Method for evaluating health state of each subsystem of rapier loom

Abstract

The invention relates to the technical field of state evaluation, and relates to a health state evaluation method for each subsystem of a rapier loom. The method comprises the following steps: 1, selecting feature parameters, extracting features based on wavelet transform, enhancing abnormal data based on GAN, and constructing a health state model; 2, performing attribute reduction and rule extraction on the evaluation indexes by using a rough set theory, and constructing an initial confidence rule base of each subsystem; 3, fusing the activated rules through an evidence reasoning algorithm, and outputting the confidence coefficient of each subsystem under each health state level; 4, data distribution stratified sampling is carried out to process missing data, and secondary fusion is carried out on all reasoning conclusions by combining a plurality of reasoning results and utilizing an ER algorithm to obtain a final reasoning result; 5, reasoning the error as a target function, and optimizing the parameters of the belief rule base by using a WOA algorithm with interpretability constraint; and step 6, updating the optimized rule base by the incremental rough set, and reserving recent data by using a sliding window, so that the model continuously learns new data. And the evaluation precision and reliability are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of condition assessment technology, and specifically to a method for assessing the health status of various subsystems of a rapier loom. Background Technology

[0002] With the diversified development of new-generation information technology, the textile industry is also moving towards intelligence and high-end development. With the improvement of industrial automation, the requirements for the reliability and stability of mechanical equipment are getting higher and higher. As a key piece of equipment, the health status assessment of rapier looms is crucial to ensuring production efficiency and product quality.

[0003] Traditional health status assessment methods rely on expert experience and regular inspections, lacking real-time accuracy and precision. Today, health status assessment of machinery is a method of "predictive maintenance," taking necessary maintenance measures by understanding the equipment's real-time operating status. However, existing health status assessment methods have the following shortcomings: Redundant information problem: Data-driven health status assessment methods (such as support vector machines, neural networks, etc.) often introduce a lot of redundant information when processing the health status assessment of rapier looms, resulting in high model complexity and low computational efficiency.

[0004] The problem of scarce abnormal data: Abnormal working conditions are rare in the actual operation of rapier looms, resulting in a lack of effective data and model bias caused by data imbalance.

[0005] Data gap issue: In actual operation, monitoring data of rapier looms are often missing due to sensor failure or human error. Existing methods are difficult to effectively handle the data gap issue, resulting in inaccurate evaluation results.

[0006] Reasoning error problem: Traditional methods (such as fuzzy reasoning, hierarchical analysis, etc.) rely on expert experience, resulting in large reasoning errors and making it difficult to adapt to the complex and ever-changing operating conditions of rapier looms.

[0007] Equipment status change issues: After long-term operation, mechanical wear and changes in operating conditions will cause static evaluation models to gradually become ineffective.

[0008] To address the aforementioned issues, Chinese Patent CN118411154B discloses a method and system for assessing the safety status of power distribution equipment. This method integrates historical data with machine learning techniques to construct a fault prediction model, improving the reliability of power supply and maintenance efficiency. However, this method relies heavily on monitoring data, while abnormal data is scarce during the operation of rapier looms. Furthermore, the data is strongly coupled and nonlinear, leading to a lack of effective data and difficulty in accurately identifying small-scale abnormal states. The model also lacks an explanation of the system's inherent change mechanisms, resulting in inaccurate assessment results.

[0009] Chinese patent CN118607065A discloses a building structure deformation prediction system based on fuzzy logic control algorithm. Based on building structure deformation data, it combines the output of a fuzzy logic controller with machine learning technology to probabilistically predict impending structural deformation. However, fuzzy inference relies on expert experience and prior knowledge, limiting its accuracy, and the model emphasizes static features while ignoring dynamic changes in the equipment.

[0010] Therefore, how to reduce the impact of uncertainties and missing effective data on the assessment results, and improve the robustness, accuracy and precision of health status assessment, is a problem that needs to be solved by those skilled in the art.

[0011] Therefore, this application proposes a method for assessing the health status of each subsystem of a rapier loom. Summary of the Invention

[0012] In order to overcome the shortcomings of the existing technology and solve the technical problems existing in the background technology, the present invention proposes a method for assessing the health status of each subsystem of a rapier loom.

