Precise dispensing nozzle frictioning control method and system based on fault real-time detection

Abstract

The invention relates to the technical field of intelligent control, and discloses a precise dispensing nozzle frictioning control method and system based on fault real-time detection.The method comprises the steps that started dispensing equipment conducts automatic alcohol spraying and frictioning clamping on a glue spraying nozzle, and multi-source state data of non-woven fabric is obtained; performing cross-modal feature fusion on the multi-source state data to generate a fusion feature map of the non-woven fabric; performing fault classification on the fused feature map to obtain a real-time fault type and a fault confidence coefficient of the dispensing equipment; dynamically generating a load control instruction set according to the real-time fault type, the fault confidence coefficient and the current process parameters of the dispensing equipment; according to the load control instruction set, driving the corresponding process node to carry out frictioning operation; after the frictioning operation is completed, performing fault re-verification on the state data of the dispensing equipment, and if the fault confidence is lower than a set threshold value, outputting a fault state and a frictioning completion state; according to the invention, the precision dispensing nozzle frictioning control efficiency based on fault real-time detection can be improved.

Description

A method and system for controlling the caking of precision dispensing nozzles based on real-time fault detection. Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a precision dispensing nozzle wiping control method and system based on real-time fault detection. Background Technology

[0002] In precision dispensing operations, the cleaning effect of the dispensing nozzle directly affects dispensing accuracy and product quality. However, existing adhesive removal control methods have significant limitations in fault detection. They rely heavily on single-type status data for fault diagnosis, failing to achieve deep integration of multi-source heterogeneous data. This results in incomplete and inaccurate capture of abnormal characteristics during equipment operation, difficulty in quickly identifying potential fault types, and a lack of scientific rigor in fault confidence assessment. Consequently, adhesive removal operations cannot effectively avoid the impact of faults, easily leading to problems such as dispensing deviation and adhesive residue.

[0003] Existing adhesive application control strategies lack dynamic adaptability and fail to flexibly adjust based on real-time fault conditions and process parameters. Their preset adhesive application sequence is fixed, lacking a parameter optimization mechanism based on fault confidence and process compensation logic for different adhesive properties and dispensing speeds. This results in insufficient matching between adhesive application intensity and duration. This not only leads to incomplete application and a high risk of residual faults, but also risks damaging equipment or wasting resources due to excessive adhesive application, significantly reducing the operating efficiency of the dispensing equipment and the stability of the production process. Summary of the Invention

[0004] This invention provides a precision dispensing nozzle wiping control method and system based on real-time fault detection to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a precision dispensing nozzle wiping control method based on real-time fault detection, comprising: S1, after startup, the dispensing equipment automatically sprays alcohol and clamps the nozzle for wiping, and acquires multi-source state data of the nonwoven fabric after the operation is completed; S2, cross-modal feature fusion is performed on the multi-source state data to generate a fused feature map of the nonwoven fabric; S3, fault classification is performed on the fused feature map to obtain the real-time fault type and fault confidence of the dispensing equipment; S4, a load control instruction set is dynamically generated based on the real-time fault type, the fault confidence, and the current process parameters of the dispensing equipment; S5, the corresponding process node is driven to perform wiping operation according to the load control instruction set; S6, after the wiping operation is completed, the state data of the dispensing equipment is re-verified for faults, and if the fault confidence is lower than a set threshold, the fault status and wiping completion status are output.

[0006] In a preferred embodiment, the dispensing device, after startup, automatically sprays alcohol and clamps the nozzle, and acquires multi-source state data of the nonwoven fabric after the operation is completed. This includes: responding to the startup completion signal of the dispensing device, sending spray duration and atomization pressure parameters to the precision nozzle assembly; the clamping mechanism performing a clamping action to fix the nonwoven fabric to the cleaning station of the precision dispensing nozzle; acquiring a surface morphology image of the nonwoven fabric under the combined action of alcohol wetting and mechanical clamping, as the first type of state data; real-time clamping force and displacement feedback data during the clamping action, as the second type of state data; and combining the first type of state data and the second type of state data into multi-source state data.

[0007] In a preferred embodiment, the step of performing cross-modal feature fusion on the multi-source state data to generate a fused feature map of the nonwoven fabric includes: extracting feature points from the second type of state data to obtain the pressure temporal features of the dispensing device; performing real-time detection scale normalization on the nonwoven fabric running length of the first type of state data to obtain the standardized running length of the nonwoven fabric; performing motion state detection on the standardized running length to obtain the spatial visual features of the nonwoven fabric; projecting the pressure temporal features and the spatial visual features into the same high-dimensional semantic space and completing temporal alignment; and performing feature fusion on the aligned multimodal feature set to obtain the fused feature map of the nonwoven fabric.

[0008] In a preferred embodiment, the step of performing feature fusion on the aligned multimodal feature set, and obtaining the fused feature map of the nonwoven fabric, includes: assigning a learnable attention weight coefficient to each modal feature in the aligned multimodal feature set; weighting and superimposing the corresponding modal features according to the attention weight coefficient; and performing nonlinear activation on the weighted and superimposed fusion result to obtain the fused feature map of the nonwoven fabric.

[0009] In a preferred embodiment, the step of classifying the fused feature map to obtain the real-time fault type and fault confidence of the dispensing equipment includes: retrieving multiple fault feature templates from a pre-stored fault feature template library, each fault feature template being associated with a preset fault category; comparing the fused feature map with each fault feature template element by element to obtain feature similarity; selecting the fault feature template with the highest similarity based on the feature similarity to obtain the real-time fault type corresponding to the fused feature map; and using the highest feature similarity value corresponding to the selected fault feature template as the fault confidence.

[0010] In a preferred embodiment, the step of comparing the fused feature map with each fault feature template element-by-element to obtain feature similarity includes: flattening the fused feature map to obtain a first feature vector of the dispensing device; flattening each fault feature template in the fault feature template library to obtain a second feature vector corresponding to the fault feature template; calculating the vector similarity between the second feature vector and the first feature vector to obtain an initial similarity value of the dispensing device; and normalizing the initial similarity value to obtain the feature similarity of the dispensing device.

[0011] In a preferred embodiment, the initial similarity value is calculated using the following formula:

[0012] In the formula, The initial similarity value, The total number of feature dimensions. The ordinal number of the feature dimension. For the first Discrimination weight coefficients for each feature dimension The first feature vector is the first feature vector. Each feature dimension The second feature vector is the first Each feature dimension.

