Electrolytic bath bolt tightening method and system based on cooperative control

Abstract

The invention belongs to the technical field of industrial automation, and provides an electrolytic bath bolt tightening method and system based on cooperative control, and the method comprises the steps: firstly, carrying out the feature extraction of image information, stress information and temperature information, and obtaining a feature vector; then, according to the feature vector and a preset threshold value dynamic adjustment model, obtaining a stress threshold value range; finally, all bolts of the electrolytic cell are cooperatively controlled on the basis of the stress threshold value range, so that the tightening force does not exceed the stress threshold value; wherein the dynamic threshold adjustment model is a federated learning algorithm, the dynamic threshold adjustment model is enabled to adapt to different temperature and stress environments through gradient updating rules and dynamic optimization, the weight is dynamically adjusted through variances of different working condition parameters, and the situation that the compression ratio of a gasket is too large due to stress overload is avoided. And when the tightening force is controlled, the real-time influence of the temperature and the stress on the bolt tightening effect is considered.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation technology, and in particular relates to a method and system for tightening bolts in an electrolytic cell based on collaborative control. Background Technology

[0002] As a core piece of equipment in fields such as water electrolysis for hydrogen production, the assembly quality of the electrolyzer directly affects electrochemical efficiency, operational safety, and product purity. During the manufacturing process of the electrolyzer, the fastening process of the electrode assembly and sealing structure is a crucial factor determining the reliability of the equipment.

[0003] Current fastening processes employ an experience-driven static fastening model, lacking a dynamic correlation mechanism between bolt preload and gasket compression ratio. When the fastening force deviates from the design value by a certain range, multiple failures can easily occur. For example, overload can lead to excessive gasket compression ratio, resulting in plastic deformation and loss of elastic compensation function, causing mechanical damage to the diaphragm; insufficient force can lead to large chamber gaps, causing problems such as leakage in the electrolytic cell during operation. Current processes rely on manual visual inspection and discrete sampling, and key parameters such as bolt elongation and gasket strain field are not included in real-time monitoring. There is a lack of real-time monitoring and feedback on bolt tightening status, making intelligent control impossible. The circumferential distribution and large number of bolts in the electrolytic cell make multi-axis collaborative control difficult, resulting in low tightening efficiency. Usually, only torque control is considered, without taking into account the impact of temperature and stress on bolt tightening effect. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a method and system for tightening bolts in an electrolytic cell based on collaborative control. The tightening force of this invention is within a threshold range, avoiding problems such as excessive gasket compression due to stress overload and large gaps in the small chambers due to insufficient force. Furthermore, when controlling the tightening force, the real-time effects of temperature and stress on the bolt tightening effect are considered.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: In a first aspect, the present invention provides a method for tightening bolts in an electrolytic cell based on collaborative control, comprising: Acquire image, stress, and temperature information of the electrolytic cell bolts; Feature vectors are obtained by extracting features from image information, stress information, and temperature information. Based on the feature vector and the preset threshold dynamic adjustment model, the stress threshold range is obtained; wherein, the threshold dynamic adjustment model is a federated learning algorithm, which dynamically optimizes the threshold dynamic adjustment model through gradient update rules, so that the threshold dynamic adjustment model can adapt to different temperature and stress environments, and dynamically adjust the weights through the variance of different working condition parameters. Based on the stress threshold range, all bolts in the electrolytic cell are controlled in a coordinated manner to ensure that the tightening force is within the threshold range.

[0006] Furthermore, when extracting features from image information, the electromagnetic wave-matter interaction characteristics of optical images are quantitatively characterized, and the continuous light intensity distribution is converted into a discrete 64-dimensional feature vector; when extracting features from temperature and stress information, the nonlinear creep response and thermal expansion effect of materials under time-varying temperature fields are captured, and the complete time series characteristics are preserved.

[0007] Furthermore, the features extracted from image information, as well as the features extracted from temperature and stress information, are fused: tensor splicing operation is used to couple the 64-dimensional optical feature space with the 128-dimensional mechanical-thermodynamic feature space in terms of feature dimensions, constructing a 192-dimensional cross-modal feature space and obtaining feature vectors.

[0008] Furthermore, when dynamically adjusting the model by dynamically optimizing the threshold through the gradient update rule, the gradient iteration formula used is as follows: ; in, i Represents model parameters; i g Represents global model parameters; J(θ): target loss function; This is the gradient operation with respect to the parameter θ; Let be the expected value of the data distribution D; p(y|x;θ) be the probability of output y given input x and parameter θ; λ is the regularization coefficient, which controls the degree of dependence on global parameters; It is the square of the L2 norm, which measures the difference between local and global parameters.

[0009] Furthermore, in the electrolytic cell system, models at different workstations are allowed to adapt to local temperature and stress environments while maintaining consistency with global knowledge; global consistency and local adaptability are balanced through the λ parameter; distributed learning is achieved, thereby improving the robustness and adaptability of the system.

[0010] Furthermore, when dynamically adjusting weights based on the variance of different operating parameters, the dynamic weight allocation mechanism is as follows: ; in, For the first i The dynamic weights of each feature can be used to find the optimal parameters that allow the algorithm to converge by adaptively fine-tuning the parameters. For the first i Data distribution of each feature; Features d i The variance; This represents the total variance of all features.

