Multi-parameter comprehensive calibration method for wet track abrasion tester of emulsified asphalt slurry mixture

Abstract

The application discloses a kind of emulsified asphalt slurry mixture wet wheel abrasion tester multi-parameter comprehensive calibration method, it is related to instrument calibration technical field, comprising: the fault case data of wet wheel abrasion tester is obtained, the key parameter category corresponding to wet wheel abrasion tester calibration is analyzed and obtained, and multi-source key parameters are collected;Variational modal decomposition using grey wolf optimization algorithm is improved to preprocess the multi-source key parameters;Using Bi-LSTM network enhanced introduction of noise reduction and sparse function autoencoder constructs comprehensive calibration model, obtains the reconstruction data of multi-source key parameters, determines the fault type according to reconstruction error distribution and marks abnormal period;According to the fault type and abnormal period, parameter collaborative calibration is carried out, and the calibration process is recorded, and the calibration effect is verified.The application solves the problem that parameter coupling is strong in wet wheel abrasion tester calibration, dynamic characteristics are complex and the like, and significantly improves the accuracy of emulsified asphalt slurry mixture abrasion resistance performance test.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of instrument calibration, and more particularly to a multi-parameter comprehensive calibration method of an emulsified asphalt slurry mixture wet track abrasion tester. BACKGROUND

[0002] As an economical and environmentally friendly pavement maintenance material, emulsified asphalt slurry mixture is widely used in preventive maintenance and minor damage repair. Performance evaluation mainly includes abrasion resistance, water damage resistance and asphalt-aggregate compatibility, etc., and the wet track abrasion test (WTAT) is one of the key methods to evaluate the durability of slurry mixture. The wet track abrasion tester measures the mass loss of the test piece under specific conditions by simulating the effects of wheel load and water erosion to evaluate the abrasion resistance of the mixture.

[0003] Currently, the calibration of the wet track abrasion tester mainly relies on single parameter verification (such as speed, abrasion time, rubber wheel pressure, etc.), and lacks a systematic and multi-parameter collaborative calibration method. Since the test results of the wet track abrasion test are affected by many factors such as instrument operating state, environmental conditions and operation specifications, relying on single parameter calibration may lead to test data deviation, affecting the accurate determination of the optimum asphalt content (OAC) of the mixture, and thus reducing the road performance of the slurry seal. In addition, wet track abrasion testers produced by different manufacturers differ in structure design, loading method, etc. If the calibration method is not unified, the test results may not be comparable, affecting engineering application. Therefore, it is of great significance to establish a multi-parameter comprehensive calibration method for the wet track abrasion tester, covering speed accuracy, rubber wheel wear state, water bath temperature stability, abrasion time control and loading pressure uniformity, etc., to improve the accuracy and repeatability of test data. SUMMARY

[0004] To solve the above technical problems, the present application provides a multi-parameter comprehensive calibration method for an emulsified asphalt slurry mixture wet track abrasion tester, which improves the accuracy of emulsified asphalt slurry mixture abrasion resistance test and provides a reliable basis for slurry mixture mix design, quality control and engineering acceptance.

[0005] The present application provides a multi-parameter comprehensive calibration method for an emulsified asphalt slurry mixture wet track abrasion tester, which improves the accuracy of emulsified asphalt slurry mixture abrasion resistance test and provides a reliable basis for slurry mixture mix design, quality control and engineering acceptance.

[0006] Obtain fault case data of the wet track abrasion tester, preprocess the fault case data, analyze and obtain key parameter categories corresponding to the calibration of the wet track abrasion tester, and collect multi-source key parameters according to the key parameter categories.

[0007] Based on the grey wolf optimization algorithm, the variational mode decomposition is improved, and the improved variational mode decomposition is used for pretreatment of the multi-source key parameters;

[0008] A comprehensive calibration model is constructed by using a Bi-LSTM network to enhance the self-encoder with noise reduction and sparse functions, the preprocessed multi-source key parameters are input into the comprehensive calibration model, the reconstruction data of the multi-source key parameters are obtained, the fault type is determined according to the reconstruction error distribution, and the abnormal period is marked;

[0009] According to the fault type and the abnormal period, parameter collaborative calibration is performed, and the calibration process is recorded, and the calibration effect is verified.

[0010] In this scheme, the fault case data of the wet track abrasion tester is obtained, the fault case data is pretreated, and the key parameter categories corresponding to the calibration of the wet track abrasion tester are analyzed and obtained, specifically:

[0011] According to the historical test data, historical operation and maintenance data and fault simulation data of the emulsified asphalt slurry mixture wet track abrasion tester, the fault case data is extracted, the fault case data is cleaned, and the pretreated fault case data is labeled using the fault type;

[0012] The parameter traceability of the fault case data is performed, the parameters involved in the fault are extracted, the extracted parameters are divided into continuous parameters, discrete parameters and periodic parameters, the region isolation forest modeling is performed on different parameters, and the super-sphere division strategy is introduced to define the data space;

[0013] The path length is calculated according to the number of spheres passed from the root node to the leaf node, the parameter importance degree is calculated based on the path length by using the feature disturbance method, the parameters meeting the preset importance degree threshold are screened, and the screened parameters are sorted according to the parameter importance degree;

[0014] According to the sorting result, a preset number of parameters are selected as the key parameter categories corresponding to the calibration of the wet track abrasion tester, and the multi-source key parameters are collected based on the key parameter categories.

