Intelligent fault diagnosis and rapid recovery system for glue preparation equipment

Abstract

The invention discloses an intelligent fault diagnosis and rapid recovery system for glue preparation equipment, and relates to the technical field of industrial automation and intelligent manufacturing, and the system comprises an edge sensing layer, a real-time diagnosis control layer, and a cloud collaboration layer. The corrosion-resistant multi-mode sensor group collects glue solution viscosity, temperature, pressure and flow parameters of glue preparation equipment and equipment vibration signals in real time, and transmits the glue solution viscosity, temperature, pressure and flow parameters and the equipment vibration signals to the edge AI calculation unit through the anti-interference communication module. A mixed diagnosis mechanism of a physical driving model and a data driving model is combined, complex glue solution rheological behaviors and equipment mechanical abnormity are effectively recognized, a residual attention mechanism dynamically fuses a physical mechanism and actually measured data difference, the misjudgment rate is remarkably reduced, the timeliness and accuracy of fault alarm are ensured, and production interruption caused by missing detection or misjudgment is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial automation and intelligent manufacturing, and in particular to an intelligent fault diagnosis and rapid recovery system for glue dispensing equipment. Background Art

[0002] Fault diagnosis and recovery for traditional glue dispensing equipment primarily rely on manual experience or single sensor monitoring, resulting in delayed response and high misjudgment rates. For example, anomalies caused by complex glue rheological behavior (such as sudden changes in viscosity) are difficult to detect in a timely manner, requiring manual intervention and analysis, resulting in extended downtime and impacting production plans. Furthermore, existing methods lack the ability to comprehensively optimize multi-dimensional parameters (such as equipment vibration, energy consumption, and safety risks). Recovery strategies are often limited to a single objective, making it difficult to balance production efficiency and equipment safety. With the increasing demand for smart manufacturing, there is an urgent need for an intelligent system that can perceive equipment status in real time, integrate physical mechanisms and data-driven models, and be verified through dynamic simulation to solve the pain points of low diagnostic accuracy and insufficient recovery efficiency in traditional technologies. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent fault diagnosis and rapid recovery system for glue dispensing equipment.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent fault diagnosis and rapid recovery system for glue dispensing equipment, the system comprising: The edge perception layer includes a corrosion-resistant multimodal sensor group, an edge AI computing unit, and an anti-interference communication module; Real-time diagnostic control layer, including hybrid diagnostic model engine, dynamic recovery strategy generator, and digital twin simulation module; The cloud collaboration layer includes an incremental federated learning module and a fault knowledge graph library; The corrosion-resistant multimodal sensor group collects the glue viscosity, temperature, pressure, flow parameters and equipment vibration signals of the glue dispensing equipment in real time, and transmits them to the edge AI computing unit through the anti-interference communication module; The hybrid diagnostic model engine embeds a deep learning model constrained by the glue rheological equation, extracts the spatiotemporal characteristics of sensor signals through a data-driven module, generates prediction values ​​through a physical-driven module, and triggers fault alarms through residual analysis; The dynamic recovery strategy generator generates multi-objective optimization recovery instructions based on real-time equipment status, production task priority and energy consumption constraints, and issues them for execution after verification by the digital twin simulation module.

[0005] As a further solution of the present invention: the physical-data dual-driven model construction method of the hybrid diagnostic model engine includes: a) Establish the power law model equation of the rheological behavior of the adhesive: ; in, is the shear stress, is the consistency coefficient, is the shear rate, is the mobility index; b) Using the power law model output as a priori knowledge layer for the LSTM network to constrain the data-driven prediction range; c) Dynamically weight the output differences between data-driven and physical-driven methods through the residual attention mechanism to generate failure probability.

[0006] As a further solution of the present invention: in the residual attention mechanism, the calculation formula for the failure probability is: ; in, is the residual of the data-driven module, is the residual of the physical driver module, and is the dynamic weight factor, is the Sigmoid activation function.