[0013] This invention is achieved through the following technical solution: A method for assessing the health status of each subsystem of a rapier loom, the method being based on the whale algorithm with interpretable constraints to optimize the confidence rule base parameters and the dynamic update mechanism of the confidence rule base, includes the following steps: Step 1: Investigate the fault mechanisms of each subsystem of the rapier loom and select characteristic parameters. Then, extract features based on wavelet transform and enhance abnormal data based on GAN to construct a health status model. Step 2: Use rough set theory to reduce the attributes and extract rules for the evaluation indicators, and construct an initial confidence rule base for each subsystem of the rapier loom; Step 3: Use the evidence reasoning algorithm to fuse the activated rules and output the confidence level of each subsystem at each health status level; Step 4: Perform stratified sampling based on data distribution to process missing data, and combine multiple inference results to use the ER algorithm to fuse all inference conclusions a second time to obtain the final inference result; Step 5: Using inference error as the objective function, optimize the parameters of the confidence rule base using the WOA algorithm with interpretability constraints; Step 6: Update the optimized rule base using incremental rough set and retain recent data using a sliding window to enable the model to continuously learn new data.

[0014] Preferably, in step one, the subsystems of the rapier loom include a warp feeding subsystem, a weaving subsystem, and a take-up subsystem; the characteristic parameters include temperature signals, vibration signals, and tension signals; and the specific steps of step one are as follows: Preprocess the feature parameters; Wavelet transform is used to decompose the signal by feature parameters, extract multi-scale time-frequency features, and perform feature selection to enhance fault-sensitive information in the signal. Use GANs to generate anomalous data and balance the distribution of the training set.

[0015] Preferably, the specific steps of wavelet transform are as follows: Signal decomposition: Multi-scale wavelet packet decomposition was performed on the vibration signal of the rapier loom. The Daubechies 4 wavelet basis function was selected, and the decomposition layer was 5, resulting in 16 sub-bands. The decomposition formula is as follows: ; in, For scale parameters, For translation parameters, These are wavelet basis functions; Feature extraction: Energy entropy calculation: The energy entropy is calculated for each layer of wavelet coefficients, representing the signal complexity. , ; Multi-scale feature fusion: The energy entropy of each layer is fused with the original time-domain features to form a comprehensive feature vector, which serves as the input to the confidence rule base. Feature selection: ReliefF algorithm weight update: ; Where NH represents the nearest neighbor of the same class, NM represents the nearest neighbor of different classes, and k represents the number of samples.

[0016] Preferably, the specific steps for using GAN to generate anomalous data and balance the distribution of the training set are as follows: The generator uses a fully connected layer + deconvolution structure. The input is random noise, and the output is a feature vector of synthesized vibration, temperature, and tension signals. Discriminator: Employs a convolutional neural network, with input being real / synthetic feature vectors and output being the discrimination probability; Loss function: Use Wasserstein GAN loss to improve training stability; ; Data balancing: Abnormal samples are generated using GAN to adjust the ratio to 3:1.

[0017] Preferably, the specific steps for extracting confidence rules using the relative reduction of rough sets in step two are as follows: Divide the data into equivalence classes based on attribute reference values; Calculate the confidence of decision rules with the same conditional attributes under different decision attributes, and extract rules with higher confidence based on rough set theory; The extracted confidence rules are used for inference, and the number and spacing of the initially defined attribute reference values ​​are optimized based on the inference results.

[0018] Preferably, the specific steps of step three are as follows: Once the model receives the input information, it matches the input parameters with the antecedent of each confidence rule. The matched confidence rules are combined and weighted using an evidence reasoning algorithm to comprehensively consider the impact of each rule on the output result, thereby obtaining the confidence score of the output.

[0019] Preferably, the specific steps for processing missing data in step four are as follows: Statistical analysis of historical data is performed to determine the data distribution characteristics of missing attributes; Stratified sampling involves dividing the data distribution into several levels or intervals, and then sampling according to the proportion of data distribution within each interval. By combining the reconstructed attribute values ​​with other known attribute information, a comprehensive evaluation is conducted using a confidence rule base and evidence reasoning algorithm to ultimately determine the health status of the subsystem.