[0015] In this scheme, the variational mode decomposition is improved based on the grey wolf optimization algorithm, and the improved variational mode decomposition is used for pretreatment of the multi-source key parameters, specifically:

[0016] The parameters of the grey wolf optimization algorithm are initialized, the value range of the mode number and the penalty factor is set, the position vector of each grey wolf is randomly initialized, the quality factor is selected to construct the fitness function, and the variational mode decomposition is performed on each parameter combination;

[0017] According to the fitness ranking, the top three optimal solutions are selected as alpha wolf, beta wolf and delta wolf, and the current global optimal solution is recorded, and the positions of the alpha wolf, the beta wolf, the delta wolf and other gray wolves are updated by using the steps of surrounding, hunting and attacking prey in the iterative calculation, the gray wolves before and after the position update are paired with each other, and the new gray wolf population is generated by uniform crossover according to the preset probability;

[0018] The fitness of the gray wolves in the new gray wolf population is calculated, a mutation operator changing with the number of iterations is introduced, the alpha wolf with the highest fitness is subjected to Gaussian mutation, the fitness of the alpha wolf before and after the mutation is compared, and the better solution is selected into the next generation by using the elite reservation strategy;

[0019] When the number of iterations reaches the maximum number of iterations or the iteration termination condition is met, the position of the alpha wolf with the highest fitness in the last iteration is output, and the optimal modal number and the penalty factor corresponding to the multi-source key parameter are obtained;

[0020] The optimal modal number and the penalty factor corresponding to the multi-source key parameter are used to configure the parameter adaptive decomposition of the variational modal decomposition, the sample entropy of each IMF component is calculated, the components meeting the preset entropy value interval are retained, and the preprocessed multi-source key parameter is obtained.

[0021] In the scheme, a Bi-LSTM network is used to enhance the self-encoder with noise reduction and sparsity function to construct a comprehensive calibration model, specifically:

[0022] A double-channel hybrid neural network structure is used to construct a comprehensive calibration model, which includes a time sequence extraction layer and a feature optimization layer, and a Bi-LSTM noise reducer is constructed by using a Bi-LSTM network and a noise reducer in the feature extraction layer;

[0023] According to the importance of the corresponding parameters, adaptive noise is added to the input multi-source key parameters, a Bi-LSTM noise reducer is used to extract bidirectional features from the noisy multivariate time series, a spatiotemporal attention mechanism is introduced, spatiotemporal attention weights are obtained, and noise reduction is performed by using an adversarial training and Gaussian mask strategy;

[0024] In the feature optimization layer, a sparse self-encoder is used for feature space compression and key information retention, the features output by the Bi-LSTM noise reducer are sparsely represented by using a KL divergence penalty term, and the independence of each dimension feature is enhanced by orthogonal regularization;

[0025] An inverse Bi-LSTM network is used to construct the decoder of the comprehensive calibration model, a skip connection is introduced to retain high-frequency details, the importance weight of each time step is calculated to focus on reconstructing the fault sensitive period, and the weighted time sequence features are used to guide the reconstruction of the multi-source key parameters.

[0026] In this scheme, the pretreated multi-source key parameters are introduced into the comprehensive calibration model to obtain reconstructed data of the multi-source key parameters, and the fault type is determined according to the reconstruction error distribution and the abnormal period is marked. Specifically,

[0027] The pretreated multi-source key parameters are standardized and time-aligned, and a multi-dimensional time series matrix is constructed as the input of the comprehensive calibration model. A Bi-LSTM denoising encoder is used to extract bidirectional time sequence features and perform spatio-temporal attention weighting;

[0028] The features output by the Bi-LSTM denoising encoder are introduced into a sparse autoencoder for compression and reconstruction. The reconstruction data is used to obtain time domain error, frequency domain error and feature space error in the bottleneck layer, respectively, to generate multi-dimensional reconstruction error.

[0029] Based on the historical normal data of the emulsified asphalt slurry mixture wet wheel abrasion tester, a dynamic threshold is set, and an adaptive sliding window is set according to short-term and long-term anomaly detection. Whether the multi-dimensional reconstruction error meets the dynamic threshold is judged according to the adaptive sliding window, and the abnormal period is detected through the judgment result.

[0030] Based on the multi-dimensional reconstruction error, principal component analysis is performed to obtain principal component reconstruction error as the principal component direction for principal component projection, and a reconstruction error distribution is generated. According to the abnormal period detection result and the reconstruction error distribution, Mahalanobis distance is used for fault feature mode matching, and the fault category is identified according to the matching result.