[0007] As a further solution of the present invention: the multi-objective optimization function of the dynamic recovery strategy generator is: ; in, For downtime, is the energy loss, is the safety risk factor, 、 、 It is a weight factor, which is dynamically adjusted according to the urgency of the production task.

[0008] As a further solution of the present invention: the weight factor 、 、 The adjustment logic includes: When the production task priority is at its peak, The weight is increased to 0.6, and reduced to 0.2 respectively; When the security risk factor exceeds the preset threshold, The weight is automatically increased to 0.5, and Decrease accordingly.

[0009] As a further solution of the present invention: before the recovery strategy is executed, the digital twin simulation module simulates the strategy effect through a high-fidelity twin model. If the simulation result meets the equipment safety threshold and the downtime is reduced by ≥20%, an instruction is issued to the execution mechanism; otherwise, the optimization strategy is regenerated.

[0010] As a further solution of the present invention: the corrosion-resistant multimodal sensor group includes a redundant sensor cross-check mechanism. When the data of a single sensor deviates from the mean by more than 5%, the infrared spectrum is triggered to analyze the glue composition and reversely correct the temperature drift error of the viscosity sensor.

[0011] As a further solution of the present invention: the infrared spectrum analysis module is communicatively connected to the edge AI computing unit, and the corrected data is updated in real time to the hybrid diagnostic model engine for dynamically adjusting the rheological equation parameters of the physical driving module.

[0012] As a further solution of the present invention: the anti-interference communication module adopts the LoRa-UWB hybrid protocol; Low-frequency LoRa transmission device status metadata, bandwidth ≤ 500kHz; High-frequency UWB transmits emergency fault signals, which have higher priority than ordinary data packets and a packet loss rate of ≤0.1%.

[0013] As a further solution of the present invention: the incremental federated learning module aggregates the local model parameters of multiple glue dispensing devices through encrypted parameters, updates the global model and then feeds it back to the edge layer to achieve cross-device knowledge sharing and model self-evolution; The fault knowledge graph library builds a semantic association network based on historical fault cases, supports matching similar faults and recovery solutions through natural language queries, and has a matching accuracy rate of ≥95%.

[0014] By adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are: 1. This invention uses a corrosion-resistant multimodal sensor group to collect multi-dimensional parameters such as glue viscosity, temperature, and vibration in real time. Combining a hybrid diagnostic mechanism with physical-driven and data-driven models, it effectively identifies complex glue rheological behavior and equipment mechanical anomalies. The residual attention mechanism dynamically integrates the differences between physical mechanisms and measured data, significantly reducing the false positive rate, ensuring the timeliness and accuracy of fault alarms, and avoiding production interruptions caused by missed detection or misjudgment. 2. This invention dynamically generates multi-objective optimized recovery instructions based on real-time equipment status, production task priorities, and safety constraints. It also pre-verifies the feasibility of the strategies through a digital twin simulation module, ensuring that recovery operations balance efficiency and safety. Compared to traditional single-objective recovery models, this invention can reduce downtime while proactively avoiding the risk of energy consumption surges or equipment damage, ensuring production continuity and optimizing resource allocation. 3. This invention relies on the incremental federated learning module to achieve encrypted sharing of model parameters and global iterative optimization among multiple devices, continuously improving the system's adaptability to complex working conditions. The fault knowledge graph library supports natural language queries for historical cases and solutions, assisting maintenance personnel in quickly locating the root cause of faults. The redundant sensor self-calibration mechanism and environmental anti-interference design further enhance the long-term reliability of the system in harsh industrial scenarios, reducing operation and maintenance dependence and labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is the first flow chart of the intelligent fault diagnosis and rapid recovery system for glue dispensing equipment; Figure 2 The second flowchart of the intelligent fault diagnosis and rapid recovery system for glue dispensing equipment; Figure 3 The third flow chart of the intelligent fault diagnosis and rapid recovery system for glue dispensing equipment; Figure 4 The fourth flowchart of the intelligent fault diagnosis and rapid recovery system for glue dispensing equipment; Figure 5 This is the data flow diagram of the edge perception layer; Figure 6 It is the workflow diagram of the hybrid diagnostic model engine; Figure 7 Generate and validate flow charts for dynamic recovery strategies; Figure 8 This is a cycle diagram of cloud collaboration layer capabilities. DETAILED DESCRIPTION