[0020] Preferably, the specific steps of step five are as follows: Initialization and expert knowledge integration: Initialize the whale population, define the search space and the maximum number of iterations; Expert knowledge-guided point selection, centered on expert experience, generates initial whale positions within their neighborhood, ensuring that the optimized starting point aligns with the actual system mechanism. A whale can be represented as: ; In the formula: The confidence level of expert knowledge; To return a A random matrix; Fitness calculation and constraint design: Calculate the fitness value for each individual whale, using mean squared error as the objective function; Interpretability constraints: Whale Behavior Simulation and Parameter Optimization: Surround the prey: Update the whale's position based on the current optimal solution to narrow the search range; Spiral hunting: Simulates the spiral approach behavior of whales towards their prey, with fine-tuned parameters in specific areas; Random search: When the exploration coefficient exceeds the threshold, the individual location is randomly selected to enhance the global search capability; Dynamic updates and verification: The rationality of the rule base is continuously verified during the iteration process, and the parameters are reset if the constraints are violated. Output the final optimized confidence rule base, ensuring that it has both high accuracy and interpretability.

[0021] Preferably, the interpretability constraints include: Constraint 1: The optimized rule confidence distribution must be consistent with the actual health status of each system of the rapier loom; ; In the formula: Under the k-th rule, the interpretability constraint of the confidence distribution is denoted as Ek, which is determined by the analysis of each actual system and has no fixed form; K is the number of confidence rules. Constraint 2: The adjustment range of the rule confidence level must be within a reasonable range preset by the experts; ; In the formula: The k-th confidence level of the n-th rule; , These represent the minimum and maximum confidence levels given by each expert, respectively.

[0022] Preferably, in step six, the specific steps of incremental rough set update are as follows: When new data arrives, only the confidence and weight of the affected rules are updated, using the following formula: ; The sliding window mechanism retains data from the most recent 7 days, while older data decays exponentially due to forgetting factors. The dynamic adjustment rule weights are as follows: .

[0023] The beneficial effects of this invention are: 1. This invention significantly improves the accuracy and reliability of health status assessment of various subsystems of rapier looms by integrating wavelet transform feature extraction, GAN data augmentation, Whale Algorithm with interpretability constraints (WOA-e), and dynamic update mechanism. This invention makes the method of this invention have high precision, strong adaptability and transparent decision-making capabilities, providing an efficient solution for intelligent maintenance of textile equipment.

[0024] 2. This invention utilizes wavelet transform to extract multi-scale features of signals and combines them with GAN to generate abnormal data, thereby increasing the fault detection rate from 82% to 95% and effectively solving the problems of data noise and imbalance.

[0025] 3. This invention introduces expert knowledge constraints to optimize the confidence distribution of rules, ensuring model interpretability and reducing inference error to 0.8% (compared to 3.2% for traditional methods).

[0026] 4. The dynamic incremental update and sliding window mechanism of this invention (7-day window, decay factor β=0.001) keeps the rule base size stable at 200±10 rules, improves the response speed to sudden operating conditions by 50%, and reduces long-term error fluctuation to <1%. 5. The stratified sampling and ER secondary fusion technology of this invention maintains a low error of 2.1% even when 30% of the data is missing, and the robustness is improved by 83%. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the design process of the present invention; Figure 2 A health status assessment model diagram for rapier looms; Figure 3 This is a flowchart of signal feature extraction based on wavelet transform; Figure 4 This is a flowchart of a GAN-based balanced dataset. Figure 5 This is a schematic diagram of the confidence rule extraction process based on rough sets. Figure 6 A flowchart for state reasoning with missing prerequisite attribute information; Figure 7 The flowchart of the WOA algorithm with interpretability constraints is shown below. Figure 8 A comparison diagram of the state inference and actual values ​​of the BRB (Bill Delivery System) subsystem; Figure 9 A comparison diagram of the state inference and actual values ​​of the BRB-WOA-e subsystem for the delivery of the scripture; Figure 10 A comparison diagram of the BRB-WOA-e state inference and actual values ​​for the weaving subsystem; Figure 11 Comparison of BRB-WOA-e state inference and true values ​​for the convolution subsystem. Detailed Implementation

[0028] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Experimental methods in the following embodiments, unless otherwise specified, are generally performed under conventional conditions or as recommended by the manufacturer.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of skill in the art. The reagents and raw materials used in this invention are readily available through conventional means, and unless otherwise specified, they shall be used in accordance with conventional methods in the art or as per the product instructions. Furthermore, any methods and materials similar to or equivalent to those described herein may be applied to the methods of this invention. The invention will now be further described with reference to the accompanying drawings and specific embodiments. The preferred embodiments and materials described herein are for illustrative purposes only.