[0011] Secondly, the present invention also provides an electrolytic cell bolt tightening system based on collaborative control, comprising: The data acquisition module is configured to acquire image information, stress information, and temperature information of the electrolytic cell bolts; The feature extraction module is configured to extract features from image information, stress information, and temperature information to obtain feature vectors. The dynamic adjustment module is configured to: obtain the stress threshold range based on the feature vector and the preset threshold dynamic adjustment model; wherein, the threshold dynamic adjustment model is a federated learning algorithm, which dynamically optimizes the threshold dynamic adjustment model through gradient update rules, so that the threshold dynamic adjustment model can adapt to different temperature and stress environments, and dynamically adjust the weights through the variance of different working condition parameters. The control module is configured to coordinate the control of all bolts in the electrolytic cell based on the stress threshold range, so that the tightening force is within the threshold range.

[0012] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the electrolytic cell bolt tightening method based on cooperative control described in the first aspect.

[0013] Fourthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the steps of the electrolytic cell bolt tightening method based on cooperative control described in the first aspect.

[0014] Fifthly, the present invention also provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the steps of the electrolytic cell bolt tightening method based on cooperative control described in the first aspect.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: In this invention, firstly, feature vectors are extracted from image information, stress information, and temperature information. Then, based on the feature vectors and a preset threshold dynamic adjustment model, a stress threshold range is obtained. Finally, based on the stress threshold range, all bolts in the electrolytic cell are controlled collaboratively to ensure that the tightening force does not exceed the stress threshold. The threshold dynamic adjustment model is a federated learning algorithm that dynamically optimizes the model through gradient update rules, adapting it to different temperature and stress environments, and dynamically adjusting weights based on the variance of different operating parameters. By keeping the tightening force within the threshold range, problems such as excessive gasket compression due to stress overload and large chamber gaps due to insufficient force are avoided. Furthermore, the real-time impact of temperature and stress on bolt tightening is considered when controlling the tightening force. Attached Figure Description

[0016] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0017] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the system structure of Embodiment 1 of the present invention; Figure 3 This is a system flowchart of Embodiment 1 of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] Example 1: In the manufacturing process of electrolytic cells, the fastening process between the electrode assembly and the sealing structure is a critical factor determining the reliability of the equipment. As described in the background section, the existing technology has the following technical problems: 1. Lack of dynamic control technology for fastening force: Existing processes employ an experience-driven static fastening mode, lacking a dynamic correlation mechanism between bolt preload and gasket compression rate. When the fastening force deviates from the design value by ±15%, multiple failures are easily triggered: overload causes gasket compression rate >60% (GB / T 5782 standard allows a range of 20-50%), resulting in plastic deformation and loss of elastic compensation function, causing mechanical damage to the diaphragm; insufficient force leads to large chamber gaps, causing problems such as leakage from the electrolytic cell during operation (ISO 15848 standard allows a leakage rate of...). ).

[0021] 2. Disconnect between process data and quality control system: Current processes rely on manual visual inspection and discrete sampling, and key parameters such as bolt elongation and gasket strain field are not included in the real-time monitoring system. Furthermore, the existing system lacks real-time monitoring and feedback on bolt tightening status, making intelligent control impossible.

[0022] Insufficient multi-axis collaborative control and limited control dimensions: Electrolytic cell bolts are circumferentially distributed and numerous, making multi-axis collaborative control difficult in traditional systems. This results in low tightening efficiency. Traditional systems typically focus only on torque control, neglecting the impact of position, temperature, and stress on bolt tightening. There is a lack of comprehensive control over position, stress, and temperature. In summary, existing electrolytic cell fastening technology suffers from three major technical problems: insufficient multi-axis collaborative control, lack of data for dynamic adjustment methods, and weak drive capability, leading to low product yield.

[0023] To address at least one of the aforementioned problems, this embodiment provides a method for tightening electrolytic cell bolts based on collaborative control, aiming to overcome the technical difficulties of traditional experience-driven processes. The method integrates visual control, position control, stress control, and temperature control to solve at least one problem in the prior art, achieving intelligent and precise control of the electrolytic cell tightening process. The method includes: S1 Bolt position recognition: The position of the bolts in the electrolytic cell can be identified through a preset visual positioning system, and the position information can be transmitted to a preset intelligent control system.

[0024] S2. Multi-axis coordinated control: The intelligent control system plans the tightening path based on the bolt position information and controls the preset multi-axis tightening device to move in coordination.

[0025] Optionally, when planning the tightening path, an intelligent path optimization algorithm based on multiphysics coupling analysis is adopted, and the specific process is as follows: Phase 1: Bolt Distribution Identification and Modeling First, a visual positioning system is used to identify the spatial coordinates of all bolts on the electrolytic cell, establishing a geometric model of the bolt distribution. The system employs an improved Hough circle detection algorithm to identify bolt hole positions and converts the bolt positions into a radial distribution pattern with the center of the electrolytic cell as the origin through polar coordinate transformation. For typical circumferentially uniformly distributed bolts, the system automatically calculates the angular intervals and radial distances between each bolt, providing an accurate geometric basis for subsequent path planning.

[0026] Phase Two: Sequence Planning Based on Mechanical Equilibrium The path planning employs a three-stage strategy: "symmetrical pre-tightening - star-shaped cross-tightening - spiral tightening." First, diagonal pre-tightening is performed, selecting geometrically symmetrical bolt pairs for initial tightening to prevent warping of the electrolytic cell. During the pre-tightening stage, the torque is controlled at 30-40% of the target value to ensure uniform stress distribution on each bolt. Then, a star-shaped cross-tightening pattern is used, tightening the bolts sequentially at fixed angular intervals (typically 120° or 90°) to achieve uniform stress distribution. Finally, a spiral tightening strategy is employed, gradually increasing the torque of each bolt to the target value along the circumference to ensure final fastening quality.