[0013] In a preferred embodiment, the step of dynamically generating a load control instruction set based on the real-time fault type, the fault confidence level, and the current process parameters of the dispensing equipment includes: querying the process parameters bound to the current nonwoven fabric dispensing task; retrieving the corresponding basic wiping action sequence from a predefined wiping strategy mapping table based on the real-time fault type; dynamically adjusting the action intensity and action duration parameters in the basic wiping action sequence based on the fault confidence level to obtain an adaptive wiping action sequence; and performing process compensation correction on the parameters in the adaptive wiping action sequence by combining the adhesive viscosity grade and dispensing speed setpoint in the process parameters to obtain the wiping control instruction set of the dispensing equipment.

[0014] In a preferred embodiment, the step of re-verifying the status data of the dispensing equipment after the glue application operation is completed, and outputting a fault status and a glue application completion status if the fault confidence level is lower than a set threshold, includes: immediately collecting verification status data of the dispensing equipment for at least one complete glue application cycle after the glue application actuator is reset; extracting features from the verification status data to obtain a verification feature vector; comparing the verification feature vector with a fault feature template corresponding to the real-time fault type in the fault feature template library to obtain the feature difference degree of the dispensing equipment; updating the fault confidence level based on the feature difference degree to obtain an updated fault confidence level; determining whether the updated fault confidence level is lower than a set threshold, and if so, outputting a glue application completion status; otherwise, triggering a fault alarm and generating a fault log and a fault status.

[0015] To address the aforementioned issues, this invention also provides a precision dispensing nozzle wiping control system based on real-time fault detection. The system includes: a human-machine interface module for automatically spraying alcohol and clamping the nozzle after startup, and acquiring multi-source state data of the nonwoven fabric after the operation is completed; a wiping process execution module for performing cross-modal feature fusion on the multi-source state data to generate a fused feature map of the nonwoven fabric; a fault detection module for classifying faults in the fused feature map to obtain the real-time fault type and fault confidence of the dispensing equipment; a load control module for dynamically generating a load control instruction set based on the real-time fault type, the fault confidence, and the current process parameters of the dispensing equipment; a control signal module for driving the corresponding process node to perform wiping operations according to the load control instruction set; and a communication module for re-verifying the fault status data of the dispensing equipment after the wiping operation is completed, and outputting the fault status and wiping completion status if the fault confidence is lower than a set threshold.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention collects multi-source state data such as non-woven fabric surface morphology images, clamping force and displacement feedback, and performs cross-modal feature fusion and attention weight allocation, combined with fault feature template element-by-element comparison, to achieve real-time and accurate determination of fault type and confidence, greatly improving the comprehensiveness and accuracy of fault detection and reducing misjudgment and missed judgment.

[0017] 2. This invention dynamically generates load control commands based on real-time fault information, confidence level, and process parameters, and adjusts the intensity and duration of the adhesive application action accordingly. Combined with a post-operation fault re-verification mechanism, it avoids excessive adhesive application or incomplete cleaning, and can promptly confirm the fault elimination status, significantly improving the cleaning quality of the dispensing nozzle and the operational stability of the equipment, and ensuring dispensing accuracy. Attached Figure Description

[0018] Figure 1 is a flowchart illustrating a precision dispensing nozzle wiping control method based on real-time fault detection according to an embodiment of the present invention; Figure 2 is a functional block diagram illustrating a precision dispensing nozzle wiping control system based on real-time fault detection according to an embodiment of the present invention; Figure 3 is a flowchart illustrating the human-machine interaction module in a precision dispensing nozzle wiping control system based on real-time fault detection according to an embodiment of the present invention; Figure 4 is a flowchart illustrating the communication module in a precision dispensing nozzle wiping control system based on real-time fault detection according to an embodiment of the present invention; Figure 5 is a flowchart illustrating the wiping process execution module in a precision dispensing nozzle wiping control system based on real-time fault detection according to an embodiment of the present invention; Figure 6 is a flowchart illustrating the fault detection module in a precision dispensing nozzle wiping control system based on real-time fault detection according to an embodiment of the present invention; Figure 7 is a flowchart illustrating the load control module in a precision dispensing nozzle wiping control system based on real-time fault detection according to an embodiment of the present invention; Figure 8 is a flowchart illustrating the control signal module in a precision dispensing nozzle wiping control system based on real-time fault detection according to an embodiment of the present invention. The realization of the purpose, functional characteristics, and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] This application provides a precision dispensing nozzle cleaning control method based on real-time fault detection. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the precision dispensing nozzle cleaning control method based on real-time fault detection can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0021] Referring to FIG. 1, it is a schematic flowchart of a precision dispensing nozzle rubber wiping control method based on real-time fault detection provided by an embodiment of the present invention. In this embodiment, the precision dispensing nozzle rubber wiping control method based on real-time fault detection includes: S1. After startup, the dispensing device automatically sprays alcohol and clamps the rubber wiping on the dispensing nozzle, and obtains multi-source state data of the non-woven fabric after the operation is completed; in the embodiment of the present invention, the dispensing device after startup automatically sprays alcohol and clamps the rubber wiping on the dispensing nozzle, and obtains multi-source state data of the non-woven fabric after the operation is completed, including: responding to the startup completion signal of the dispensing device, sending the spray duration and atomization pressure parameters to the precision nozzle assembly; the clamping mechanism performs a rubber wiping clamping action to fix the non-woven fabric at the cleaning station of the precision dispensing nozzle; collecting the surface topography image of the non-woven fabric under the combined action of alcohol wetting and mechanical clamping as the first type of state data; the real-time clamping force and displacement feedback data during the rubber wiping clamping action as the second type of state data; and汇集 the first type of state data and the second type of state data into multi-source state data.

[0022] After the dispensing device completes all startup self-check processes, its control module generates a continuous and stable startup completion electrical signal, which is directly transmitted to the control unit supporting this process through a dedicated shielded signal transmission line inside the device. The control unit internally pre-stores a set of spray duration and atomization pressure parameters corresponding to each type of dispensing process one by one. After receiving the startup completion electrical signal, the control unit automatically identifies the type of dispensing process currently being executed, accurately retrieves the corresponding set of parameters, and then the control unit sends these two sets of parameters to the parameter receiving module自带 by the precision nozzle assembly in the form of electrical signals byte by byte through a dedicated wired transmission link. After receiving the parameters, the parameter receiving module immediately starts an integrity verification program, checks the parameter information bit by bit, and after confirming that the parameters are complete and error-free, stores the parameters in its internal non-volatile dedicated storage area, waiting for subsequent spray processes to call.