[0016] The specific embodiments of the present invention will be further described below in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0017] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0018] Please see the attached Figure 1 -Attached Figure 8 The present invention provides an intelligent fault diagnosis and rapid recovery system for glue dispensing equipment, the system comprising: The edge perception layer includes a corrosion-resistant multimodal sensor group, an edge AI computing unit, and an anti-interference communication module; Real-time diagnostic control layer, including hybrid diagnostic model engine, dynamic recovery strategy generator, and digital twin simulation module; The cloud collaboration layer includes an incremental federated learning module and a fault knowledge graph library; The corrosion-resistant multimodal sensor group collects the glue viscosity, temperature, pressure, flow parameters and equipment vibration signals of the glue dispensing equipment in real time, and transmits them to the edge AI computing unit through the anti-interference communication module; The hybrid diagnostic model engine embeds a deep learning model constrained by the glue rheological equation. The data-driven module extracts the spatiotemporal characteristics of sensor signals, the physical-driven module generates predictions, and triggers fault alarms through residual analysis. The dynamic recovery strategy generator generates multi-objective optimization recovery instructions based on real-time equipment status, production task priority and energy consumption constraints, and issues them for execution after verification by the digital twin simulation module.

[0019] In one embodiment of the present invention, the physical-data dual-driven model construction method of the hybrid diagnostic model engine includes: a) Establish the power law model equation of the rheological behavior of the adhesive: ; in, is the shear stress, is the consistency coefficient, is the shear rate, is the mobility index; b) Using the power law model output as a priori knowledge layer for the LSTM network to constrain the data-driven prediction range; c) Dynamically weight the output differences between data-driven and physical-driven methods through the residual attention mechanism to generate failure probability.

[0020] In one embodiment of the present invention: in the residual attention mechanism, the calculation formula for the failure probability is: ; in, is the residual of the data-driven module, is the residual of the physical driver module, and is the dynamic weight factor, is the Sigmoid activation function.

[0021] In one embodiment of the present invention, the multi-objective optimization function of the dynamic recovery strategy generator is: ; in, For downtime, is the energy loss, is the safety risk factor, 、 、 It is a weight factor, which is dynamically adjusted according to the urgency of the production task.

[0022] In one embodiment of the present invention: the weighting factor 、 、 The adjustment logic includes: When the production task priority is at its peak, The weight is increased to 0.6, and reduced to 0.2 respectively; When the security risk factor exceeds the preset threshold, The weight is automatically increased to 0.5, and Decrease accordingly.

[0023] In one embodiment of the present invention: before the recovery strategy is executed, the digital twin simulation module simulates the strategy effect through a high-fidelity twin model. If the simulation result meets the equipment safety threshold and the downtime is reduced by ≥20%, an instruction is issued to the execution mechanism; otherwise, the optimization strategy is regenerated.

[0024] In one embodiment of the present invention: the corrosion-resistant multimodal sensor group includes a redundant sensor cross-check mechanism. When the data of a single sensor deviates from the mean by more than 5%, infrared spectroscopy is triggered to analyze the composition of the glue and reversely correct the temperature drift error of the viscosity sensor.

[0025] In one embodiment of the present invention: the infrared spectroscopy analysis module is communicatively connected to the edge AI computing unit, and the corrected data is updated in real time to the hybrid diagnostic model engine for dynamically adjusting the rheological equation parameters of the physical driving module.