[0030] Health status assessment methods for various subsystems of a rapier loom, such as... Figure 1 As shown, it includes the following steps: The fault mechanisms of each subsystem of the rapier loom were investigated and characteristic parameters were selected. Feature extraction was performed based on wavelet transform, and a health status model was constructed based on GAN-enhanced abnormal data. Based on the research on the mechanical structure, process flow and functional realization of rapier looms, this invention divides the complete rapier loom equipment into three core functional systems: warp feeding subsystem, weaving subsystem, and take-up subsystem.

[0031] In order to effectively monitor and assess the health status of looms, it is necessary to select sensitive and representative parameters for monitoring.

[0032] Therefore, based on expert evaluation and actual operational data, temperature, vibration, and tension were identified as key monitoring indicators affecting weaving performance and the normal operation of each subsystem. A health status assessment model was constructed from three perspectives: the warp feeding system, the weaving system, and the take-up system of the rapier loom. For example... Figure 2 As shown.

[0033] Add a data preprocessing step before inputting the data into the confidence rule base.

[0034] First, wavelet transform is performed on the raw monitoring data to extract the time-frequency features of the signal. Feature enhancement allows for better capture of key information in the data, improving the accuracy of subsequent assessments. The details of the wavelet transform-based signal feature extraction technique are as follows, and its flowchart is shown below. Figure 3 As shown.

[0035] ① Signal Decomposition: Multi-scale wavelet packet decomposition was performed on the vibration signal of the rapier loom. The Daubechies 4 (db4) wavelet basis function was selected, and the decomposition level was 5, resulting in 16 sub-bands. The decomposition formula is as follows: ; in, For scale parameters, For translation parameters, These are wavelet basis functions.

[0036] ② Feature extraction: Energy entropy calculation, calculating the energy entropy for each layer of wavelet coefficients to characterize signal complexity: , ; Multi-scale feature fusion: The energy entropy of each layer is fused with the original time-domain features (mean, variance) to form a comprehensive feature vector, which serves as the input to the confidence rule base.

[0037] ③ Feature selection: ReliefF algorithm weight update: ; Where NH represents the nearest neighbor of the same class, NM represents the nearest neighbor of different classes, and k represents the number of samples.

[0038] Then, a GAN is used to generate anomalous data to balance the distribution of the training set. By generating 200 sets of anomalous data (such as sudden tension changes and bearing high temperatures), the proportion of faulty samples is increased from 5% to 30%, thus solving the data imbalance problem. The flowchart is as follows: Figure 4 As shown, the technical details are as follows: The generator uses a fully connected layer + deconvolution structure. The input is random noise (100 dimensions), and the output is a feature vector of synthesized vibration, temperature, and tension signals.

[0039] Discriminator: Employs a convolutional neural network (CNN), with inputs being real / synthetic feature vectors and outputs being the discrimination probability.

[0040] Loss function: Use Wasserstein GAN (WGAN) loss to improve training stability. ; Data balancing: To address the problem of scarce abnormal data (normal:abnormal = 10:1), abnormal samples are generated using GAN to adjust the ratio to 3:1.

[0041] Rough set theory is used to perform attribute reduction and rule extraction on evaluation indicators to construct an initial confidence rule base for each subsystem of the rapier loom; The process of extracting confidence rules based on rough sets is as follows: Figure 5 As shown.

[0042] Taking the establishment of a confidence rule base for the warp delivery subsystem as an example, the health status assessment indicators of the warp delivery system, i.e., the prerequisite attribute parameters, are four: warp tension (WT), warp delivery motor vibration frequency (FD), bearing temperature signal of the reduction gear (BT), and spindle vibration signal (FS). Simultaneously, the health status level of the warp delivery subsystem {healthy, sub-healthy, pathological, faulty} is used as the output of the confidence rule base, i.e., the result attribute.