[0031] In this scheme, the threshold is dynamically adjusted according to the aging of the emulsified asphalt slurry mixture wet wheel abrasion tester. Specifically,

[0032] The baseline normal data of the emulsified asphalt slurry mixture wet wheel abrasion tester corresponding to the multi-source key parameters is obtained, and an aging coefficient vector of the multi-source key parameters is defined. According to the baseline normal data, the aging coefficient vector, the device usage intensity and the usage time, an aging model is constructed;

[0033] In the aging model, particle swarm optimization is used to optimize the aging coefficient. The value range of the aging coefficient is initialized, and the fitness of the particles is calculated according to the deviation between the predicted normal data and the actual normal data of the current multi-source key parameters.

[0034] The fitness of the current particles is introduced into the chaos algorithm for chaos processing, and the differential idea is used to determine the optimal position moving direction for position updating. Through iterative position updating, the optimal solution of the aging coefficient is obtained until the maximum iteration number is reached.

[0035] The optimal solution of the aging coefficient is used to weight the threshold corresponding to the historical normal data, and the threshold is dynamically adjusted.

[0036] In the scheme, according to the fault type and the abnormal period, parameter collaborative calibration is carried out, and the calibration process is recorded, and the calibration effect is verified, specifically:

[0037] According to different fault types, a hierarchical response mechanism is preset, a calibration priority is set in the hierarchical response mechanism through the current fault type and the abnormal period, and a calibration time is determined based on the calibration priority;

[0038] The main key parameters and the error image corresponding to the associated key parameters of the current fault type are obtained, a Stacking strategy is introduced to construct a retrieval model, a KNN model is trained based on the distance feature vectors corresponding to the Euclidean distance, Manhattan distance and Frechet distance, a preset number of calibration instances are output, a score is combined with the calibration effect, and a compensation coefficient of the main key parameters and the associated key parameters is determined using the score result;

[0039] The calibration process is recorded using a preset electronic form, the difference value of the main key parameters before and after calibration is compared, and a control chart is established to track the difference value for stability monitoring and verification.

[0040] The second aspect of the application provides a multi-parameter comprehensive calibration system for an emulsified asphalt thin slurry mixture wet wheel abrasion tester, comprising a memory and a processor, the memory comprising a multi-parameter comprehensive calibration method program for an emulsified asphalt thin slurry mixture wet wheel abrasion tester, and when the multi-parameter comprehensive calibration method program for an emulsified asphalt thin slurry mixture wet wheel abrasion tester is executed by the processor, the steps of the multi-parameter comprehensive calibration method for an emulsified asphalt thin slurry mixture wet wheel abrasion tester are implemented.

[0041] Compared with the prior art, the application has the following beneficial effects:

[0042] The application solves the problems of strong parameter coupling and complex dynamic characteristics in wet wheel abrasion tester calibration, and significantly improves the accuracy of emulsified asphalt thin slurry mixture abrasion resistance test. In the technical scheme of the application, the improved variational mode decomposition is used to adaptively decompose non-stationary signals such as speed, temperature and pressure, effectively separating noise and fault characteristics, reducing the calibration error of key parameters, and the LSTM-VAE model accurately captures the time dependence of dynamic parameters, improving the calibration accuracy and reliability. Through latent space mapping, compound chain fault modes are automatically identified, multi-parameter collaborative optimization is realized, and the false alarm risk is greatly reduced. Adaptive calibration and closed-loop calibration are used to improve the engineering applicability and automation level. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments or examples of the present application, the drawings needed to be used in the embodiments or examples will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0044] Figure 1 A flow chart of a multi-parameter comprehensive calibration method of an emulsified asphalt slurry mixture wet wheel abrasion tester is shown.

[0045] Figure 2 A flow chart of constructing a comprehensive calibration model in an embodiment is shown.

[0046] Figure 3 A flow chart of dynamically adjusting a threshold according to equipment aging in an embodiment is shown.

[0047] Figure 4 A block diagram of a multi-parameter comprehensive calibration system of an emulsified asphalt slurry mixture wet wheel abrasion tester is shown. DETAILED DESCRIPTION

[0048] In order to more clearly illustrate the technical solutions in the embodiments or examples of the present application, the drawings needed to be used in the embodiments or examples will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0049] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below.

[0050] Figure 1 A flow chart of a multi-parameter comprehensive calibration method of an emulsified asphalt slurry mixture wet wheel abrasion tester is shown.

[0051] As Figure 1 shown, the present embodiment provides a multi-parameter comprehensive calibration method of an emulsified asphalt slurry mixture wet wheel abrasion tester, comprising:

[0052] S102, obtaining fault case data of a wet wheel abrasion tester, pre-processing the fault case data, analyzing and obtaining key parameter categories corresponding to calibration of the wet wheel abrasion tester, and collecting multi-source key parameters according to the key parameter categories;

[0053] S104, improving variational mode decomposition based on a grey wolf optimization algorithm, and pre-processing the multi-source key parameters using the improved variational mode decomposition;

[0054] In S106, a comprehensive calibration model is constructed by using a Bi-LSTM network to enhance an autoencoder with noise reduction and sparsity functions, the preprocessed multi-source key parameters are introduced into the comprehensive calibration model, the reconstruction data of the multi-source key parameters are obtained, the fault type is determined according to the reconstruction error distribution, and the abnormal period is marked;

[0055] In S108, parameter collaborative calibration is performed according to the fault type and the abnormal period, the calibration process is recorded, and the calibration effect is verified.