[0027] Phase 3: Multi-axis collaborative optimization For multi-axis tightening devices, the system assigns bolts to different tightening axes according to their spatial location, enabling parallel operation. The path planning algorithm considers the working range and motion interference constraints of each tightening axis, optimizing the coordination timing of each axis. By establishing a time-space constraint model for multi-axis motion, tool collisions are avoided while improving tightening efficiency. The system employs a graph theory-based path optimization algorithm to transform the multi-axis coordination problem into a constrained traveling salesman problem, minimizing the total travel time and path length.

[0028] Phase 4: Dynamic Path Adjustment Mechanism During the tightening process, the system monitors feedback information from stress and temperature sensors in real time and dynamically adjusts the tightening path. When local stress concentration or temperature anomalies are detected, the path planning algorithm automatically recalculates the tightening sequence of the remaining bolts. For stress concentration areas, bolts around the affected area are prioritized to release stress; for areas with large temperature gradients, the tightening sequence is adjusted to compensate for thermal expansion effects. The dynamic adjustment mechanism employs a sliding window prediction algorithm to predict the optimal path for subsequent operations based on the tightening results of preceding bolts.

[0029] Phase 5: Temperature Compensation Path Optimization Considering the impact of temperature variations in the electrolytic cell's working environment on the tightening effect, a temperature compensation mechanism is integrated into the path planning. The system calculates the thermal expansion at each bolt location based on temperature field distribution data collected by temperature sensors, and adjusts the tightening sequence and target torque accordingly. For areas with higher temperatures, the initial tightening torque is appropriately reduced to compensate for subsequent thermal expansion; for areas with large temperature gradients, a segmented tightening strategy is adopted, allowing the temperature to homogenize before final tightening. The temperature compensation algorithm establishes a coupled relationship model of temperature, stress, and torque based on the material's coefficient of thermal expansion and elastic modulus.

[0030] Phase 6: Safety Constraints and Anomaly Handling The path planning process strictly considers material mechanics safety constraints, including maximum stress limits, fatigue strength limits, and stress gradient control. The system establishes a safety threshold monitoring mechanism; when a predicted tightening operation may lead to stress exceeding limits, it automatically adjusts tightening parameters or inserts a buffer step. For detected anomalies, such as bolt stripping or washer damage, the system initiates an emergency replanning procedure to recalculate the optimal path to ensure overall tightening quality. The anomaly handling mechanism employs a decision tree algorithm based on an expert system, selecting appropriate handling strategies according to the anomaly type and severity.

[0031] This intelligent path planning method achieves high-precision, high-efficiency, and high-reliability control of electrolytic cell bolt tightening through multi-physics coupling analysis, dynamic feedback adjustment, and safety constraint optimization, significantly improving the assembly quality and operational stability of the electrolytic cell.

[0032] S3. Precise Position Control: The position of the tightening shaft is monitored in real time by a preset displacement sensor to ensure accurate positioning of the multi-axis tightening device.

[0033] S4. Precise Torque Control: During the tightening process, a preset torque control module monitors the torque value in real time and performs precise control according to a preset torque curve. Optionally, precise control based on the preset torque curve includes: Torque curve design principle: The preset torque curve is designed based on the physical characteristics of bolt tightening, and is divided into four typical stages: free screwing stage, linear rise stage, yield transition stage, and target holding stage. The mathematical model of the torque curve is a piecewise function, with each stage having a different torque-angle relationship and control strategy. The tightening control of an M12 bolt is used as an example for illustration: S4.1 Torque curve parameter settings: For M12 stainless steel bolts (material 316L, yield strength 205MPa) commonly used in electrolytic cells, the preset torque curve parameters are as follows: Free tightening stage: torque range 0-5 N·m, angle range 0-180°. Linear rise stage: torque range 5-35 N·m, angle range 180-540°. Yield transition stage: torque range 35-42 N·m, angle range 540-600°. Target holding stage: torque stabilizes at 42±1 N·m.

[0034] S4.2 Real-time torque monitoring and control: During tightening, the torque sensor collects real-time torque values ​​at a frequency of 1000Hz, and the control system compares the measured torque with a preset curve in real time. When the measured torque deviates from the preset curve by more than ±3%, the system immediately adjusts the motor speed. During the free tightening phase, the motor runs at a constant speed of 300rpm; after entering the linear increase phase, the system switches to torque closed-loop control mode, adjusting the speed according to the torque deviation. The control algorithm uses a PID controller with a proportional coefficient Kp=0.8, an integral coefficient Ki=0.3, and a derivative coefficient Kd=0.1.

[0035] S4.3, Dual monitoring of angle and torque: In addition to torque control, the system synchronously monitors the bolt rotation angle, recording angle changes via a high-precision encoder. During the linear rise phase, the system verifies the linear relationship between torque and angle increments, with the slope maintained within the range of 0.083 N·m / °. If an abnormal slope is detected (such as bolt stripping causing a decrease in torque instead of an increase), the system immediately stops tightening and issues an alarm. During the yield transition phase, the angle increment significantly exceeds the torque increment; at this point, the system reduces the tightening speed to 50 rpm to ensure a smooth transition to the target torque.