[0043] In the experiment, 500 sets of preprocessed data (wavelet transform features, GAN anomaly enhancement data) were used as the training set for rule extraction, and 300 sets of data were used as the test set to assess health status using the extracted rules. The number and spacing of reference values ​​were optimized using inference error, and the optimal number of reference values ​​for each attribute was determined by controlling variables. It was found that a larger number of reference values ​​does not necessarily lead to lower inference error, and excessive redundant information increases computational complexity.

[0044] Therefore, rough set theory is used to reduce the attributes of the four premise attributes, remove redundant attributes (the highly correlated attributes of FD and FS, so FD is removed), and retain the core attributes.

[0045] Through equivalence class partitioning and confidence calculation, the final confidence rule base extracted by the delivery subsystem contains 225 valid rules, of which the first... A rule can be described as: ; By applying rough set theory to reduce the number of attribute values ​​in the warp feeding subsystem of the rapier loom and extracting rules for state reasoning, an optimized post-confidence rule base for the system was constructed. The size of the rule base was reduced from 300 to 225, and the computational efficiency was improved by 40%. This study preliminarily verified the effectiveness of the rule base construction method and the reliability of state reasoning.

[0046] Furthermore, the attribute reference values ​​in the initial confidence rule base of the weaving and winding subsystems can be reasonably and effectively determined, and a confidence rule base for assessing the health status of each subsystem of the rapier loom can be established for subsequent status assessment.

[0047] Evidence-based reasoning and the use of stratified sampling and secondary fusion methods to process missing data and optimize the final reasoning results; For the health status assessment of the warp feeding subsystem of a rapier loom, if the status assessment index data is missing, it is difficult to simply infer the health status using the BRB system.

[0048] We need to supplement the missing information by statistically analyzing the distribution of historical data and performing stratified sampling.

[0049] Obtained through statistical distribution and stratified sampling. Multiple sample values These sampled values ​​were combined with other precondition attributes to form multiple input combinations. Then, for each input combination, reasoning is performed one by one according to the reasoning method of the confidence rule base. This process includes calculating the rule activation level of each input combination, matching the rule antecedents, and using the ER algorithm to effectively fuse the evidence, thereby obtaining an evaluation result with confidence. .

[0050] After reasoning for all input combinations, the ER algorithm is used again to fuse the reasoning results of all inputs a second time, producing the final evaluation result. The inference error is obtained by comparing it with the evaluation results in the simulation data. : ; This process not only improves reasoning ability under incomplete input information but also enhances the accuracy and reliability of assessment results, providing an effective data processing and reasoning strategy for the health status assessment of rapier looms, such as... Figure 6 The diagram shows the state reasoning process when the prerequisite attribute information is missing.

[0051] The WOA algorithm with interpretability constraints optimizes the parameters of the confidence rule base. To further improve the assessment accuracy, this invention selects the WOA algorithm to optimize the rule weights, premise attribute weights, and output confidence scores of the confidence rule base, and uses the root mean square error (RMSE) to reflect the model's accuracy, thereby reflecting the accuracy and reliability of the optimized model in health status assessment. However, since the interpretability of the BRB is always compromised during the optimization process, a WOA algorithm with interpretability constraints is proposed to optimize the BRB.

[0052] like Figure 7 The flowchart of the WOA algorithm with interpretability constraints is shown below. The core optimization process is as follows.

[0053] (1) Initialization: Define the population size of whales Search space size Number of iterations .

[0054] (2) Point Scattering Operation: The traditional WOA algorithm uses a random point scattering method, which has low efficiency in utilizing expert knowledge. The strategy adopted in this invention is to construct the solution space centered on expert knowledge and scatter points within a specific region around the expert knowledge, thereby enhancing the interpretability of the model. The whale's position vector is denoted as... , No. A whale can be represented as: ; In the formula: The confidence level of expert knowledge; To return a A random matrix.

[0055] (3) Calculate the fitness value of each whale, which is the mean square error value.