[0056] It should be noted that the fault case data is extracted according to the historical test data, historical operation and maintenance data, and fault simulation data of the emulsified asphalt slurry mixture wet wheel abrasion tester, the fault case data is cleaned, and the preprocessed fault case data is labeled by using the fault type, for example, mechanical system failure (bearing wear, belt slip), control system failure (PID tuning, sensor drift), measurement system failure (pressure unevenness, timing error), etc.; the parameter tracing of the fault case data is performed, the parameters involved in the fault are extracted, the extracted parameters are divided into continuous parameters, discrete parameters, and periodic parameters, the regional isolation forest modeling is performed on different parameters, the data space is defined by introducing a hyper-sphere division strategy, the local sensitive area is formed by radius constraint, for the continuous parameters such as rotation speed, an adaptive radius hyper-sphere is used, wherein the radius constraint automatically adjusts the size of the hyper-sphere according to the parameter distribution density; for the discrete parameters such as wear state, a fixed radius hyper-sphere is used; for the periodic parameters such as temperature, an angle-constrained ring division is introduced.

[0057] The path length is calculated according to the number of spheres passed from the root node to the leaf node, the feature contribution degree is defined by the characteristic perturbation method based on the path length, the parameter importance degree is calculated, the parameters meeting the preset importance degree threshold are screened, and the screened parameters are sorted according to the parameter importance degree; preferably, the key parameters are screened by multi-criteria, the stability is verified by bootstrap sampling, the engineering correlation is calculated according to the Spearman correlation coefficient with the abrasion amount, and the multi-criteria screening is performed by using the importance degree, the stability and the engineering correlation. A preset number of parameters are selected as the key parameter categories corresponding to the wet wheel abrasion tester calibration according to the sorting result, and the multi-source key parameters are collected based on the key parameter categories. Preferably, the calibration of the wet wheel abrasion tester involves the rotational speed accuracy, the rubber wheel wear state, the water bath temperature stability, the abrasion time control and the loading pressure uniformity; the actual rotational speed is measured by using a high-precision photoelectric encoder, and the rotational speed accuracy is obtained by comparing with the set value; the rubber wheel wear state is obtained by calculating the wear depth and uniformity through a 3D profile scanner to obtain the rubber wheel surface morphology; the water bath temperature stability is obtained by using a high-precision temperature sensor to record the water bath temperature fluctuation and evaluating the temperature control performance; the abrasion time control is obtained by using a high-resolution timer to measure the actual abrasion time and comparing with the set time; and the loading pressure uniformity is obtained by arranging pressure sensors on the contact surface between the rubber wheel and the test piece to detect the pressure distribution uniformity.

[0058] It should be noted that the improved grey wolf optimization algorithm is used to intelligently optimize two key parameters (mode number and penalty factor) of the variational mode decomposition, and the optimal parameter combination is obtained through multi-stage iterative calculation. The parameters of the grey wolf optimization algorithm are initialized, the value range of the mode number and the penalty factor is set, the size of the grey wolf population is set, the maximum number of iterations is defined, the position vector of each grey wolf is randomly initialized, the mode number is ensured to be an integer, the penalty factor is within the set range, a quality factor is selected to construct a fitness function, and the variational mode decomposition is performed on each parameter combination; the quality factor of each IMF component is calculated, the quality factor=(signal energy / noise energy)×(main frequency bandwidth / total bandwidth), and the fitness function is constructed according to the weighted sum of all IMF classification quality factors. According to the fitness ranking, the top three optimal solutions are selected as alpha wolf, beta wolf and delta wolf, and the current global optimal solution is recorded. In the iterative calculation, the steps of surrounding, hunting and attacking prey are used to update the positions of alpha wolf, beta wolf, delta wolf and other grey wolves, the grey wolves before and after position updating are paired, uniform crossover is performed according to the preset probability, the mode number and the parameter value of the penalty factor are exchanged with a probability of 0.5, the newly generated mode number is ensured to be an integer, and a new grey wolf population is generated; the fitness of the grey wolf in the new grey wolf population is calculated, and a mutation operator changing with the number of iterations is introduced to perform Gaussian mutation on the alpha wolf with the highest fitness, the mutation probability decreases with the increase of the number of iterations, the fitness of the alpha wolf before and after mutation is compared, and the better solution is selected into the next generation using the elite reservation strategy; when the number of iterations reaches the maximum number of iterations or the iteration termination condition is met, the position of the alpha wolf with the highest fitness in the last iteration is output, and the optimal mode number and the penalty factor corresponding to the multi-source key parameter are obtained; the optimal mode number and the penalty factor corresponding to the multi-source key parameter are used to configure the variational mode decomposition for parameter adaptive decomposition, the sample entropy of each IMF component is calculated, the components meeting the preset entropy value interval are retained, and the preprocessed multi-source key parameter is obtained. Through the combination of the intelligent optimization algorithm and the adaptive signal decomposition, an accurate feature extraction basis is provided for the multi-parameter collaborative calibration of the wet track abrasion tester.