[0023] It should be noted that there is an unclear expression "汇集" in the translation of . It may need to be further clarified according to the accurate meaning in the original text for a more precise translation.The clamping mechanism consists of a fixed clamp, a movable clamp, and a linear drive component that drives the movable clamp. The fixed clamp is bolted to one side of the precision dispensing nozzle cleaning station, and the movable clamp is correspondingly installed on the other side of the cleaning station. The linear drive component is rigidly connected to the tail of the movable clamp. When the control unit issues a clamping command, the linear drive component receives the command and starts working, pushing the movable clamp towards the fixed clamp at a constant speed. At this time, the non-woven fabric has been pre-laid flat on the anti-slip bearing surface of the cleaning station, and the edge of the non-woven fabric is in the preset clamping area between the fixed clamp and the movable clamp. The movable clamp continues to move until it contacts the edge of the non-woven fabric and applies a stable clamping force until the non-woven fabric is completely attached to the bearing surface of the cleaning station and its planar position is not offset. At this time, the positioning detection element of the linear drive component triggers a stop signal, the linear drive component stops power output, maintains the current clamping state, and completes the fixing operation of the non-woven fabric in the cleaning station.

[0024] A micro-drip device is fixedly installed next to the cleaning station using a bracket. The device's outlet points vertically downwards, directly onto the non-woven fabric area to be tested. A valve structure at the outlet provides precise volume control. Once the non-woven fabric is fixed, the control unit of the micro-drip device receives a start signal, and the valve slowly opens, gradually dripping alcohol onto the fabric surface. The alcohol naturally permeates and diffuses through the fibers of the non-woven fabric until the entire area to be tested is completely and evenly wetted. Throughout this process, the clamping mechanism maintains a stable clamping state without any loosening or displacement. Meanwhile, at the cleaning station... A high-definition industrial camera is fixedly mounted on the top via a lifting bracket. The camera lens is vertically downward aimed at the non-woven fabric to be inspected. The camera is pre-adjusted with a fixed shooting angle and focal length to ensure that every detail of the area to be inspected can be clearly imaged. When the non-woven fabric is fully wetted, the wetting detection element sends a shooting trigger signal, and the high-definition industrial camera immediately takes a high-definition picture. The image of the non-woven fabric surface morphology is the first type of state data. After the picture is taken, the image is directly transmitted to the image-dedicated storage area of ​​the data acquisition unit through a dedicated high-speed image transmission line.

[0025] A force-sensitive sensor is embedded in the clamping surface of the movable gripper of the clamping mechanism. The detection surface of the sensor is flush with the clamping surface of the movable gripper, so it will not affect the clamping effect. At the same time, a displacement sensor is installed at the power output end of the linear drive component. The force-sensitive sensor detects the clamping force applied by the movable gripper to the edge of the non-woven fabric in real time. The detected clamping force analog signal is converted into a digital electrical signal in real time. The displacement sensor detects the movement displacement of the movable gripper from the initial position to the clamping position in real time. The detected displacement feedback analog signal is also converted into a digital electrical signal in real time. The signal transmission lines of the two sensors are directly connected to the data acquisition unit. The clamping force data and displacement feedback data are transmitted synchronously to the data acquisition unit's dedicated numerical storage area in the same time sequence. These two sets of data together constitute the second type of status data.

[0026] The data acquisition unit is internally divided into two independent storage areas: a dedicated image storage area and a dedicated numerical storage area. The image information of the first type of state data is stored in the dedicated image storage area, while the clamping force and displacement numerical information of the second type of state data is stored in the dedicated numerical storage area. After receiving the two types of data, the data acquisition unit will integrate the image information of the first type of state data with the clamping force and displacement numerical information of the second type of state data according to a preset fixed data format. During the integration process, a unique association identifier will be added to each set of integrated data to ensure that each nonwoven fabric surface morphology image can correspond one-to-one with the corresponding clamping force and displacement values ​​without any confusion. The complete dataset containing the two types of data after integration is the multi-source state data. The multi-source state data will be uniformly stored in the integrated dedicated storage area of ​​the data acquisition unit, waiting to be retrieved and used in subsequent processes.

[0027] S2. Perform cross-modal feature fusion on the multi-source state data to generate a fused feature map of the nonwoven fabric; In this embodiment of the invention, the step of performing cross-modal feature fusion on the multi-source state data to generate a fused feature map of the nonwoven fabric includes: extracting feature points from the second type of state data to obtain the pressure time-series features of the dispensing device; performing real-time detection scale normalization on the nonwoven fabric running length of the first type of state data to obtain the standardized running length of the nonwoven fabric; performing motion state detection on the standardized running length to obtain the spatial visual features of the nonwoven fabric; projecting the pressure time-series features and the spatial visual features onto the same high-dimensional semantic space and completing time-series alignment; performing feature fusion on the aligned multimodal feature set to obtain the fused feature map of the nonwoven fabric.

[0028] The step of performing feature fusion on the aligned multimodal feature set, and obtaining the fused feature map of the nonwoven fabric, includes: assigning a learnable attention weight coefficient to each modal feature in the aligned multimodal feature set; weighting and superimposing the corresponding modal features according to the attention weight coefficient; and performing nonlinear activation on the weighted and superimposed fusion result to obtain the fused feature map of the nonwoven fabric.

[0029] Feature points are extracted from the second type of state data to obtain the pressure time-series characteristics of the dispensing equipment. Specifically, for the continuous data of the dispensing equipment pressure changing over time recorded in the second type of state data, the pressure value corresponding to each time node is continuously monitored. When the pressure value starts to increase continuously from a stable state and the increase exceeds the normal fluctuation range of the stable state, the position corresponding to that moment is the starting position of the pressure rising phase. When the pressure value reaches its maximum value and stops increasing and then shows a downward trend, the position corresponding to that moment is the pressure peak position. When the pressure value remains constant after the peak and the fluctuation range is controlled within a very small range, the interval corresponding to that time period is the holding position of the pressure stabilization phase. When the pressure value starts to decrease continuously from a stable state and the decrease exceeds the normal fluctuation range of the stable state, the position corresponding to that moment is the ending position of the pressure falling phase. The pressure values ​​and corresponding time information corresponding to these positions are extracted, and all the extracted feature points are arranged in chronological order. The resulting ordered feature point set is the pressure time-series characteristic of the dispensing equipment.