[0036] S4.4 Temperature Compensation Control: Considering the influence of temperature on bolt material properties, the system dynamically adjusts the target torque based on temperature sensor data. The compensation formula is: T_compensated = T_target × (1 + α × ΔT), where α is the temperature compensation coefficient of 0.001 / °C, and ΔT is the difference between the actual temperature and the standard temperature (20°C). For example, when the ambient temperature is 35°C, the target torque is adjusted to 42 × (1 + 0.001 × 15) = 42.63 N·m.

[0037] S4.5 Multi-axis synchronous control: For the four-axis tightening device, the system ensures that each axis executes synchronously according to the same torque curve. Through a master-slave control architecture, the master controller sends unified torque commands to each slave controller. Each axis executes PID control independently but receives synchronization signals from the master controller. When the tightening progress of any axis lags behind by more than 10°, the other axes automatically decelerate and wait, ensuring that all axes reach the target torque synchronously. The synchronization control time window is set to ±0.5 seconds, meaning that all axes must complete the same stage of tightening within 0.5 seconds.

[0038] S4.6 Anomaly Detection and Handling: The system has built-in multiple anomaly detection mechanisms. Torque drop detection: When the torque value drops by more than 5 N·m within 0.1 seconds, it is determined that the bolt is stripped or broken.

[0039] Torque overload detection: When the torque value exceeds the set value by 120%, it is determined to be overload tightening.

[0040] Angle anomaly detection: When the angle increment deviates from the expected value by more than 15°, it is determined that the thread is damaged.

[0041] Vibration anomaly detection: Vibration during the tightening process is monitored by an accelerometer. When the vibration amplitude exceeds 0.5g, it is determined to be an abnormality of the tool or workpiece.

[0042] S4.7 Quality Assessment and Recording: After tightening is completed, the system automatically generates a quality assessment report, including the final torque value and its deviation from the target value, the total tightening angle and the angle distribution of each stage, the maximum torque fluctuation range during the tightening process, the temperature compensation amount and actual compensation effect, and abnormal event records and handling results.

[0043] The quality assessment uses a comprehensive scoring mechanism with a maximum score of 100 points. Torque accuracy accounts for 40 points, angle consistency for 30 points, process stability for 20 points, and anomaly handling for 10 points. A score of 90 or above is excellent, 80-90 is good, 70-80 is acceptable, and below 70 is unacceptable, requiring retightening.

[0044] S5. Real-time stress monitoring: Real-time monitoring of stress changes during bolt tightening using a preset stress sensor to ensure the safety of bolt connections.

[0045] S6. Real-time temperature monitoring: The system monitors the temperature changes of the bolt and the surrounding environment in real time through a preset temperature sensor, and adjusts the tightening parameters based on the temperature data.

[0046] S7. Real-time data acquisition and feedback: The preset data acquisition module collects data such as torque, angle, position, stress and temperature during the tightening process in real time, and transmits the data to the preset central controller for analysis and processing.

[0047] S8. Intelligent Adjustment and Alarm: Based on real-time data, the intelligent control system automatically adjusts the tightening parameters and issues an alarm message when an abnormality occurs.

[0048] The optional content of step S8 is: S8.1. Feature extraction is performed on image, stress, and temperature information to obtain feature vectors. This feature extraction is implemented using Python, and after quantization, the extracted features are input into the model. When extracting features from image information, the ResNet50 network (input channel 3 corresponds to RGB three-color light waves, output 64-dimensional feature space) is used to quantify the electromagnetic wave-matter interaction features of optical images, converting the continuous light intensity distribution into a discrete 64-dimensional feature vector.

[0049] When extracting features from temperature and stress information, the stress-temperature time series data collected by the sensor is processed by an LSTM network (128 memory units in the hidden layer) to capture the nonlinear creep response and thermal expansion effect of the material under the action of time-varying temperature field, and retain the complete time series features.

[0050] Features extracted from image information, as well as features extracted from temperature and stress information, are fused: Tensor splicing operation is used to couple the optical feature space (64-dimensional) with the mechanical-thermodynamic feature space (128-dimensional time step × feature dimension) in the feature dimension, constructing a 192-dimensional cross-modal feature space and obtaining feature vectors, providing a unified representation for subsequent physical field coupling analysis.

[0051] Specifically, the optical image features extracted by ResNet50 can capture the microscopic details of electromagnetic wave-matter interactions, while the LSTM network focuses on the macroscopic dynamic response of the mechanical-thermodynamic time series. The combination of the two achieves complementarity of spatial and temporal dimensions. The concatenation of 64-dimensional visual features and 128-dimensional temporal features preserves the local characteristics of optical information while integrating the global temporal regularity of mechanical response. The 192-dimensional fused feature space provides standardized input for multiphysics analysis, solving the problem of dimensional mismatch between optical images and mechanical data in traditional methods. Tensor concatenation avoids information loss that may be caused by feature weighted summation, ensuring the integrity of each modality feature. The deep convolutional structure of ResNet50 has a strong filtering capability for optical noise, while LSTM can suppress short-term fluctuations in sensor temporal data. The fused features are more adaptable to complex environmental changes, and this architecture is particularly suitable for the analysis of nonlinear behavior of materials under varying temperature fields. A unified feature space facilitates joint optimization in subsequent network layers. Compared to feature-level fusion schemes that process features in stages, it is easier to achieve end-to-end training. The 192-dimensional vector can be directly input into fully connected layers or attention mechanisms for physical field coupling modeling. Table 1 shows the standardization of heterogeneous data: Table 1 Data Standardization