[0056] (4) Constraint Operation: The following constraints are added to ensure that the BRB-WOA-e model is interpretable during the optimization process. The specific constraints are as follows: Constraint 1: The optimized rules often produce errors that conflict with the actual subsystems of the rapier loom.

[0057] Therefore, the optimized rule confidence distribution needs to be consistent with the actual health status of each subsystem to ensure the rationality of the confidence rules.

[0058] ; In the formula: under the k-th rule, the interpretability constraint of the confidence distribution is denoted as Ek, and K is the number of confidence rules.

[0059] Constraint 2: To maintain model interpretability, the changes in rule confidence during optimization must conform to expert knowledge, and a reasonable range of constraints must be set.

[0060] ; In the formula: The k-th confidence level of the n-th rule; This is to combine the minimum confidence level given by each expert; This is the maximum confidence level given by each expert.

[0061] (5) Encircling prey: When hunting, whales encircle their prey to update their position. This behavior can be described as follows: , ; in, This represents the current iteration number. The distance between the humpback whale and its prey. This indicates the location of the whale with the best current adaptation. This indicates the whale's position in the current iteration. and This is the coefficient vector updated in each iteration.

[0062] (6) Catching prey: When whales catch their prey, they swim towards it in a spiral motion, exhaling bubbles as they swim. , ; in It is the distance between the whale and its prey; yes Internal random number, It is a constant (representing the spiral pattern).

[0063] (7) Searching for prey: The process of searching for prey is to find the optimal solution, that is: , ; in, Indicates the whale's random location; when At that time, a whale individual is randomly selected.

[0064] This invention uses 200 sets of test data to verify the effectiveness of state reasoning before and after the optimization of the confidence rule base parameters.

[0065] like Figure 8 The inference error is 0.0036, indicating a large error between the system output value and the true value.

[0066] like Figure 9 As shown, the state inference results output by the true value and BRB-WOA-e were compared, and the mean squared error was 0.00020, a significant reduction. Furthermore, this invention compared the inference errors of the two methods with different numbers of test groups, as shown in Table 1. It can be seen that the BRB inference error is larger after optimizing the reference value, while the optimized confidence rule base maintains a low level of state inference error across different numbers of test groups; with 600 test data groups, its inference error is only 0.00028.

[0067] This indicates that the method of using rough sets for rule extraction has a certain degree of stability when constructing a confidence rule base, and that the WOA-e algorithm can significantly reduce inference errors by optimizing the parameters of the confidence rule base.

[0068] Table 1: Comparison of inference errors of different methods ; The method of establishing, reasoning, and optimizing the confidence rule base of this invention can effectively assess the health status of the subsystem while keeping the reasoning error at a low level.

[0069] Therefore, confidence rule bases were established for the winding subsystem of the weaving subsystem, and experimental verification was conducted using 200 sets of test data, such as... Figure 10 and Figure 11 BRB-WOA-e state inference was applied to both the weaving and take-up subsystems. The goodness of fit between the actual values ​​and the system output values ​​in the figure shows that the state inference results obtained using the BRB-WOA-e method have a small error compared to the actual health state, demonstrating sensitivity and accuracy to changes in health state. This further proves the reliability and effectiveness of the invention in each subsystem, providing accurate health state information for the weaving and take-up subsystems of rapier looms.

[0070] To compare the inference performance of the BRB-WOA-e method during the optimization process, rule extraction and state inference were performed on the initial confidence rule base, the confidence rule base after optimizing the reference value, and the WOA-e optimized confidence rule base of the warp feeding, weaving, and winding subsystems, respectively. The inference error data of each subsystem at each stage are shown in Table 2.

[0071] Table 2: Inference Errors of Each Subsystem ; As shown in Table 2, optimizing the number of reference values ​​significantly reduced the error in the confidence rule base extracted using rough sets. The inference error of the convolution subsystem showed the most significant reduction, decreasing by 0.01918 compared to before optimization. Further optimization of the confidence rule base parameters using the WOA-e algorithm further reduced the inference errors of each subsystem, with an average inference error of 3.57 × 10⁻⁶. -4 .

[0072] The optimized rule base is updated using incremental rough set analysis, and recent data is retained using a sliding window, enabling the model to continuously learn new data, avoiding "model drift" and achieving long-term reliability.