[0030] The nonwoven fabric running length of the first type of state data is normalized in real time to obtain the standardized running length of the nonwoven fabric. Specifically, the actual running length information of the nonwoven fabric transmitted in the first type of state data is continuously collected. The rated running length range is determined as the standard length interval according to the nonwoven fabric production process requirements. The upper and lower limits of the standard length interval are determined by the design and usage specifications of the nonwoven fabric during the production process. The real-time collected actual running length of the nonwoven fabric is compared with the upper and lower limits of the standard length interval to calculate the proportion of the real-time running length relative to the standard length interval. The real-time running length is converted into the corresponding specific value within the standard length interval through linear mapping. During the conversion process, the change trend of the nonwoven fabric running length is kept completely consistent with the actual change trend. The value obtained after conversion is the standardized running length of the nonwoven fabric.

[0031] Motion state detection is performed on the standardized running length to obtain the spatial visual features of the nonwoven fabric. Specifically, based on the obtained standardized running length of the nonwoven fabric, the standardized running length value corresponding to each time node is recorded at fixed time intervals. The difference between the standardized running lengths in two adjacent time intervals is calculated. This difference directly reflects the length change rate of the nonwoven fabric in the corresponding time period. The change in the difference of multiple consecutive time intervals reflects the length change trend of the nonwoven fabric during operation. At the same time, combined with the actual transmission path of the nonwoven fabric on the production equipment, the specific spatial position coordinates of the nonwoven fabric in the production equipment in each time interval are determined. The length change rate and length change trend corresponding to each time interval are matched one by one with the corresponding spatial position coordinates. The resulting set of associated information containing the time dimension length change feature and the spatial dimension position feature is the spatial visual feature of the nonwoven fabric.

[0032] The pressure temporal features and the spatial visual features are projected into the same high-dimensional semantic space and time-aligned. Specifically, a feature mapping rule for the high-dimensional semantic space is constructed. For the pressure temporal features, the pressure value and time information corresponding to each feature point are converted into feature vector dimensions in the high-dimensional semantic space, and each feature point corresponds to a specific coordinate point in the high-dimensional semantic space. For the spatial visual features, the length change rate, length change trend, and spatial position coordinates corresponding to each time interval are converted into feature vector dimensions in the same high-dimensional semantic space, and the association information of each time interval corresponds to a specific coordinate point in the high-dimensional semantic space. The projection operation of the two features into the same high-dimensional semantic space is completed through this mapping rule. After the projection is completed, the time interval division method of the spatial visual features is adjusted based on the time node of the pressure temporal features, so that the start and end times of each time interval of the spatial visual features completely coincide with the corresponding time node of the pressure temporal features, ensuring accurate correspondence of the two features in the time dimension, and finally obtaining the aligned multimodal feature set.

[0033] Learnable attention weight coefficients are assigned to each modal feature in the aligned multimodal feature set. Specifically, a dynamic evaluation mechanism for feature importance is established for the stress temporal feature modality and spatial visual feature modality in the aligned multimodal feature set. Before feature fusion, the mechanism performs pre-representation tests on the two modal features. The stress temporal feature and spatial visual feature are input into the feature representation model respectively, and the accuracy of each feature in representing the nonwoven fabric production state is tested. Based on the accuracy, weight values ​​are initially assigned. In the subsequent continuous feature fusion process, the representation effect of the fusion result on the nonwoven fabric features is monitored in real time. When it is found that the addition of a certain modal feature significantly improves the representation effect, the weight value corresponding to that modality is increased. When it is found that the addition of a certain modal feature has no positive impact on the representation effect or even interferes with it, the weight value corresponding to that modality is decreased. The weight values ​​obtained through this dynamic adjustment method, which can be continuously optimized according to the contribution of features, are the learnable attention weight coefficients.

[0034] Based on the attention weight coefficients, the corresponding modal features are weighted and superimposed. Specifically, the pressure value and time information corresponding to each feature point in the pressure time series features are extracted. Each feature value is multiplied by the attention weight coefficient assigned to the pressure time series feature modality to obtain a weighted feature value set of the pressure time series features. The length change rate value, change trend value, and spatial position coordinate corresponding to each time interval in the spatial visual features are extracted. Each feature value is multiplied by the attention weight coefficient assigned to the spatial visual feature modality to obtain a weighted feature value set of the spatial visual features. The feature values ​​corresponding to the same time node in the weighted feature value set of the pressure time series features and the weighted feature value set of the spatial visual features are added point by point. The set containing feature values ​​of all time nodes after addition is the fusion result after weighted superposition.

[0035] The weighted superposition fusion result is nonlinearly activated to obtain the fusion feature map of the nonwoven fabric. Specifically, a nonlinear activation process is used to transform the weighted superposition fusion result. During the transformation, for the part of the fusion result with values ​​greater than a preset threshold, the numerical difference is increased according to a specific nonlinear transformation law to highlight the effective feature information of the nonwoven fabric corresponding to the part of the value. For the part of the fusion result with values ​​less than the preset threshold, the numerical difference is reduced according to the same nonlinear transformation law to suppress the invalid interference information corresponding to the part of the value. After this nonlinear transformation processing, the feature set formed has a clear spatial position correspondence, clear numerical hierarchy differences, and can completely characterize the nonwoven fabric's operating state and dispensing pressure state, which is the fusion feature map of the nonwoven fabric.

[0036] S3. Classify the fused feature map for faults to obtain the real-time fault type and fault confidence of the dispensing equipment. In this embodiment of the invention, classifying the fused feature map for faults to obtain the real-time fault type and fault confidence of the dispensing equipment includes: retrieving multiple fault feature templates from a pre-stored fault feature template library, each fault feature template being associated with a preset fault category; comparing the fused feature map with each fault feature template element by element to obtain feature similarity; selecting the fault feature template with the highest similarity based on the feature similarity to obtain the real-time fault type corresponding to the fused feature map; and using the highest feature similarity value corresponding to the selected fault feature template as the fault confidence.

[0037] The step of comparing the fused feature map with each fault feature template element by element to obtain feature similarity includes: flattening the fused feature map to obtain a first feature vector of the dispensing equipment; flattening each fault feature template in the fault feature template library to obtain a second feature vector corresponding to the fault feature template; calculating the vector similarity between the second feature vector and the first feature vector to obtain an initial similarity value of the dispensing equipment; and normalizing the initial similarity value to obtain the feature similarity of the dispensing equipment.