[0059] Figure 2 A flowchart for constructing a comprehensive calibration model in an embodiment is shown.

[0060] According to the embodiment of the present application, a Bi-LSTM network is used to enhance the self-encoder with noise reduction and sparsity function to construct a comprehensive calibration model, specifically:

[0061] S202, a double-channel hybrid neural network structure is used to construct a comprehensive calibration model, and the comprehensive calibration model comprises a time sequence extraction layer and a feature optimization layer. A Bi-LSTM network and a noise reduction encoder are used to construct a Bi-LSTM noise reduction encoder in the feature extraction layer;

[0062] S204, adding adaptive noise to the input multi-source key parameters according to the importance of the corresponding parameters, using a Bi-LSTM denoising encoder to perform bidirectional feature extraction on the noisy multivariate time series, introducing a spatiotemporal attention mechanism, obtaining spatiotemporal attention weights for weighted fusion, and using an adversarial training and Gaussian mask strategy for denoising;

[0063] S206, using a sparse autoencoder in the feature optimization layer to compress the feature space and retain key information, using KL divergence penalty terms to perform sparse representation of the features output by the Bi-LSTM denoising encoder, and using orthogonal regularization to enhance the independence of each dimension feature;

[0064] S208, using a reverse Bi-LSTM network to build a decoder for the comprehensive calibration model, introducing a skip connection to retain high-frequency details, calculating the importance weight of each time step to focus on reconstructing the fault-sensitive period, using the weighted time series features to guide the reconstruction of the multi-source key parameters, and using a multi-task head structure to locate the fault type.

[0065] It should be noted that the comprehensive calibration model is constructed by using the Bi-LSTM network to enhance the self-encoder with the functions of noise reduction and sparsity, wherein the bidirectional long short-term memory network captures the dependence before and after the time sequence, the noise reduction coding enhances the robustness to sensor noise, the sparsity constraint avoids overfitting, and the feature interpretability is improved. In the training process of the comprehensive calibration model, two-stage training is performed, the Bi-LSTM noise reduction encoder is pre-trained first, and then the entire network is fine-tuned end-to-end. The preprocessed multi-source key parameters are standardized and time-aligned, a multi-dimensional time sequence matrix is constructed as the input of the comprehensive calibration model, the Bi-LSTM noise reduction encoder is used to extract bidirectional time sequence features and perform spatio-temporal attention weighting, and the abnormal sensitive period is focused on. The features output by the Bi-LSTM noise reduction encoder are introduced into the sparse self-encoder for compression and reconstruction, the features are extracted through the bottleneck layer, the original dimension is restored by using the decoder, the time domain error is obtained by using the reconstructed data, the frequency domain error is obtained after FFT transformation, the feature space error is obtained in the bottleneck layer, and multi-dimensional reconstruction errors are generated. Based on the historical normal data of the emulsified asphalt slurry mixture wet wheel abrasion tester, a dynamic threshold is set, a self-adaptive sliding window is set according to short-term and long-term anomaly detection, whether the multi-dimensional reconstruction error meets the dynamic threshold is judged according to the self-adaptive sliding window, and anomaly period detection is performed according to the judgment result. Short-term anomaly (<5 seconds), such as detection of instantaneous fluctuation of rotating speed and pressure drop; long-term anomaly (> 30 seconds), such as wear trend deviation and temperature drift. Principal component analysis is performed based on the multi-dimensional reconstruction error, the principal component reconstruction error is obtained as the principal component direction for principal component projection, the reconstruction error distribution is generated, the fault feature mode matching is performed by using Mahalanobis distance according to the anomaly period detection result and the reconstruction error distribution, and the fault category is identified according to the matching result. For example, the feature mode corresponding to the belt slip is the high-frequency burr of rotating speed, and the typical period is the acceleration stage.

[0066] Figure 3 A flowchart for dynamically adjusting the threshold according to the aging of the device in the embodiment is shown.