[0026] In one embodiment of the present invention: the anti-interference communication module adopts the LoRa-UWB hybrid protocol; Low-frequency LoRa transmission device status metadata, bandwidth ≤ 500kHz; High-frequency UWB transmits emergency fault signals, which have higher priority than ordinary data packets and a packet loss rate of ≤0.1%.

[0027] In one embodiment of the present invention: the incremental federated learning module aggregates the local model parameters of multiple glue dispensing devices through encrypted parameters, updates the global model and then feeds it back to the edge layer to achieve cross-device knowledge sharing and model self-evolution; The fault knowledge graph library builds a semantic association network based on historical fault cases, supports matching similar faults and recovery solutions through natural language queries, and has a matching accuracy rate of ≥95%.

[0028] Example 1: System Architecture Implementation and Hybrid Diagnostic Model Application Implementation scenario: The glue dispensing production line of an automobile manufacturer requires real-time monitoring of glue viscosity, temperature, and equipment vibration to ensure glue coating quality.

[0029] 1. Edge perception layer deployment: Corrosion-resistant multi-modal sensor group: Install infrared viscosity detector (model: VISC-IR2025), chemical-resistant pressure sensor (model: PT-X6) in the glue tank, real-time collection of glue viscosity (range: 500-2000mPa s), temperature (20-80°C), flow rate (5-50L / min) and vibration frequency (0-200Hz).

[0030] Anti-interference communication module: The LoRa module (frequency band: 868MHz, bandwidth: 125kHz) is used to transmit regular data, and the UWB module (frequency band: 6.5GHz, bandwidth: 500MHz) is used to transmit emergency signals. This ensures communication stability under electromagnetic interference (such as that caused by inverters) within the workshop, with a measured packet loss rate of ≤0.08%.

[0031] 2. Real-time diagnosis of control layer operation: Hybrid Diagnostic Model Engine: Data-driven module: Deploys a lightweight CNN-LSTM model based on TensorFlow Lite (input layer: 32x32 time series data, 64 LSTM units) to extract abnormal waveforms (such as amplitude mutations ≥ 15%) from vibration signals.

[0032] Physical driving module: embedded power law model (parameters: , n=0.6), the theoretical value of shear stress was calculated in real time and compared with the actual value measured by the sensor.

[0033] Residual analysis: When the residual exceeds a threshold (±10%), a fault alarm (such as "abnormal glue fluidity") is triggered, and the edge AI computing unit's FPGA (model: Xilinx Zynq UltraScale+) accelerates the calculation, with a measured diagnostic delay of 48ms.

[0034] 3. Cloud collaboration layer optimization: Incremental federated learning module: Each device's local model uploads encrypted parameters (AES-256 encryption) to the cloud daily, aggregates them to generate a global model, and feeds it back to the edge layer. After 30 days of optimization, the model's detection accuracy for abnormal glue viscosity increased from 92% to 97%.

[0035] Fault Knowledge Graph: Built on a Neo4j graph database, it links 1,200 historical fault cases. For example, when entering "viscosity fluctuation," the system matches similar cases (such as "temperature sensor drift") and recommends calibration procedures, with a measured matching accuracy of 96.3%. Implementation Effect

[0036] Fault detection response time ≤ 50ms, 10 times faster than traditional PLC systems (500ms); The misjudgment rate of complex fluid behavior was reduced from 8% to 2.5%; Production line downtime was reduced by 22% and annual maintenance costs were reduced by 35%.

[0037] Example 2: Dynamic recovery strategy generation and digital twin simulation verification Implementation scenario: An electronic component packaging factory's glue dispensing equipment needs to quickly resume production due to abnormal glue viscosity. Specific implementation method: 1. Fault triggering and locating: The sensor detects that the viscosity of the glue drops suddenly to 300 mPa s (normal range: 800-1200 mPa s), the hybrid diagnosis model engine determined it to be "glue-liquid ratio error" and located it to be a feed pump valve failure.