[0038] The formula for calculating the initial similarity value is as follows:

[0039] In the formula, The initial similarity value, The total number of feature dimensions. The ordinal number of the feature dimension. For the first Discrimination weight coefficients for each feature dimension The first feature vector is the first feature vector. Each feature dimension The second feature vector is the first Each feature dimension.

[0040] Each feature dimension of the first feature vector comes from a one-dimensional feature sequence formed after the flattening operation of the fused feature map. This fused feature map is obtained by weighted superposition of pressure temporal features and spatial visual features followed by nonlinear activation.

[0041] Each feature dimension of the second feature vector comes from a one-dimensional feature sequence formed after the fault feature template is flattened. This fault feature template is retrieved from a pre-stored fault feature template library.

[0042] The total number of feature dimensions is determined by the length of the first or second feature vector, since they are generated using the same flattening rule and have the same length.

[0043] The ordinal number of a feature dimension is a sequential count of each feature dimension from the first to the last, used to locate the position of each feature in the vector.

[0044] No. The distinguishing weight coefficients for each feature dimension are pre-set according to the degree of influence of different feature dimensions in the fault detection process. After being set, they are stored in the weight configuration area of ​​the fault detection unit and can be directly retrieved and used during calculation.

[0045] This calculation process is used to measure the similarity between the first feature vector and the second feature vector. By assigning preset distinguishing weights to different feature dimensions, the influence of the feature dimensions that are more important for fault detection is strengthened, while the interference of the secondary feature dimensions is weakened. The final value directly reflects the degree of matching between the two vectors after weighting. The higher the value, the more consistent the feature information corresponding to the two vectors is, and the higher the degree of matching of the corresponding fault category.

[0046] The closer the information of the first feature vector and the second feature vector is in their corresponding feature dimensions, the larger the calculated initial similarity value is. When all the corresponding feature dimensions of the two vectors are completely identical, the initial similarity value reaches its maximum. The greater the difference in the information of the two vectors in their corresponding feature dimensions, the smaller the calculated initial similarity value is. When the corresponding feature dimensions of the two vectors are completely unrelated, the initial similarity value approaches its minimum.

[0047] Access the fault feature template library that is pre-built and stored in the dedicated storage area of ​​the fault detection unit. This template library is built based on historical fault data and simulated fault scenario data of the dispensing equipment. Each fault feature template in the library corresponds to a specific fault manifestation of the dispensing equipment. At the same time, the fault detection unit has a dedicated association table that clearly records the unique identifier of each fault feature template and its corresponding preset fault category. When the fault detection unit receives the fused feature map and completes the preprocessing, it automatically triggers the template retrieval command and extracts all preset fault feature templates from the fault feature template library according to the preset retrieval rules.

[0048] The generated fusion feature map of the nonwoven fabric is retrieved from the fusion feature storage area of ​​the feature processing unit. This fusion feature map is a set of feature information in a two-dimensional structure. Starting from the upper left corner of the fusion feature map in a fixed row-first order, the feature information corresponding to each position is extracted row by row and column by column. During the extraction process, the original attributes of the feature information are kept unchanged. Then, all the extracted feature information is arranged continuously in the extraction order to form a one-dimensional feature sequence without gaps. This one-dimensional feature sequence is the first feature vector of the dispensing device. After the first feature vector is generated, it is directly stored in the first vector-specific partition of the vector storage area.

[0049] For each fault feature template retrieved from the fault feature template library, the same operating standard as the fusion feature map flattening is adopted. Starting from the top left corner of each fault feature template in row-first order, the feature information of each position in the template is extracted row by row and column by column. During the extraction process, the original state of the template feature information is maintained. Then, all the extracted feature information is arranged continuously in the extraction order into a one-dimensional feature sequence without gaps. Each fault feature template generates a corresponding one-dimensional feature sequence through this operation. This one-dimensional feature sequence is the second feature vector corresponding to the fault feature template. At the same time, a unique identifier consistent with the fault feature template from which it originates is added to each second feature vector. All second feature vectors are uniformly stored in the second vector-specific partition of the vector storage area.

[0050] The first feature vector is retrieved from the first dedicated partition of the vector storage area, and each second feature vector is retrieved one by one from the second dedicated partition. During the retrieval process, a dedicated identifier is used to ensure that each second feature vector corresponds one-to-one with the corresponding fault feature template. Since the first feature vector and the second feature vector are generated using the same flattening operation, their vector lengths are completely identical. The first feature vector and each second feature vector are matched position by position, and the feature information of the corresponding positions of the two vectors is compared position by position. The number of matching position features is recorded. The number of matching positions is divided by the total length of the vectors, and the result is the quantified value of the matching degree between the first feature vector and the second feature vector. This quantified value is the initial similarity value of the dispensing device. Each second feature vector will be used to calculate a corresponding initial similarity value with the first feature vector. All initial similarity values ​​are bound to and stored with the dedicated identifier of the corresponding second feature vector.

[0051] Retrieve all generated initial similarity values ​​from the similarity storage area and place these initial similarity values ​​into the same comparison set. Determine the maximum and minimum values ​​in the set by comparing them one by one. For each initial similarity value in the set, perform a uniform numerical adjustment operation. First, subtract the minimum value in the comparison set from the initial similarity value to obtain the corresponding difference. Then, divide the difference by the difference between the maximum and minimum values ​​in the comparison set. Through this operation, all initial similarity values ​​are mapped to a fixed value range. The value obtained after this adjustment operation is the feature similarity of the dispensing device. All feature similarities replace the original initial similarity values, are bound to the exclusive identifier of the corresponding second feature vector, and are re-stored.

[0052] All generated feature similarities are imported into the numerical sorting module and sorted in descending order of value. The feature similarity with the largest value in the sorting results is selected. The second feature vector from which the feature similarity originated is obtained by reverse lookup using the unique identifier corresponding to the feature similarity. The corresponding fault feature template is then retrieved by looking up the unique identifier of the second feature vector. The preset fault category pre-bound to the fault feature template in the dedicated association table is then retrieved. This preset fault category is the real-time fault type corresponding to the fused feature map. After the real-time fault type is generated, it is stored in the fault detection result area.

[0053] The maximum feature similarity obtained after sorting the values ​​is directly determined as the reliability quantification index of the current fault detection result. This maximum feature similarity value is the fault confidence level. The fault confidence level is bound to the real-time fault type and stored in the same data record in the fault detection result area for reference when issuing subsequent fault warnings and processing instructions.