[0067] According to the embodiment of the present application, the threshold is dynamically adjusted according to the aging of the emulsified asphalt slurry mixture wet wheel abrasion tester, specifically:

[0068] S302, the baseline normal data of the emulsified asphalt slurry mixture wet wheel abrasion tester corresponding to the multi-source key parameters is obtained, the aging coefficient vector of the multi-source key parameters is defined, and the aging model is constructed according to the baseline normal data, the aging coefficient vector, the use intensity and the use time of the device;

[0069] S304, the particle swarm algorithm is used for aging coefficient optimization in the aging model, the aging coefficient value range is initialized, and the fitness of the particles is calculated according to the deviation between the predicted normal data and the actual normal data of the current multi-source key parameters;

[0070] S306, the chaos algorithm is introduced to perform chaos processing on the fitness of the current particle, and the optimal position moving direction is determined by means of the differential thought to update the position, and the position is updated by iteration until the maximum iteration number is reached to obtain the optimal solution of the aging coefficient;

[0071] S308, the optimal solution of the aging coefficient is used to weight the threshold corresponding to the historical normal data to dynamically adjust the threshold.

[0072] It should be noted that the multi-source key parameters are collected in the new equipment acceptance period (first 3 months), the abnormal working condition data is removed, the statistical characteristics are extracted, and the baseline normal data is obtained. The aging influence equation in the aging model is represented as , represents the equipment use intensity coefficient, represents the aging coefficient, represents the use time, represents the baseline normal data, represents the normal data after aging influence, the fitness of the particle is calculated according to the deviation between the predicted normal data of the current multi-source key parameters and the actual normal data , and the fitness function is: In the particle swarm algorithm, the Logistic mapping is used to generate the initial population, and the differential strategy is introduced to improve the convergence speed and global search success rate. The threshold is intelligently adjusted by quantifying the aging state of the equipment, and maintenance decision support is provided according to the quantified aging state of the equipment. When the aging coefficient of a certain key parameter in the multi-source key parameters is greater than the preset threshold, an operation and maintenance warning is generated. In addition, the aging model is used to predict the next calibration time, and a warning information is generated before the next calibration time.

[0073] It should be noted that the hierarchical response mechanism is preset according to different fault types, and the preferred hierarchical response mechanism includes immediate calibration (P0): faults directly affecting the test accuracy such as temperature out of control, abnormal speed, planned calibration (P1): progressive faults such as wear, uneven pressure, observation monitoring (P2): abnormal auxiliary parameters such as environmental humidity. The calibration priority is set by querying the hierarchical response mechanism based on the current fault type and abnormal period, and the calibration time is determined based on the calibration priority; the error image corresponding to the main key parameter and the associated key parameter of the current fault type is obtained, for example, the main key parameter of bearing wear is speed fluctuation and vibration, and the associated key parameter is wear uniformity, and the preferred error image includes time domain features, frequency domain features and morphological features obtained by dynamic time warping algorithm; the Stacking strategy is introduced to construct a retrieval model, a KNN model is trained based on the distance feature vectors corresponding to the Euclidean distance, Manhattan distance and Frechet distance, a preset number of calibration instances are output, and the calibration effect is scored to determine the compensation coefficient of the main key parameter and the associated key parameter; the calibration process is recorded using a preset electronic form, and the preferred preset electronic form includes calibration time, fault code, deviation before compensation, compensation coefficient, etc. The difference value of the main key parameter before and after calibration is compared, and an X-bar-R control chart is established to track the difference value for stability monitoring and verification, wherein the stability criteria include 7 consecutive points on the same side of the center line, any point exceeding the control limit, and obvious non-random pattern. The closed-loop calibration is integrated into the wet wheel wear tester hardware to support unattended automatic calibration every 24 hours, reducing the need for manual intervention. In addition, a self-learning calibration knowledge base is constructed, and typical fault calibration schemes are stored in a structured manner. A closed-loop feedback is formed during each calibration, and when the calibration verification is passed, it is stored in the calibration knowledge base. The calibration knowledge base is automatically clustered and updated every month, and the multi-dimensional reconstructed error of the multi-source key parameters of the current emulsified asphalt thin slurry mixture wet wheel wear tester, the error image corresponding to the main key parameter and the associated key parameter are obtained. Similarity calculation is performed in the calibration knowledge base using the multi-dimensional reconstructed error, the error image corresponding to the main key parameter and the associated key parameter, and a fault calibration scheme that meets the similarity threshold is obtained. The current calibration scheme is generated through the screened fault calibration scheme; in addition, the remaining life is predicted through the control chart trend, and the predictive maintenance warning is generated according to the remaining life.

[0074] Figure 4 A block diagram of a multi-parameter comprehensive calibration system of an emulsified asphalt thin slurry mixture wet wheel wear tester is shown.

[0075] The second embodiment of the present application provides a multi-parameter comprehensive calibration system 4 of the emulsified asphalt slurry mixture wet wheel abrasion tester, comprising a memory 41 and a processor 42, the memory comprises a multi-parameter comprehensive calibration method program of the emulsified asphalt slurry mixture wet wheel abrasion tester, and when the multi-parameter comprehensive calibration method program of the emulsified asphalt slurry mixture wet wheel abrasion tester is executed by the processor, the steps of the multi-parameter comprehensive calibration method of the emulsified asphalt slurry mixture wet wheel abrasion tester are implemented.