[0038] 2. Dynamic recovery strategy generation: Multi-objective optimization function application: ; According to the urgency of the production task (the order volume increased by 30% on that day), the weight factor is adjusted to =0.6, =0.2, =0.2.

[0039] The strategy generator calculated the optimal instructions: temporarily increase the heating power to 85°C (originally 70°C) and start the backup feed pump instead of immediately shutting down for maintenance. The estimated downtime was shortened from 30 minutes to 6 minutes.

[0040] 3. Digital twin simulation verification: High-fidelity twin model: Built using ANSYS Twin Builder, it simulates the rheological behavior of the adhesive after the heating power is increased. The simulation shows that the viscosity recovers to 750 mPa within 5 minutes. s, and the vibration amplitude of the equipment does not exceed the safety threshold (≤0.5mm).

[0041] After the simulation results are verified, the instructions are sent to the actuators (such as heaters and pump controllers).

[0042] 4. Recovery effect monitoring: In actual implementation, the viscosity of the glue recovered to 820 mPa within 6 minutes. s, the production line was down for only 5 minutes and 48 seconds, and energy loss was reduced by 18%, which was within the simulation prediction error range ( 3%) Implementation effect: Dynamic strategies reduce the average recovery time from a single failure by 80%; The digital twin simulation accuracy is ≥97%, avoiding secondary failures; Energy consumption losses decreased by 25% year-on-year, saving approximately 120,000 yuan in electricity bills annually.

[0043] Example 3: Environmental Adaptability Enhancement and Self-Evolution Capability Verification Implementation scenario: In the glue preparation workshop of a chemical plant, the ambient humidity is ≥85% and there is highly corrosive gas. DETAILED DESCRIPTION

[0044] 1. Sensor self-calibration implementation: Redundant sensor cross-check: Deploy three viscosity sensors (main sensor + 2 redundant sensors) at the same measurement point. When the data of a sensor deviates from the mean by more than 5% (for example, the main sensor shows 1200 mPa s, the redundant sensor shows 1140 mPa s and 1160 mPa s), triggering the calibration process.

[0045] Infrared spectrum-assisted correction: A near-infrared spectrometer (model: NIRScan Nano) was used to analyze the composition of the glue solution, detect concentration changes caused by solvent volatilization, and reversely correct the temperature drift error of the main sensor (the original deviation of +8% was corrected to ±1.5%).

[0046] 2. Anti-interference communication protocol optimization: LoRa-UWB hybrid transmission: Four UWB anchor points (covering a 50m radius) are deployed in the workshop to locate the source of fault signals in real time. Emergency signal transmission delay is ≤ 2ms, and bandwidth utilization is increased by 40%.

[0047] In actual measurements under strong electromagnetic interference (such as when the inverter starts up), the UWB signal bit error rate is ≤0.05%, and the LoRa data packet retransmission rate is ≤1%.

[0048] 3. Incremental Federated Learning and Model Evolution: Cross-device knowledge sharing: Five dispensing machines share local model parameters through federated learning (aggregation cycle: 24 hours). To address the problem of sensor drift in high-humidity environments, a "humidity compensation factor" is added to the global model.

[0049] After 7 days of training, the model's fault detection accuracy in high humidity environments increased from 88% to 94%, and the false alarm rate decreased to 3.2%. Implementation Effect

[0050] Sensor calibration stabilizes measurement errors within +2%, reducing operation and maintenance costs by 40%; The communication protocol ensures signal reliability in complex workshop environments, with a fault signal transmission success rate of ≥99.9%; Federated learning increases the speed at which models adapt to new environments by 60%, supporting rapid expansion of production lines.

[0051] Although the present invention is disclosed above with reference to preferred embodiments, this is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent variations, and modifications made to the above embodiments in accordance with the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection defined by the claims of the present invention.