[0073] Based on the existing confidence rule base, a dynamic update mechanism is introduced. As the equipment operating time and environment change, the rule weights, reference values, and other parameters in the confidence rule base are updated periodically to adapt to changes in equipment status, environment, and other factors.

[0074] Incremental rough set update: When new data is input, the incremental rough set theory is used to dynamically adjust attribute reduction and rule extraction, avoiding retraining the entire model.

[0075] The core steps are as follows: ① When a new data block Upon arrival, only the differences between the data and historical data (new / deleted objects) are calculated.

[0076] ② Dynamically adjust equivalence class partitioning and update approximate sets: ; ③ Recalculate the rule confidence, retain rules with a confidence level > 0.6, and delete redundant rules.

[0077] Combining sliding window with forgetting factor: Set the sliding window capacity to W=1000 data points, and only retain the most recent W data points for rule updates, balancing computational efficiency and model timeliness.

[0078] The technical details are as follows: Data eviction: When new data arrives and the window is full, evict the old data with the lowest weight (according to exponential decay forgetting, decay factor). ); Weight update: After each iteration, according to... The formula for attenuating the weight of historical data and prioritizing the influence of recent data is as follows: ; After introducing a dynamic update mechanism, computational efficiency was improved. The incremental update time was reduced from 120 seconds for full training to 5 seconds (for 1000 data points), and the number of rules was dynamically maintained at 200-250 to avoid unlimited growth. As the amount of data increased, the error stabilized within ±0.02.

[0079] Any aspects not covered in this invention are applicable to existing technologies.

Claims

1. A method for assessing the health status of various subsystems of a rapier loom, characterized in that, The method optimizes the confidence rule base parameters and the dynamic update mechanism of the confidence rule base based on the whale algorithm with interpretability constraints, and includes the following steps: Step 1: Investigate the fault mechanisms of each subsystem of the rapier loom and select characteristic parameters. Then, extract features based on wavelet transform and enhance abnormal data based on GAN to construct a health status model. Step 2: Use rough set theory to reduce the attributes and extract rules for the evaluation indicators, and construct an initial confidence rule base for each subsystem of the rapier loom; Step 3: Use the evidence reasoning algorithm to fuse the activated rules and output the confidence level of each subsystem at each health status level; Step 4: Perform stratified sampling based on data distribution to process missing data, and combine multiple inference results to use the ER algorithm to fuse all inference conclusions a second time to obtain the final inference result; Step 5: Using inference error as the objective function, optimize the parameters of the confidence rule base using the WOA algorithm with interpretability constraints; Step 6: Update the optimized rule base using incremental rough set and retain recent data using a sliding window to enable the model to continuously learn new data.

2. The method for assessing the health status of each subsystem of a rapier loom according to claim 1, characterized in that, In step one The rapier loom's subsystems include a warp feeding subsystem, a weaving subsystem, and a take-up subsystem; characteristic parameters include temperature signals, vibration signals, and tension signals; and the specific steps of step one are as follows: Preprocess the feature parameters; Wavelet transform is used to decompose the signal by feature parameters, extract multi-scale time-frequency features, and perform feature selection to enhance fault-sensitive information in the signal. Use GANs to generate anomalous data and balance the distribution of the training set.

3. The method for assessing the health status of each subsystem of a rapier loom according to claim 2, characterized in that, The specific steps of wavelet transform are as follows: Signal decomposition: Multi-scale wavelet packet decomposition was performed on the vibration signal of the rapier loom. The Daubechies 4 wavelet basis function was selected, and the decomposition layer was 5, resulting in 16 sub-bands. The decomposition formula is as follows: ； in, For scale parameters, For translation parameters, These are wavelet basis functions; Feature extraction: Energy entropy calculation: The energy entropy is calculated for each layer of wavelet coefficients, representing the signal complexity. ， ； Multi-scale feature fusion: The energy entropy of each layer is fused with the original time-domain features to form a comprehensive feature vector, which serves as the input to the confidence rule base. Feature selection: ReliefF algorithm weight update: ； Where NH represents the nearest neighbor of the same class, NM represents the nearest neighbor of different classes, and k represents the number of samples.