[0052] S8.2. Based on the feature vector and the preset threshold dynamic adjustment model, the stress threshold range is obtained; wherein, the threshold dynamic adjustment model is a federated learning algorithm (GIN), which dynamically optimizes the threshold dynamic adjustment model through gradient update rules, so that the threshold dynamic adjustment model can adapt to different temperature and stress environments, and dynamically adjust the weights through the variance of different working condition parameters. When dynamically adjusting the model by dynamically optimizing the threshold using the gradient update rule, the gradient iteration formula used is as follows: ; in, i These represent the model parameters, specifically the neural network weights and bias parameters for the current client (tightening workstation). i g ∠_θ represents the global model parameters; ∠_θ represents the gradient operation on the parameters θ; E_{d~D} represents the expected value of the data distribution D; p(y|x; θ) represents the probability of output y given input x and parameters θ; λ represents the regularization coefficient, which controls the degree of dependence on global parameters; ||θ - θ_g||_2^2 represents the square of the L2 norm, which measures the difference between local parameters and global parameters.

[0053] Specifically, the local model parameters θ include all trainable parameters of the visual feature extraction network, the LSTM temporal network, and the fusion network. In the electrolytic cell system, these parameters reflect the adaptability of the current workstation to specific operating conditions. Global model parameters... i g The global model parameters, aggregated by the server using a federated averaging algorithm, incorporate the knowledge and experience of all tightening workstations involved in training, reflecting the general rules and best practices of electrolytic cell tightening. The target loss function J(θ) measures the performance of the current model parameters θ. In tightening systems, it is typically the mean square error loss between the predicted torque and the actual torque; a smaller value indicates more accurate model prediction. The gradient operator ▽_θ represents the partial derivative with respect to the parameter θ, indicating the direction and magnitude of parameter optimization. The gradient direction points to the direction of the fastest growth of the loss function; the negative gradient direction is the optimization direction. E_{d~D} - the expectation of the data distribution; E represents the mathematical expectation (mean) operation. d~D indicates that the data sample d follows a distribution D. In practical applications, the expectation is approximated by the average of a small batch of data. The conditional probability density p(y|x;θ) is the probability distribution of the output y given the input features x and model parameters θ; x: the fused 192-dimensional feature vector (including visual, stress, and temperature features); y: the predicted stress threshold range, reflecting the model's confidence in the stress threshold prediction in tightening systems. The log-likelihood function log p(y|x;θ) is the logarithmic form of the probability density function, which facilitates numerical calculation and gradient solving. Maximizing the log-likelihood is equivalent to maximizing the original probability, but the value is more stable, corresponding to the cross-entropy loss or mean squared error loss in neural networks. The regularization coefficient λ (lambda) controls the penalty strength for the local parameter θ deviating from the global parameter θ_g. Its value is typically in the range of 0.001-0.1, adjusted according to data heterogeneity. A larger λ value results in a local model that is closer to the global model; a smaller λ value indicates stronger local adaptability. Regularization term. It is used to calculate the square of the Euclidean distance between local parameters and global parameters, preventing the local model from deviating too much from global knowledge and maintaining the consistency of federated learning.

[0054] A balance between local adaptability and global consistency is achieved through the λ parameter: when λ=0, optimization is based entirely on local data, which may lead to overfitting; when λ→∞, the local model is completely equivalent to the global model, losing its adaptability; an appropriate λ value allows the model to adapt to local conditions while maintaining global best practices.

[0055] The specific design of the threshold dynamic adjustment model to adapt to different temperature and stress environments: The model uses a federated learning framework to share tightening experience under different temperature and stress environments. The global parameter θ_g contains the optimal control strategy under various operating conditions, while the local parameter θ is fine-tuned for the current environmental conditions. The λ parameter is dynamically adjusted according to environmental changes: the λ value is larger under standard operating conditions to maintain consistency with global experience; the λ value decreases under extreme operating conditions, allowing for more local adaptive adjustments.

[0056] In this embodiment, the regularized gradient update formula for federated learning comprises two parts: First item: ▽ θ log p(y|x; θ), the standard likelihood gradient, is used to maximize the model's performance on local data. The second term, λ▽_θ||θ - θ_g||_2^2, is the gradient of the regularization term, preventing the local model parameters from deviating too far from the global model; λ controls the strength of this constraint.

[0057] In electrolytic cell systems, models at different workstations can adapt to local temperature and stress environments while maintaining consistency with global knowledge. By controlling the model's dependence on global parameters through λ, transfer learning (i.e., edge devices perform personalized optimization based on global knowledge) can be achieved.

[0058] When dynamically adjusting weights based on the variance of different operating parameters, the dynamic weight allocation mechanism is as follows: ; in, For the first i Dynamic weights of each feature; For the first i Data distribution of each feature; Features d i The variance; This represents the total variance of all features.

