quality, and the weight value is determined through statistical analysis of historical processing data. The improved analytic hierarchy process integrates these indicators into a comprehensive abnormality index through weighted summation. The index value ranges from 0 to 100, and the higher the value, the more serious the abnormality.

[0068] The multi-level prompt mechanism of the abnormality prompt engine visually displays the abnormality degree through color states. When the yellow prompt state is started, the border of the system interface and the related parameter values turn yellow, and the device control module automatically reduces the operating speed of the processing equipment, for example, the conveying track speed is reduced from the original 1 meter / second to 0.7 meters / second, and the drilling speed is reduced from 3000 revolutions / minute to 2500 revolutions / minute, leaving time for possible adjustments. When the orange prompt state is triggered, the processing equipment is suspended, and the adaptive cleaning execution unit starts the emergency cleaning process. The cleaning nozzle performs comprehensive flushing on the glass surface according to the preset emergency path, the flushing pressure is increased by 20% compared with the regular cleaning, and the cleaning fluid temperature is increased by 5°C to enhance the cleaning effect. The red prompt state corresponds to the most serious abnormality, at this time the abnormality prompt engine issues a continuous sound and light alarm, the system emergency stops, all equipment stops running, the conveying track is locked, the drilling main shaft rises to a safe position, and the human-machine interface displays fault positioning and possible cause analysis, such as "stain residue thickness in the drilling area exceeds the standard" "equipment vibration is abnormal" etc.

[0069] The specific parameters of the optimal cleaning signal are transmitted to the adaptive cleaning execution unit through the equipment linkage execution module. The cleaning area coordinates are given in three-dimensional coordinates, accurate to the millimeter level, ensuring that the cleaning nozzle can accurately aim at the area to be processed. The cleaning liquid type priority is determined according to the glass material and the type of stains, for example, for oil stains, the priority of water-based cleaning agent is higher than that of solvent-based cleaning agent; for mineral residues, the priority of acidic cleaning agent is higher. The cleaning pressure range is set according to the glass thickness, when the thickness is less than 3mm, the pressure range is controlled at 0.2-0.3MPa, when the thickness is greater than 5mm, the pressure range is adjusted to 0.3-0.4MPa. The coverage path planning adopts spiral or reciprocating trajectory, ensuring that the cleaning liquid can uniformly cover the target area, the path interval is determined according to the nozzle diameter, usually 80% of the nozzle diameter, to avoid blind areas.

[0070] The intelligent cleaning nozzle of the adaptive cleaning execution unit is built-in signal recognition module, which can analyze the parameter instructions in the optimal cleaning signal, adjust the moving speed and angle of the nozzle through the stepping motor, and control the cleaning flow through the electromagnetic valve, the flow regulation accuracy reaches 1 liter / minute. The integrated cleaning liquid concentration online detection device monitors the conductivity value of the cleaning liquid in real time through the conductivity sensor, according to the deviation of the value and the standard concentration conductivity, automatically adjusts the proportion valve of the concentrated liquid and water, for example, when the conductivity is detected to be lower than the standard value, the addition amount of concentrated liquid is increased, otherwise the water amount is increased, the concentration adjustment response time is within 5 seconds.

[0071] Real-time monitoring signals are continuously collected by sensors distributed in various parts of the equipment. The temperature of the equipment is monitored by thermocouple sensors installed on the main shaft of the drilling machine and the cleaning pump body, and the temperature data is updated every 2 seconds. The humidity of the equipment is collected by a humidity sensor above the cleaning tank, which is used to determine whether the cleaning environment meets the requirements. The current of the equipment is monitored by a current transformer in the main circuit, which reflects the load condition of the equipment. The cleaning depth is recorded by a displacement sensor at the cleaning nozzle, accurate to 0.1 millimeter. The drilling pressure is measured by a pressure sensor on the drill bit, which provides real-time feedback on the stress during drilling. The coverage radius is analyzed by the cleaning liquid spray image taken by the visual recognition detection unit, ensuring that the actual coverage range is consistent with the planned path. These real-time monitoring signals are transmitted to the man-machine interaction and prompt module and the strategy generation module through the data bus, forming a feedback link of closed-loop control.