[0054] S4. Dynamically generate a load control instruction set based on the real-time fault type, the fault confidence level, and the current process parameters of the dispensing equipment; In this embodiment of the invention, the step of dynamically generating a load control instruction set based on the real-time fault type, the fault confidence level, and the current process parameters of the dispensing equipment includes: querying the process parameters bound to the current nonwoven fabric dispensing task; retrieving the corresponding basic wiping action sequence from a predefined wiping strategy mapping table based on the real-time fault type; dynamically adjusting the action intensity parameter and action duration parameter in the basic wiping action sequence based on the fault confidence level to obtain an adaptive wiping action sequence; and performing process compensation correction on the parameters in the adaptive wiping action sequence by combining the adhesive viscosity grade and dispensing speed setting value in the process parameters to obtain the wiping control instruction set of the dispensing equipment.

[0055] Access the dedicated storage area of ​​the task management unit of the dispensing equipment. This area pre-stores the process parameters for all nonwoven fabric dispensing tasks. Each dispensing task is uniquely bound to the corresponding process parameters through a unique task identifier. When the current nonwoven fabric dispensing task is started, a corresponding task identifier is generated. After receiving the process parameter query command, the task management unit matches the current task identifier with the task identifiers in the storage area one by one. If a match is successful, all process parameters bound to that identifier are retrieved. The retrieved process parameters include core information related to the wiping action, such as the adhesive viscosity grade and dispensing speed setting.

[0056] Access the predefined adhesive wiping strategy mapping table, which is stored in the device's strategy management unit. The mapping table records all preset fault types and the basic adhesive wiping action sequence corresponding to each fault type in a one-to-one correspondence. After receiving the determined real-time fault type, the strategy management unit uses the real-time fault type as the search keyword to perform precise matching in the adhesive wiping strategy mapping table. After matching the corresponding entry, the basic adhesive wiping action sequence associated with that entry is directly retrieved.

[0057] The retrieved basic adhesive application sequence and fault confidence level are input into the device's dynamic parameter adjustment unit. This unit pre-sets the correspondence between fault confidence level and the adjustment range of action parameters. The higher the fault confidence level, the higher the accuracy of the current fault matching, and the smaller the adjustment range of the corresponding action intensity and duration parameters. Conversely, the lower the fault confidence level, the lower the accuracy of the current fault matching, and the larger the adjustment range of the corresponding action intensity and duration parameters. According to this correspondence, the dynamic parameter adjustment unit makes targeted modifications to the intensity and duration parameters of each action in the basic adhesive application sequence. The output action sequence after modification is the adaptive adhesive application sequence.

[0058] The adhesive viscosity grade and dispensing speed setting value from the retrieved process parameters are input into the process compensation and correction unit of the equipment. This unit pre-sets the correlation rules between adhesive viscosity grade and wiping action intensity, as well as the correlation rules between dispensing speed setting value and wiping action duration. The higher the adhesive viscosity grade, the greater the required wiping action intensity, and the higher the dispensing speed setting value, the longer the required wiping action duration. According to these correlation rules, the process compensation and correction unit performs secondary correction on the action intensity parameters and action duration parameters in the adaptive wiping action sequence. After the correction is completed, the action sequence is converted into electrical signal commands that can be recognized by the wiping actuator of the dispensing equipment. The set of all commands is the wiping control command set of the dispensing equipment.

[0059] In this embodiment of the invention, S5, the corresponding process node is driven to perform a gluing operation according to the load control instruction set; the gluing control instruction set stored in the output area of ​​the process compensation correction unit is retrieved. This instruction set contains the action instructions and parameter configurations of the corresponding process node. The instruction set is transmitted completely to the instruction buffer area of ​​the drive control unit through the dedicated signal transmission bus inside the equipment. The drive control unit parses the instruction set frame by frame to clarify the action type and execution requirements corresponding to each process node.

[0060] After parsing the instructions, the drive control unit identifies all process nodes involved in the adhesive application operation, including the clamping mechanism, the non-woven fabric conveying mechanism, and the adhesive application execution mechanism at the cleaning station. For each process node's execution mechanism, the drive control unit generates a corresponding electric drive signal and transmits the signal to the power input port of each execution mechanism through a dedicated drive line.

[0061] After receiving the drive signal, the clamping mechanism adjusts the clamping force parameters according to the instructions to keep the non-woven fabric stable and fixed. After receiving the drive signal, the non-woven fabric conveying mechanism conveys the non-woven fabric to the cleaning station directly below the precision dispensing nozzle according to the instructions. After receiving the drive signal, the adhesive wiping actuator drives the gripper holding the non-woven fabric to wipe the precision dispensing nozzle according to the preset motion trajectory and intensity. The duration of the wiping action fully meets the parameter requirements in the instruction set.

[0062] During the adhesive application process, each process node's actuator is equipped with a status feedback sensor. The force sensor of the clamping mechanism provides real-time feedback on the clamping force, the position sensor of the conveying mechanism provides real-time feedback on the position of the nonwoven fabric, and the displacement sensor of the adhesive application actuator provides real-time feedback on the trajectory of the wiping action. All feedback signals are transmitted to the drive control unit through a dedicated feedback line. The drive control unit compares the feedback status with the command requirements in real time. If a deviation is found, it immediately adjusts the output parameters of the drive signal to ensure that the adhesive application process fully complies with the command requirements.

[0063] After the adhesive wiping actuator completes all wiping actions, the drive control unit confirms that the execution status of all process nodes has met the instruction requirements, and then generates an adhesive wiping operation completion signal. This signal is transmitted to the task management unit through the internal signal bus of the equipment. The task management unit updates the status of the current nonwoven fabric dispensing task to adhesive wiping completion, and waits for the start instruction of the subsequent dispensing process.

[0064] S6. After the glue application operation is completed, the status data of the glue dispensing equipment is re-verified for faults. If the fault confidence level is lower than the set threshold, the fault status and glue application completion status are output.

[0065] In this embodiment of the invention, the step of re-verifying the status data of the dispensing equipment after the glue application operation is completed, and outputting the fault status and glue application completion status if the fault confidence is lower than a set threshold, includes: immediately collecting the verification status data of the dispensing equipment for at least one complete glue application cycle after the glue application actuator is reset; extracting features from the verification status data to obtain a verification feature vector; comparing the verification feature vector with the fault feature template corresponding to the real-time fault type in the fault feature template library to obtain the feature difference degree of the dispensing equipment; updating the fault confidence based on the feature difference degree to obtain an updated fault confidence; determining whether the updated fault confidence is lower than a set threshold, and if so, outputting the glue application completion status; otherwise, triggering a fault alarm and generating a fault log and fault status.