[0059] Specifically, This represents the importance coefficient of the i-th feature (visual, stress, or temperature) in the fusion process. The weight value ranges from 0 to 1, and the sum of the weights of all features equals 1 (normalized). The larger the weight, the greater the influence of the feature on the final decision. Let d_visual represent the data distribution of the i-th feature, and d_k represent the data set of the i-th feature over time. For visual features: d_visual = {f_1, f_2, ..., f_n}, where f_k is the 64-dimensional visual feature vector at time k. For stress features: d_stress = {s_1, s_2, ..., s_n}, where s_k is the stress sensor data at time k. For temperature features: d_temp = {t_1, t_2, ..., t_n}, where t_k is the temperature sensor data at time k.

[0060] feature d i variance The formula used to measure the volatility and uncertainty of the i-th feature data is as follows: , where μ_i is the mean of feature d_i; the larger the variance, the more drastic the change in the feature and the richer the information it contains.

[0061] Total variance of all features The variance of all features (j = visual, stress, temperature) is summed and used as a normalization factor to ensure that the sum of all weights is 1.

[0062] The greater the influence of the operating condition parameter (the higher the variance), the greater its weight; the weight is dynamically adjusted by calculating the variance of different operating condition parameters. This ensures that the tightening system can be reasonably optimized under different working conditions.

[0063] The methods and differences in determining the variance of "image information, stress information, and temperature information": 1. Method for determining the variance of image information: Calculation method: The variance of image information is determined by statistical analysis of the 64-dimensional visual feature vector over time. Specifically, images of the same bolt location are acquired within T consecutive time windows. Each image is processed by a ResNet50 network to extract a 64-dimensional feature vector. The variance of each feature dimension over time is calculated, and the average value is taken as the overall variance of the image information.

[0064] Analysis of sources of variance: Illumination variation variance: Fluctuations in ambient light intensity cause changes in pixel brightness values, which are reflected in the feature space as periodic changes in low-frequency components.

[0065] Geometric deformation variance: Small shifts in camera viewpoint or changes in the angle of reflection on the bolt surface cause positional shifts in edge detection features.

[0066] Noise interference variance: Thermal noise and electromagnetic interference from CCD sensors introduce random fluctuations in high-frequency characteristic components.

[0067] Physical meaning: Image variance Var(d_visual) reflects the stability of an optical measurement system. When the optical conditions of the working environment are complex, the image features of the same physical target at different times show large differences. At this time, the system needs to enhance the weight of visual processing to improve positioning accuracy.

[0068] 2. Method for determining the variance of stress information: Calculation Method: The variance of stress information is determined by statistical analysis of real-time data from multi-point stress sensors during the tightening process. Within each bolt tightening cycle, the system collects data from N stress sensors at a frequency of 1000Hz, forming a time series matrix. The time variance of each sensor's measurement is calculated, and then a spatially weighted average is used to obtain the overall variance of the stress information.

[0069] Analysis of sources of variance: Material creep variance: The time-dependent plastic deformation of metallic materials under sustained load, which is characterized by the slow release and redistribution of stress.

[0070] Load transfer variance: The force transmission path changes during bolt tightening, causing the stress response at different measuring points to exhibit nonlinear changes.

[0071] Temperature coupling variance: The interaction between the thermal expansion effect of the material and the stress field generates thermal stress components when a temperature gradient exists.

[0072] Physical meaning: Stress variance Var(d_stress) reflects the dynamic characteristics of a solid mechanical system. When the stress distribution is uneven or close to the material limit, the variance value increases, and the system automatically increases the weight of stress control to prioritize structural safety and the rationality of load distribution.

[0073] 3. Method for determining the variance of temperature information: Calculation Method: The variance of temperature information is determined by two-dimensional statistical analysis of the spatial-temporal data of a distributed temperature sensor network. The system deploys M temperature sensors at key locations in the electrolytic cell, continuously collecting temperature data within a time window T, forming an M×T temperature field matrix. The spatial variance (temperature difference at different locations at the same time) and temporal variance (temperature change at different times at the same location) are calculated separately, and the comprehensive variance of the temperature information is obtained by weighted summation.

[0074] Analysis of sources of variance: Thermal conduction delay variance: Differences in thermal conduction rates caused by different materials and geometries result in a spatial gradient in temperature distribution.

[0075] Heat source distribution variance: The location and intensity of frictional heat generation during tightening are uneven, resulting in local hot spots and temperature fluctuations.

[0076] Environmental disturbance variance: The influence of changes in ambient temperature and air convection on the local temperature field.

[0077] Physical meaning: Temperature variance Var(d_temp) reflects the degree of non-equilibrium of a thermodynamic system. When the temperature field is unevenly distributed or changes drastically, the thermal expansion effect of the material has a significant impact on the tightening quality, and the system accordingly increases the weight of temperature compensation control.

[0078] The underlying effect of image feature changes: Low variance state (Var(d_visual) < threshold): The optical environment is stable, and the visual features have good reproducibility, indicating high and consistent bolt position recognition accuracy. At this point, the weight of w_visual is relatively small, and the system mainly relies on stress and temperature sensors for precise control, with the vision system primarily serving as a position confirmation mechanism. The control strategy is biased towards efficiency optimization, and the tightening speed can be appropriately increased.

[0079] High variance state (Var(d_visual) > threshold): Optical measurements are unstable and may be affected by interference factors such as occlusion, reflection, or uneven illumination. In this state, the weight of w_visual increases, the system strengthens image processing and filtering algorithms, and employs multi-frame image fusion technology to improve positioning robustness. The control strategy prioritizes stability, reducing tightening speed to allow more time for visual correction.