4. The method for assessing the health status of each subsystem of a rapier loom according to claim 2, characterized in that, The specific steps for generating outlier data using GANs and balancing the training set distribution are as follows: The generator uses a fully connected layer + deconvolution structure. The input is random noise, and the output is a feature vector of synthesized vibration, temperature, and tension signals. Discriminator: Employs a convolutional neural network, with input being real / synthetic feature vectors and output being the discrimination probability; Loss function: Use Wasserstein GAN loss to improve training stability; ； Data balancing: Abnormal samples are generated using GAN to adjust the ratio to 3:

1.

5. The method for assessing the health status of each subsystem of a rapier loom according to claim 1, characterized in that, The specific steps for extracting confidence rules using the relative reduction of rough sets in step two are as follows: Divide the data into equivalence classes based on attribute reference values; Calculate the confidence of decision rules with the same conditional attributes under different decision attributes, and extract rules with higher confidence based on rough set theory; The extracted confidence rules are used for inference, and the number and spacing of the initially defined attribute reference values ​​are optimized based on the inference results.

6. The method for assessing the health status of each subsystem of a rapier loom according to claim 1, characterized in that, The specific steps for step three are as follows: Once the model receives the input information, it matches the input parameters with the antecedent of each confidence rule. The matched confidence rules are combined and weighted using an evidence reasoning algorithm to comprehensively consider the impact of each rule on the output result, thereby obtaining the confidence score of the output.

7. The method for assessing the health status of each subsystem of a rapier loom according to claim 1, characterized in that, The specific steps for handling missing data in step four are as follows: Statistical analysis of historical data is performed to determine the data distribution characteristics of missing attributes; Stratified sampling involves dividing the data distribution into several levels or intervals, and then sampling according to the proportion of data distribution within each interval. By combining the reconstructed attribute values ​​with other known attribute information, a comprehensive evaluation is conducted using a confidence rule base and evidence reasoning algorithm to ultimately determine the health status of the subsystem.

8. The method for assessing the health status of each subsystem of a rapier loom according to claim 1, characterized in that, The specific steps for step five are as follows: Initialization and expert knowledge integration: Initialize the whale population, define the search space and the maximum number of iterations; Expert knowledge-guided point selection, centered on expert experience, generates initial whale positions within their neighborhood, ensuring that the optimized starting point aligns with the actual system mechanism. A whale can be represented as: ； In the formula: The confidence level of expert knowledge; To return a A random matrix; Fitness calculation and constraint design: Calculate the fitness value for each individual whale, using mean squared error as the objective function; Interpretability constraints: Whale Behavior Simulation and Parameter Optimization: Surround the prey: Update the whale's position based on the current optimal solution to narrow the search range; Spiral hunting: Simulates the spiral approach behavior of whales towards their prey, with fine-tuned parameters in specific areas; Random search: When the exploration coefficient exceeds the threshold, the individual location is randomly selected to enhance the global search capability; Dynamic updates and verification: The rationality of the rule base is continuously verified during the iteration process, and the parameters are reset if the constraints are violated. Output the final optimized confidence rule base, ensuring that it has both high accuracy and interpretability.

9. The method for assessing the health status of each subsystem of a rapier loom according to claim 8, characterized in that, Interpretability constraints include: Constraint 1: The optimized rule confidence distribution must be consistent with the actual health status of each system of the rapier loom; ； In the formula: Under the k-th rule, the interpretability constraint of the confidence distribution is denoted as Ek, which is determined by the analysis of each actual system and has no fixed form; K is the number of confidence rules. Constraint 2: The adjustment range of the rule confidence level must be within a reasonable range preset by the experts; ； In the formula: The k-th confidence level of the n-th rule; , These represent the minimum and maximum confidence levels given by each expert, respectively.

10. The method for assessing the health status of each subsystem of a rapier loom according to claim 1, characterized in that, In step six, the specific steps for incremental rough set updating are as follows: When new data arrives, only the confidence and weight of the affected rules are updated, using the following formula: ； The sliding window mechanism retains data from the most recent 7 days, while older data decays exponentially due to forgetting factors. The dynamic adjustment rule weights are as follows: .

Images