[0066] After the wiping actuator completes the wiping action, its onboard displacement sensor detects the actuator's position information in real time. When the actuator's position coordinates are completely consistent with the preset initial position coordinates, a reset completion electrical signal is immediately generated. This signal is transmitted to the verification data acquisition unit through a dedicated line. Upon receiving the signal, the verification data acquisition unit immediately starts the acquisition process, collecting verification status data of the dispensing equipment within at least one complete wiping cycle. The start point of a complete wiping cycle is the moment when the clamping mechanism receives the clamping command and begins to move, and the end point is the moment when the wiping actuator completes the wiping and returns to the initial position. The collected verification status data includes images of the non-woven fabric surface morphology taken by a high-definition industrial camera directly above the cleaning station, as well as real-time clamping force and displacement feedback data obtained by the force-sensitive sensor of the clamping mechanism's moving claw and the displacement sensor of the linear drive component. All collected verification status data are stored in the dedicated verification data storage area in chronological order of collection time.

[0067] The verification data acquisition unit transmits the verification status data stored in the dedicated storage area to the feature extraction unit. For the nonwoven fabric surface morphology image, the feature extraction unit first selects the core area directly related to the cleaning effect of the precision dispensing nozzle, removes the background and irrelevant parts at the image edges to complete key area cropping, and then uses pixel interpolation to adjust the cropped image to a fixed size for scale normalization. Subsequently, it extracts texture features such as the direction distribution of nonwoven fabric fibers, fiber density, and surface pore size. For real-time clamping force and displacement feedback data, it extracts temporal features such as the peak and valley values ​​of the clamping force and the start and end times of displacement changes. Then, it integrates the extracted texture and temporal features into a two-dimensional feature set, and extracts information from the feature set row by row and column by column in a fixed row-first order, arranging it into a one-dimensional feature sequence without gaps. This one-dimensional feature sequence is the verification feature vector, which is then stored in the verification vector-specific partition of the feature vector storage area.

[0068] The feature extraction unit transmits the verification feature vector to the feature matching unit. The feature matching unit retrieves the unique association identifier corresponding to the previously determined real-time fault type. Using this identifier, it accurately retrieves the corresponding fault feature template from the fault feature template library. Then, using the same flattening operation standard as when generating the verification feature vector, it converts the fault feature template into a one-dimensional fault feature vector. Subsequently, it compares the verification feature vector and the fault feature vector bit by bit, comparing the feature information at corresponding positions of the two vectors row by row and column by column. It records the number of inconsistent feature information at corresponding positions of the two vectors. Dividing the number of inconsistent positions by the total length of the vectors, the result is the feature difference degree of the dispensing equipment. After the feature difference degree is generated, it is bound and stored in the difference degree storage area along with the corresponding verification feature vector and fault feature vector.

[0069] The feature matching unit transmits the feature difference to the confidence update unit. The confidence update unit pre-stores the corresponding rules between the feature difference and the fault confidence adjustment range. When the feature difference value is higher, it means that the current verification state deviates more from the fault feature, and the corresponding fault confidence is lowered more. When the feature difference value is lower, it means that the current verification state deviates less from the fault feature, and the corresponding fault confidence is lowered less. The confidence update unit first retrieves the original fault confidence stored in the confidence storage area, and then calculates the corresponding adjustment value according to the preset rules. The original fault confidence is subtracted from the adjustment value, and the result is the updated fault confidence. The updated fault confidence overwrites the original fault confidence and is bound to the corresponding feature difference and real-time fault type.

[0070] The confidence update unit transmits the updated fault confidence to the status judgment unit. The status judgment unit retrieves a pre-set fault confidence threshold from the device configuration storage area and compares the updated fault confidence with the set threshold. If the updated fault confidence is lower than the set threshold, the status judgment unit immediately generates a glue application completion status signal. This signal is transmitted to the task management unit via the device's internal signal bus. Upon receiving the signal, the task management unit updates the glue application process status of the current nonwoven fabric dispensing task to completion and sends a start preparation signal to the subsequent glue application process. If the updated fault confidence is not lower than the set threshold, the status judgment unit immediately sends a trigger signal to the device's fault alarm module. Upon receiving the signal, the fault alarm module immediately activates the audible and visual alarm device, emitting a continuous audible and visual prompt. Simultaneously, the status judgment unit generates a fault log, which contains core information such as the execution time of the current glue application operation, real-time fault type, characteristic difference, and fault confidence before and after the update. This log is stored in the device's log storage area. Subsequently, the status judgment unit updates the device's operating status to a fault status and displays it synchronously on the device's status display panel for real-time viewing by operators.

[0071] Figure 2 shows a functional block diagram of a precision dispensing nozzle wiping control system based on real-time fault detection provided in an embodiment of the present invention.

[0072] The precision dispensing nozzle wiping control system 100 based on real-time fault detection described in this invention can be installed in an electronic device. Depending on the functions implemented, the precision dispensing nozzle wiping control system 100 may include a human-machine interface module 101, a wiping process execution module 102, a fault detection module 103, a load control module 104, a control signal module 105, and a communication module 106. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0073] In this embodiment, the functions of each module / unit are as follows: The human-machine interaction module 101 is used to automatically spray alcohol and clamp the glue nozzle after the dispensing equipment is started, and to acquire multi-source state data of the nonwoven fabric after the operation is completed; the glue wiping process execution module 102 is used to perform cross-modal feature fusion on the multi-source state data to generate a fused feature map of the nonwoven fabric; the fault detection module 103 is used to classify the fused feature map for faults to obtain the real-time fault type and fault confidence of the dispensing equipment; the load control module 104 is used to dynamically generate a load control instruction set according to the real-time fault type, the fault confidence and the current process parameters of the dispensing equipment; the control signal module 105 is used to drive the corresponding process node to perform glue wiping operation according to the load control instruction set; the communication module 106 is used to re-verify the state data of the dispensing equipment after the glue wiping operation is completed, and if the fault confidence is lower than a set threshold, output the fault status and glue wiping completion status.

[0074] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0075] The modules described as separate components may or may not be physically separate. The components shown as modules 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.