[0080] The fundamental effect of stress characteristic changes: Low variance state (Var(d_stress) < threshold): Stress distribution is uniform and stable, the material is within the elastic deformation range, and the bolt connection is in good condition. At this time, the weight of w_stress is relatively small, and the system performs tightening according to the standard torque curve, focusing on positional accuracy and temperature compensation. The control strategy emphasizes efficiency and can use a higher tightening speed.

[0081] High variance state (Var(d_stress) > threshold): Stress fluctuations are severe, potentially indicating stress concentration, material defects, or abnormal tightening. In this state, the weight of w_stress increases significantly, the system activates a safety protection mode, monitors the stress change rate in real time, and pauses tightening for stress release if necessary. The control strategy shifts to safety priority, employing staged tightening and stress relaxation techniques.

[0082] The fundamental effect of temperature characteristic changes: Low variance state (Var(d_temp) < threshold): The temperature field distribution is uniform and stable, and the thermal expansion effect is predictable and consistent. In this state, w_temp has a smaller weight, the system uses a standard temperature compensation coefficient, and focuses on mechanical tightening accuracy. The control strategy favors standardized operation, executing according to preset parameters.

[0083] High variance state (Var(d_temp) > threshold): Uneven temperature distribution or drastic changes occur, and thermal expansion significantly affects tightening quality. In this state, the weight of w_temp increases, and the system activates a dynamic temperature compensation algorithm to adjust the target torque value and tightening timing in real time. The control strategy shifts to adaptive mode, dynamically optimizing tightening parameters based on temperature changes.

[0084] Synergistic Effect Principle: The dynamic weighting of the three characteristic variances enables intelligent coordinated control of multiple physics. When an abnormal fluctuation occurs in a certain physical quantity, the system automatically increases the control weight of that physical quantity while adjusting the priority of other control parameters to ensure that high-quality tightening results are maintained even under complex working conditions. This adaptive weighting mechanism gives the system judgment capabilities similar to human experts, enabling it to adjust control strategies according to real-time conditions.

[0085] Deep learning modeling of visual, pressure, and temperature feature data relies on GIN for training, where θ represents the network's parameters. The model is dynamically optimized using gradient update rules to adapt to different operating conditions (such as varying temperature and pressure environments).

[0086] S8.3 Based on the stress threshold range, all bolts in the electrolytic cell are controlled in a coordinated manner to ensure that the tightening force is within the threshold range.

[0087] To achieve the above method, this embodiment also provides an electrolytic cell bolt tightening system based on collaborative control, suitable for automated tightening operations of electrolytic cell bolts. This system integrates vision control, stress control, and temperature control, enabling high-precision and high-efficiency bolt tightening operations while ensuring the safety and stability of the bolt connections. The system includes: Multi-axis tightening device: can include multiple tightening axes, each equipped with an independent servo motor, torque sensor, displacement sensor and temperature sensor, which can realize multi-axis collaborative control; through multi-axis collaborative control, tightening efficiency is significantly improved.

[0088] The intelligent control system can include a central controller, motion control module, torque control module, stress control module, temperature control module, and data acquisition module. It enables multi-axis collaborative control, precise torque control, stress monitoring, temperature monitoring, and real-time data acquisition. Integrating vision control, position control, stress control, and temperature control achieves comprehensive intelligent control; and precise control of torque, position, stress, and temperature ensures tightening quality.

[0089] Visual positioning system: may include industrial cameras and image processing modules, used to identify bolt positions and guide the tightening device to accurate positioning.

[0090] Stress monitoring system: can include stress sensors and data acquisition modules, used to monitor stress changes in real time during bolt tightening.

[0091] Temperature monitoring system: can include temperature sensors and data acquisition modules, used to monitor temperature changes of bolts and the surrounding environment in real time.

[0092] Human-machine interface: used to set tightening parameters, display tightening status and alarm information.

[0093] The method and system of this embodiment are applicable to the tightening requirements of bolts of different specifications; real-time stress monitoring and temperature monitoring ensure the safety and stability of bolt connections; and the impact of temperature changes on tightening effect is considered to improve the environmental adaptability of the system.

[0094] Example 2: This embodiment provides an electrolytic cell bolt tightening system based on collaborative control, including: The data acquisition module is configured to acquire image information, stress information, and temperature information of the electrolytic cell bolts; The feature extraction module is configured to extract features from image information, stress information, and temperature information to obtain feature vectors. The dynamic adjustment module is configured to: obtain the stress threshold range based on the feature vector and the preset threshold dynamic adjustment model; wherein, the threshold dynamic adjustment model is a federated learning algorithm, which dynamically optimizes the threshold dynamic adjustment model through gradient update rules, so that the threshold dynamic adjustment model can adapt to different temperature and stress environments, and dynamically adjust the weights through the variance of different working condition parameters. The control module is configured to coordinate the control of all bolts in the electrolytic cell based on the stress threshold range, so that the tightening force is within the threshold range.

[0095] The working method of the system is the same as the electrolytic cell bolt tightening method based on collaborative control in Embodiment 1, and will not be repeated here.

[0096] Example 3: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the electrolytic cell bolt tightening method based on cooperative control described in Embodiment 1.

[0097] Example 4: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the steps of the electrolytic cell bolt tightening method based on cooperative control described in Embodiment 1.

[0098] Example 5: This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the electrolytic cell bolt tightening method based on cooperative control described in Embodiment 1.