[0072] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0073] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A glass processing intelligent cleaning and drilling collaborative system, characterized in that, The application relates to a multi-modal information acquisition module, a strategy generation module and a device linkage execution module. The multi-modal information acquisition module comprises a data integration node unit, which forms a multi-dimensional signal. The strategy generation module receives the multi-dimensional signal, and comprises a prediction unit, an optimization unit and a regulation unit. The prediction unit processes the multi-dimensional signal to output a stain residue signal and a drilling deviation risk signal.

2. The glass processing intelligent cleaning and drilling collaborative system of claim 1, wherein, The optimization unit processes the multi-dimensional signal and the stain residue signal to obtain an optimal cleaning signal.

3. The glass processing intelligent cleaning and drilling collaborative system of claim 2, wherein, The regulation unit defines an accuracy constraint condition according to the multi-dimensional signal to output a collaborative operation threshold. The device linkage execution module comprises an adaptive cleaning execution unit, a drilling device group control unit and a man-machine interaction and prompting module. The adaptive cleaning execution unit adjusts cleaning parameters according to the optimal cleaning signal. The drilling device group control unit adjusts drilling device parameters according to the collaborative operation threshold and the multi-dimensional signal, and collects real-time monitoring signals. The man-machine interaction and prompting module processes the drilling deviation risk signal and the stain residue signal and outputs a comprehensive abnormality index. The multi-modal information acquisition module comprises a distributed sensing array unit and a visual recognition detection unit. The multi-modal information acquisition module is arranged on a cleaning tank, a drilling machine and a conveying track, and collects sensing signals. The visual recognition detection unit obtains glass surface feature data through image acquisition technology and forms a surface signal. The data integration node unit receives the sensing signals and the surface signal to realize heterogeneous data fusion and form the multi-dimensional signal. The prediction unit comprises a convolutional neural network-based image analysis model and a fuzzy inference fusion model. The prediction multi-dimensional signal comprises glass surface grayscale data, device vibration frequency data, visual recognition feature gradient data, historical operation parameters and historical stain residue thickness data. The glass surface grayscale data, the historical operation parameters, the device vibration frequency data, the visual recognition feature gradient data and the historical stain residue thickness data are input into the image analysis model to output the stain residue signal. The glass surface grayscale data, the device vibration frequency data and the visual recognition feature gradient data are fused through an improved fuzzy inference fusion model to output the drilling deviation risk signal.

4. The glass processing intelligent cleaning and drilling collaborative system of claim 3, wherein, The image analysis model comprises an image feature extraction submodule, a multi-signal fusion submodule, an abnormality quantification submodule, and a prediction result display module. The image feature extraction submodule performs time-frequency domain decomposition on the equipment vibration frequency data, extracts the energy change characteristics of the frequency band that meets the frequency band range of the stain adhesion precursor as the stain adhesion precursor signal, the multi-signal fusion submodule adopts a bidirectional convolution network with attention mechanism weighting to process the glass surface gray data, the equipment vibration frequency data, and the visual recognition feature gradient data and outputs the data to the abnormality quantification submodule, the abnormality quantification submodule defines an abnormality index algorithm, and the prediction result display module generates a dynamic abnormality distribution map of the abnormality index based on a gradient color block and uses color blocks of different colors to represent different abnormality indexes at different positions.

5. The glass processing intelligent cleaning and drilling collaborative system of claim 2, wherein, The preferred unit processes the preferred multi-dimensional signal and the stain residue signal through a cleaning-stain non-linear relationship model, outputs a cleaning signal, and processes the cleaning coverage signal using a particle swarm algorithm to obtain an optimal cleaning signal; the preferred multi-dimensional signal includes glass surface roughness and cleaning fluid flow signal, and the cleaning-stain non-linear relationship model is constructed by the glass surface roughness, the stain residue signal, and the cleaning fluid flow signal.