[0076] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0077] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0078] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for controlling the adhesive application of a precision dispensing nozzle based on real-time fault detection, characterized in that, The method includes: S1, after startup, the dispensing equipment automatically sprays alcohol and clamps the nozzle to apply adhesive, and acquires multi-source state data of the nonwoven fabric after the operation is completed; S2, cross-modal feature fusion is performed on the multi-source state data to generate a fused feature map of the nonwoven fabric; S3, fault classification is performed on the fused feature map to obtain the real-time fault type and fault confidence of the dispensing equipment; S4, a load control instruction set is dynamically generated based on the real-time fault type, the fault confidence, and the current process parameters of the dispensing equipment; S5, the corresponding process node is driven to perform adhesive application according to the load control instruction set; S6, after the adhesive application is completed, the state data of the dispensing equipment is re-verified for faults. If the fault confidence is lower than a set threshold, the fault status and adhesive application completion status are output.

2. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 1, characterized in that, The dispensing equipment, after startup, automatically sprays alcohol and clamps the nozzle, and acquires multi-source state data of the nonwoven fabric after the operation is completed. This includes: responding to the startup completion signal of the dispensing equipment, sending spray duration and atomization pressure parameters to the precision nozzle assembly; the clamping mechanism performing the clamping action to fix the nonwoven fabric to the cleaning station of the precision dispensing nozzle; acquiring the surface morphology image of the nonwoven fabric under the combined action of alcohol wetting and mechanical clamping, as the first type of state data; real-time clamping force and displacement feedback data during the clamping action, as the second type of state data; and combining the first type of state data and the second type of state data into multi-source state data.

3. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 2, characterized in that, The step of fusing cross-modal features of the multi-source state data to generate a fused feature map of the nonwoven fabric includes: extracting feature points from the second type of state data to obtain the pressure time-series features of the dispensing device; performing real-time detection scale normalization on the nonwoven fabric running length of the first type of state data to obtain the standardized running length of the nonwoven fabric; performing motion state detection on the standardized running length to obtain the spatial visual features of the nonwoven fabric; projecting the pressure time-series features and the spatial visual features into the same high-dimensional semantic space and completing time-series alignment; and performing feature fusion on the aligned multimodal feature set to obtain the fused feature map of the nonwoven fabric.

4. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 3, characterized in that, The step of performing feature fusion on the aligned multimodal feature set, and obtaining the fused feature map of the nonwoven fabric, includes: assigning a learnable attention weight coefficient to each modal feature in the aligned multimodal feature set; weighting and superimposing the corresponding modal features according to the attention weight coefficient; and performing nonlinear activation on the weighted and superimposed fusion result to obtain the fused feature map of the nonwoven fabric.

5. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 1, characterized in that, The step of classifying the fused feature map to obtain the real-time fault type and fault confidence of the dispensing equipment includes: retrieving multiple fault feature templates from a pre-stored fault feature template library, each fault feature template being associated with a preset fault category; comparing the fused feature map with each fault feature template element by element to obtain feature similarity; selecting the fault feature template with the highest similarity based on the feature similarity to obtain the real-time fault type corresponding to the fused feature map; and using the highest feature similarity value corresponding to the selected fault feature template as the fault confidence.

6. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 5, characterized in that, The step of comparing the fused feature map with each fault feature template element by element to obtain feature similarity includes: flattening the fused feature map to obtain a first feature vector of the dispensing equipment; flattening each fault feature template in the fault feature template library to obtain a second feature vector corresponding to the fault feature template; calculating the vector similarity between the second feature vector and the first feature vector to obtain an initial similarity value of the dispensing equipment; and normalizing the initial similarity value to obtain the feature similarity of the dispensing equipment.

7. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 6, characterized in that, The formula for calculating the initial similarity value is as follows: In the formula, The initial similarity value, The total number of feature dimensions. The ordinal number of the feature dimension. For the first Discrimination weight coefficients for each feature dimension The first feature vector is the first feature vector. Each feature dimension The second feature vector is the first Each feature dimension.

8. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 1, characterized in that, The step of dynamically generating a load control instruction set based on the real-time fault type, the fault confidence level, and the current process parameters of the dispensing equipment includes: querying the process parameters bound to the current nonwoven fabric dispensing task; retrieving the corresponding basic wiping action sequence from a predefined wiping strategy mapping table based on the real-time fault type; dynamically adjusting the action intensity and action duration parameters in the basic wiping action sequence based on the fault confidence level to obtain an adaptive wiping action sequence; and performing process compensation correction on the parameters in the adaptive wiping action sequence by combining the adhesive viscosity grade and dispensing speed setpoint in the process parameters to obtain the wiping control instruction set of the dispensing equipment.

9. The precision dispensing nozzle erasure control method based on real-time fault detection as described in claim 5, characterized in that, The step of re-verifying the status data of the dispensing equipment after the glue application operation is completed, and outputting the fault status and glue application completion status if the fault confidence is lower than a set threshold, includes: immediately collecting the verification status data of the dispensing equipment for at least one complete glue application cycle after the glue application actuator is reset; extracting features from the verification status data to obtain a verification feature vector; comparing the verification feature vector with the fault feature template corresponding to the real-time fault type in the fault feature template library to obtain the feature difference degree of the dispensing equipment; updating the fault confidence based on the feature difference degree to obtain the updated fault confidence; determining whether the updated fault confidence is lower than a set threshold, and if so, outputting the glue application completion status; otherwise, triggering a fault alarm and generating a fault log and fault status.

10. A precision dispensing nozzle wiping control system based on real-time fault detection, used to implement the precision dispensing nozzle wiping control method based on real-time fault detection as described in claim 1, the system comprising: The human-machine interaction module is used to automatically spray alcohol and wipe and clamp the glue nozzle after the dispensing equipment is started, and to acquire multi-source status data of the non-woven fabric after the operation is completed. The adhesive application process execution module is used to perform cross-modal feature fusion on the multi-source state data to generate a fused feature map of the nonwoven fabric. The fault detection module is used to classify the fused feature map for faults and obtain the real-time fault type and fault confidence of the dispensing equipment. The load control module is used to dynamically generate a load control instruction set based on the real-time fault type, the fault confidence level, and the current process parameters of the dispensing equipment; the control signal module is used to drive the corresponding process node to perform the adhesive wiping operation based on the load control instruction set. The communication module is used to re-verify the status data of the dispensing equipment after the glue application operation is completed. If the fault confidence level is lower than a set threshold, the module outputs the fault status and the glue application completion status.

Images