[0076] The third aspect of the present application provides a computer readable storage medium, the computer readable storage medium comprises a multi-parameter comprehensive calibration method program of the emulsified asphalt slurry mixture wet wheel abrasion tester, and when the multi-parameter comprehensive calibration method program of the emulsified asphalt slurry mixture wet wheel abrasion tester is executed by the processor, the steps of the multi-parameter comprehensive calibration method of the emulsified asphalt slurry mixture wet wheel abrasion tester are implemented.

[0077] In several embodiments provided in the present application, it should be understood that the disclosed method and system can be implemented in other ways. The system embodiments described above are only illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division mode, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed components can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms. In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be realized in the form of hardware or hardware plus software function unit.

[0078] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, the aforementioned program can be stored in a computer readable storage medium, and the program executes the steps including the above-mentioned method embodiments when executed; and the aforementioned storage medium includes mobile storage device, read-only memory (ROM), random access memory (RAM), magnetic disc or optical disc and various program code storage media.

[0079] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A multi-parameter comprehensive calibration method for a wet wheel abrasion tester of emulsified asphalt slurry mixture, characterized in that, Includes the following steps: Obtain fault case data of wet wheel wear tester, preprocess the fault case data, analyze and obtain the key parameter categories corresponding to wet wheel wear tester calibration, and collect multi-source key parameters according to the key parameter categories; The variational mode decomposition is improved based on the gray wolf optimization algorithm, and the improved variational mode decomposition is used to preprocess the multi-source key parameters. A comprehensive calibration model is constructed by using a Bi-LSTM network to enhance the self-encoder with noise reduction and sparsity functions. The preprocessed multi-source key parameters are imported into the comprehensive calibration model to obtain the reconstructed data of the multi-source key parameters. The fault type is determined and the abnormal time period is marked according to the reconstruction error distribution. Perform parameter co-calibration based on the fault type and abnormal time period, record the calibration process, and verify the calibration effect; A comprehensive calibration model is constructed using a Bi-LSTM network to enhance the noise reduction and sparsity of the autoencoder. Preprocessed multi-source key parameters are imported into this model to obtain reconstructed data. Based on the reconstruction error distribution, the fault type is determined and abnormal time periods are marked. Specifically: A comprehensive calibration model is constructed using a dual-channel hybrid neural network structure. The comprehensive calibration model includes a temporal extraction layer and a feature optimization layer. In the temporal extraction layer, a Bi-LSTM network and a noise reduction encoder are used to construct a Bi-LSTM noise reduction encoder. The preprocessed multi-source key parameters are standardized and time-aligned. Adaptive noise is added to the input multi-source key parameters according to their importance. A Bi-LSTM noise reduction encoder is used to extract bidirectional features from the noisy multivariate time series. A spatiotemporal attention mechanism is introduced to obtain spatiotemporal attention weighted fusion. Adversarial training and Gaussian masking strategies are used for noise reduction. In the feature optimization layer, a sparse autoencoder is used to compress the feature space and retain key information. The features output by the Bi-LSTM noise reduction encoder are sparsely represented using the KL divergence penalty term. Orthogonal regularization is used to enhance the independence of each dimension of features. The decoder is used to recover the original dimension. The reconstructed data is used to obtain the time domain error, frequency domain error and feature space error in the bottleneck layer, respectively, to generate multi-dimensional reconstruction error. A decoder for the comprehensive calibration model is constructed using an inverse Bi-LSTM network. Skip connections are introduced to preserve high-frequency details. The importance weights of each time step are calculated to reconstruct the fault-sensitive period. Weighted time-series features are used to guide the reconstruction of multi-source key parameters. The reconstructed multi-source key parameters are imported into the output layer. A multi-task head structure is used to locate the fault type. Principal component analysis is performed based on the multi-dimensional reconstruction error. The principal component reconstruction error is used as the principal component direction for principal component projection to generate the reconstruction error distribution. Based on the abnormal period detection results and the reconstruction error distribution, Mahalanobis distance is used to perform fault feature pattern matching. The fault category is identified based on the matching results. A dynamic threshold is set based on historical normal data from a wet wheel abrasion tester for emulsified asphalt slurry mixtures. An adaptive sliding window is set based on short-term and long-term anomaly detection. The multi-dimensional reconstruction error is judged based on the adaptive sliding window to determine whether it meets the dynamic threshold. Anomaly detection is performed based on the judgment result.

2. The multi-parameter comprehensive calibration method for a wet wheel abrasion tester of emulsified asphalt slurry mixture according to claim 1, characterized in that, Fault case data of the wet wheel wear tester are obtained, and the fault case data is preprocessed to analyze and obtain the key parameter categories corresponding to the calibration of the wet wheel wear tester, specifically: Based on the historical test data, historical operation and maintenance data and fault simulation data of the wet wheel abrasion tester for emulsified asphalt slurry mixture, fault case data is extracted, the fault case data is cleaned, and the preprocessed fault case data is labeled with fault type. We perform parameter tracing on fault case data, extract the parameters involved in the fault, and divide the extracted parameters into continuous parameters, discrete parameters, and periodic parameters. We then perform regional isolated forest modeling for different parameters and introduce a hypersphere partitioning strategy to define the data space. The path length is calculated based on the number of spheres traversed from the root node to the leaf node. Based on the path length, the importance of parameters is calculated using the feature perturbation method. Parameters that meet the preset importance threshold are selected, and the selected parameters are sorted according to their importance. Based on the sorting results, a preset number of parameters are selected as the key parameter categories corresponding to the wet wheel wear tester calibration, and multi-source key parameters are collected based on the key parameter categories.