[0099] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for screwing bolts of a cell based on cooperative control, characterized in that, The method comprises the following steps: acquiring image information, stress information and temperature information of the electrolytic cell bolt; extracting features from the image information, stress information and temperature information to obtain a feature vector; obtaining a stress threshold range according to the feature vector and a preset threshold dynamic adjustment model; wherein the threshold dynamic adjustment model is a federated learning algorithm, the threshold dynamic adjustment model is dynamically optimized through a gradient update rule, the threshold dynamic adjustment model is adapted to different temperature and stress environments, and the weight is dynamically adjusted through the variance of different working condition parameters; based on the stress threshold range, all bolts of the electrolytic cell are cooperatively controlled so that the tightening force is within the threshold range.

2. The synergic control based electrolyzer bolt tightening method of claim 1, wherein, When the features of the image information are extracted, the electromagnetic wave-substance interaction features of the optical image are quantitatively represented, and the continuous light intensity distribution is converted into a discrete 64-dimensional feature vector; when the features of the temperature information and the stress information are extracted, the nonlinear creep response and thermal expansion effect of the material under the action of the time-varying temperature field are captured, and the complete time sequence features are retained.

3. The synergic control based electrolyzer bolt tightening method of claim 2, wherein, The features extracted from the image information and the features extracted from the temperature information and the stress information are fused: the 64-dimensional optical feature space and the 128-dimensional mechanical-thermal feature space are coupled in the feature dimension by using a tensor splicing operation, a 192-dimensional cross-modal feature space is constructed, and a feature vector is obtained.

4. The synergic control based electrolyzer bolt tightening method as claimed in claim 1 wherein, When the threshold dynamic adjustment model is dynamically optimized through the gradient update rule, the gradient iteration formula used is: ； ； where, When the weight is dynamically adjusted through the variance of different working condition parameters, the dynamic weight distribution mechanism is: represents the model parameters; When the threshold dynamic adjustment model is dynamically optimized through the gradient update rule, the specific meanings of the parameters in the gradient iteration formula are as follows: θ represents the local model parameters of the current client, including all trainable weights of the visual feature extraction network, the LSTM time sequence network and the feature fusion network; θ_g represents the global model parameters aggregated through the federated averaging algorithm, which integrates the tightening experience knowledge of all participating training workstations; E_{d~D} represents the mathematical expectation operation on the data distribution D, which is approximately calculated by the average value of a small batch of data; p(y|x; θ) represents the conditional probability density of predicting the stress threshold range y given the 192-dimensional fused feature vector x and the model parameter θ; g represents the global model parameters; J(θ): objective loss function; is the gradient operation on the parameters θ; is the expectation value on the data distribution D; p(y|x; θ) is the probability of the output y given the input x and the parameters θ; λ is the regularization coefficient, which controls the degree of dependence on the global parameters; is the square of the L2 norm, which measures the difference between the local parameters and the global parameters.

5. The synergic control based electrolyzer bolt tightening method as claimed in claim 1 wherein, λ is a regularization coefficient, the value range is 0.001-0.1, which is used to control the punishment intensity of the deviation of the local model parameters from the global model parameters; ||θ-θ_g||_2^2 is an L2 regularization term, which calculates the square of the Euclidean distance of the local parameters and the global parameters, and prevents the local model from deviating too much from the global knowledge; ； Wherein, The dynamic weight of the first i characteristic, the dynamic weight information can be found by the adaptive fine-tuning parameters of the algorithm, and the optimal parameters that make the algorithm converge; The data distribution of the first i characteristic; The variance of the characteristic d i The total variance of all characteristics.​ 6. The synergic control based electrolyzer bolt tightening method as claimed in claim 1, wherein, The threshold dynamic adjustment model adapts to different temperature and stress environments through the following mechanism: the global parameter θ_g contains the optimal control strategy under various working conditions, the local parameter θ is fine-tuned for the current environmental conditions, the regularization coefficient λ is dynamically adjusted according to the environmental changes, the λ value is larger in the standard working condition to keep consistent with the global experience, and the λ value is smaller in the extreme working condition to allow more local adaptive adjustment, and the knowledge sharing and collaborative optimization of the tightening experience under different temperature and stress environments are realized through the federated learning framework. The method comprises the following steps: ​ 7. A cell bolt tightening system based on coordinated control, characterized in that, ​ The data acquisition module is configured to acquire image information, stress information and temperature information of the electrolytic cell bolt; The feature extraction module is configured to perform feature extraction on the image information, stress information and temperature information to obtain a feature vector; The dynamic adjustment module is configured to obtain a stress threshold range according to the feature vector and a preset threshold dynamic adjustment model, wherein the threshold dynamic adjustment model is a federal learning algorithm, the threshold dynamic adjustment model is dynamically optimized through a gradient update rule, the threshold dynamic adjustment model is adapted to different temperature and stress environments, and the weight is dynamically adjusted through the variance of different working condition parameters; The control module is configured to cooperatively control all the bolts of the electrolytic cell based on the stress threshold range, so that the tightening force is within the threshold range.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the electrolytic cell bolt tightening method based on cooperative control according to any one of claims 1-6.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and capable of running on the processor, characterized in that, The processor executes the program to implement the steps of the electrolytic cell bolt tightening method based on cooperative control according to any one of claims 1-6.

10. A computer program product, characterised in that, The computer program product comprises a computer program, and the computer program is executed by the processor to implement the steps of the electrolytic cell bolt tightening method based on cooperative control according to any one of claims 1-6.

Images