Claims

1. An intelligent fault diagnosis and rapid recovery system for glue dispensing equipment, characterized in that: The system comprises: The edge perception layer includes a corrosion-resistant multimodal sensor group, an edge AI computing unit, and an anti-interference communication module; Real-time diagnostic control layer, including hybrid diagnostic model engine, dynamic recovery strategy generator, and digital twin simulation module; The cloud collaboration layer includes an incremental federated learning module and a fault knowledge graph library; The corrosion-resistant multimodal sensor group collects the glue viscosity, temperature, pressure, flow parameters and equipment vibration signals of the glue dispensing equipment in real time, and transmits them to the edge AI computing unit through the anti-interference communication module; The hybrid diagnostic model engine embeds a deep learning model constrained by the glue rheological equation, extracts the spatiotemporal characteristics of sensor signals through a data-driven module, generates prediction values ​​through a physical-driven module, and triggers fault alarms through residual analysis; The dynamic recovery strategy generator generates multi-objective optimization recovery instructions based on real-time equipment status, production task priority and energy consumption constraints, and issues them for execution after verification by the digital twin simulation module.

2. The system according to claim 1, wherein: The physical-data dual-driven model construction method of the hybrid diagnostic model engine includes: a) Establish the power law model equation of the rheological behavior of the adhesive: ； in, is the shear stress, is the consistency coefficient, is the shear rate, is the mobility index; b) Using the power law model output as a priori knowledge layer for the LSTM network to constrain the data-driven prediction range; c) Dynamically weight the output differences between data-driven and physical-driven methods through the residual attention mechanism to generate failure probability.

3. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 2 is characterized by: In the residual attention mechanism, the calculation formula for the failure probability is: ; in, is the residual of the data-driven module, is the residual of the physical driver module, and is the dynamic weight factor, is the Sigmoid activation function.

4. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 1 is characterized by: The multi-objective optimization function of the dynamic recovery strategy generator is: ； in, For downtime, is the energy loss, is the safety risk factor, 、 、 It is a weight factor, which is dynamically adjusted according to the urgency of the production task.

5. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 4 is characterized by: The weighting factor 、 、 The adjustment logic includes: When the production task priority is at its peak, The weight is increased to 0.6, and reduced to 0.2 respectively; When the security risk factor exceeds the preset threshold, The weight is automatically increased to 0.5, and Decrease accordingly.

6. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 1 is characterized by: Before executing the recovery strategy, the digital twin simulation module simulates the strategy effect through a high-fidelity twin model. If the simulation result meets the equipment safety threshold and the downtime is reduced by ≥20%, an instruction is issued to the execution mechanism; otherwise, the optimization strategy is regenerated.

7. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 1 is characterized by: The corrosion-resistant multimodal sensor group includes a redundant sensor cross-check mechanism. When the data of a single sensor deviates from the mean by more than 5%, infrared spectroscopy is triggered to analyze the composition of the glue liquid and reversely correct the temperature drift error of the viscosity sensor.

8. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 7 is characterized by: The infrared spectrum analysis module is communicated with the edge AI computing unit, and the corrected data is updated to the hybrid diagnostic model engine in real time for dynamically adjusting the rheological equation parameters of the physical drive module.

9. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 1 is characterized by: The anti-interference communication module adopts the LoRa-UWB hybrid protocol; Low-frequency LoRa transmission device status metadata, bandwidth ≤ 500kHz; High-frequency UWB transmits emergency fault signals, which have higher priority than ordinary data packets and a packet loss rate of ≤0.1%.

10. The intelligent fault diagnosis and rapid recovery system for glue dispensing equipment according to claim 1 is characterized by: The incremental federated learning module aggregates the local model parameters of multiple dispensing devices through encrypted parameters, updates the global model and feeds it back to the edge layer to achieve cross-device knowledge sharing and model self-evolution; The fault knowledge graph library builds a semantic association network based on historical fault cases, supports matching similar faults and recovery solutions through natural language queries, and has a matching accuracy rate of ≥95%.

Images