3. The multi-parameter comprehensive calibration method for a wet wheel abrasion tester of emulsified asphalt slurry mixture according to claim 1, characterized in that, An improvement to the variational mode decomposition based on the gray wolf optimization algorithm is applied, and the improved variational mode decomposition is used to preprocess the multi-source key parameters, specifically as follows: Initialize the parameters of the gray wolf optimization algorithm, set the range of values ​​for the number of modes and the penalty factor, randomly initialize the position vector of each gray wolf, select the quality factor to construct the fitness function, and perform variational mode decomposition for each parameter combination; The top three optimal solutions are selected based on fitness ranking as α wolf, β wolf, and δ wolf, and the current global optimal solution is recorded. In the iterative calculation, the positions of α wolf, β wolf, δ wolf and other gray wolves are updated using the steps of surrounding, hunting and attacking prey. Gray wolves before and after the position update are paired up and crossed uniformly according to a preset probability to generate a new gray wolf population. In the new gray wolf population, the fitness of gray wolves is calculated, and a mutation operator that changes with the number of iterations is introduced to perform Gaussian mutation on the α wolf with the highest fitness. The fitness of α wolves before and after mutation is compared, and an elite retention strategy is used to select a better solution to enter the next generation. When the number of iterations reaches the maximum number of iterations or the iteration termination condition is met, output the position of the α wolf with the highest fitness in the last iteration, and obtain the optimal number of modes and penalty factor corresponding to the multi-source key parameters. The optimal number of modes and penalty factor corresponding to the multi-source key parameters are used to configure variational mode decomposition for adaptive parameter decomposition. The sample entropy of each IMF component is calculated, and the components that meet the preset entropy value range are retained to obtain the preprocessed multi-source key parameters.

4. The multi-parameter comprehensive calibration method for a wet wheel abrasion tester of emulsified asphalt slurry mixture according to claim 1, characterized in that, The threshold for dynamic aging adjustment of the wet wheel abrasion tester for emulsified asphalt slurry mixture is as follows: Obtain the baseline normal data of the multi-source key parameters corresponding to the wet wheel abrasion tester for emulsified asphalt slurry mixture, define the aging coefficient vector of the multi-source key parameters, and construct an aging model based on the baseline normal data, aging coefficient vector, equipment usage intensity, and usage time. The aging model uses a particle swarm optimization algorithm to optimize the aging coefficient, initializes the range of aging coefficient values, and calculates the fitness of particles based on the deviation between the predicted normal data and the actual normal data of the current multi-source key parameters. A chaotic algorithm is introduced to process the fitness of the current particle in a chaotic manner, and the optimal position movement direction is determined by the difference idea to update the position. The position update is iterated until the maximum number of iterations is reached to obtain the optimal solution of the aging coefficient. The thresholds corresponding to historical normal data are weighted using the optimal solution of the aging coefficient, and the thresholds are dynamically adjusted.

5. The multi-parameter comprehensive calibration method for a wet wheel abrasion tester of emulsified asphalt slurry mixture according to claim 1, characterized in that, Based on the fault type and abnormal time period, parameter co-calibration is performed, the calibration process is recorded, and the calibration effect is verified, specifically as follows: A graded response mechanism is preset according to different fault types. The calibration priority is queried and set in the graded response mechanism by the current fault type and abnormal time period. The calibration time is determined based on the calibration priority. Obtain the error profiles corresponding to the main key parameters and related key parameters of the current fault type, introduce the Stacking strategy to build a retrieval model, train the KNN model based on the distance feature vectors corresponding to Euclidean distance, Manhattan distance and Fraser distance, output a preset number of calibration instances, score the calibration effect, and use the scoring results to determine the compensation coefficients of the main key parameters and related key parameters. The calibration process is recorded using a pre-set electronic form. The differences in key parameters before and after calibration are compared, and a control chart is established to track these differences for stability monitoring and verification.

6. A multi-parameter integrated calibration system for a wet wheel abrasion tester of emulsified asphalt slurry mixture, characterized in that, The system includes a memory and a processor. The memory includes a multi-parameter integrated calibration method program for an emulsified asphalt slurry mixture wet wheel abrasion tester. When the processor executes the multi-parameter integrated calibration method program for the emulsified asphalt slurry mixture wet wheel abrasion tester, it implements the steps of the multi-parameter integrated calibration method for the emulsified asphalt slurry mixture wet wheel abrasion tester as described in any one of claims 1 to 5.

Images