6. The glass processing intelligent cleaning and drilling collaborative system of claim 5, wherein, The preferred unit also includes a cleaning fluid characteristic matching library, a drilling path connectivity analysis submodule, and a cleaning path planning submodule. The cleaning fluid characteristic matching library stores historical flow signals of historical cleaning fluids, the drilling path connectivity analysis submodule calculates a through coefficient of a drilling path according to the cleaning pressure and the coverage radius, and the cleaning path planning submodule obtains the optimal cleaning signal through a particle swarm algorithm according to the through coefficient, the cleaning-stain non-linear relationship model, and the cleaning fluid flow signal.

7. The glass processing intelligent cleaning and drilling collaborative system of claim 2, wherein, The regulation unit defines an accuracy constraint condition according to the regulation multi-dimensional signal, outputs a collaborative operation threshold, and dynamically adjusts the collaborative operation threshold to be not lower than a collaborative operation minimum value through PID control, the regulation multi-dimensional signal includes a real-time positioning offset signal and a glass thickness detection signal, the real-time positioning offset signal and the glass thickness detection signal define the accuracy constraint condition, and the PID control dynamically adjusts the cleaning nozzle moving speed and the drilling equipment feeding step distance to make the collaborative operation threshold not lower than the collaborative operation minimum value.

8. The glass processing intelligent cleaning and drilling collaborative system of claim 1, wherein, The human-computer interaction and prompting module includes a two-dimensional monitoring platform and an abnormality prompting engine. The two-dimensional monitoring platform receives the multi-dimensional signal and the real-time monitoring signal and dynamically displays the stain cleaning process, and the abnormality prompting engine processes the drilling deviation risk signal and the stain residue signal according to an improved analytic hierarchy process and outputs a comprehensive abnormality index. When the comprehensive abnormality index exceeds an abnormality index threshold, the human-computer interaction and prompting module controls the adaptive cleaning execution unit to perform emergency cleaning. The two-dimensional monitoring platform supports viewing at any angle and can display the spatial relationship between the drilling position and the cleaning area in real time, and simulate glass deformation and residual changes under different operation schemes based on the finite element method.

9. The glass processing intelligent cleaning and drilling collaborative system of claim 8, wherein, The abnormality prompt engine includes a multi-level prompt mechanism color state, which is a yellow prompt state, an orange prompt state and a red prompt state respectively. When the stain residue thickness in the stain residue signal rises to a first thickness threshold, the yellow prompt state is started, and the operating speed of the working equipment is reduced to a prompt operating speed. When the stain residue thickness in the stain residue signal rises to a second thickness threshold, the operation of the working equipment is suspended and the self-adaptive cleaning execution unit is controlled to perform emergency cleaning. When the stain residue thickness in the stain residue signal rises to a third thickness threshold, the abnormality prompt engine sends an emergency fault signal and controls the glass processing intelligent cleaning and drilling cooperative system to stop urgently.

10. The glass processing intelligent cleaning and drilling collaborative system of claim 1, wherein, The optimal cleaning signal includes but is not limited to cleaning area coordinates, cleaning fluid type priority, cleaning pressure range and coverage path planning. The self-adaptive cleaning execution unit includes an intelligent cleaning nozzle and an integrated cleaning fluid concentration online detection device. The intelligent cleaning nozzle can identify the optimal cleaning signal and adjust the cleaning flow and cleaning pressure as needed. The integrated cleaning fluid concentration online detection device receives the optimal cleaning signal and adjusts the matching concentration of the cleaning fluid in real time based on the conductivity detection method. The real-time monitoring signal includes but is not limited to equipment temperature, equipment humidity, equipment current, cleaning depth, drilling pressure and coverage radius.

Images