A dynamic balance test method for a grinding wheel assembly for a sand washing machine

Abstract

The present invention provides a dynamic balancing test method for a grinding wheel assembly for a sand washing machine, and relates to the technical field of equipment dynamic balancing test. The method includes matching the grinding wheel assembly model with the test rotating shaft model, performing correlation analysis on the vibration state index set, training the unbalanced state prediction network, fitting the grinding wheel assembly vibration state monitoring matrix, generating an unbalanced point mass list, an eccentric distance list, a detection angle list, and a detection time list, performing positioning according to the unbalanced point eccentric distance list, the detection angle list, and the detection time list, obtaining an unbalanced point positioning list, and adding the unbalanced point positioning list and the unbalanced point mass list to the dynamic balancing test result. The method solves the technical problems in the prior art that the dynamic balancing test is inefficient, the test result accuracy is not high, and the data simulation stability is difficult to guarantee. The method achieves the technical effect of improving the efficiency and test accuracy of the dynamic balancing test and ensuring the stability and reliability of the test results.

Description

Technical Field

[0001] The invention relates to the technical field of equipment dynamic balance testing, and in particular to a dynamic balance testing method for a grinding wheel assembly for a sand washing machine. Background Art

[0002] Traditional dynamic balancing tests rely on modal analysis, and the dynamic balancing test of grinding wheel assemblies for sand washing machines mainly relies on modal analysis methods. Modal analysis is a method that simulates the dynamic characteristics of grinding wheel assemblies under different unbalanced motion states through simulation models, and judges whether the current test motion state is consistent. However, modal analysis relies on complex simulation models and a large amount of computing resources, the dynamic balancing test process is slow, and the test efficiency is low. And due to the limited accuracy and complexity of the simulation model, it is difficult to fully simulate the dynamic behavior of the grinding wheel assembly in the real environment in the model, resulting in certain errors in the test results and low accuracy of the test results. In addition, the simulation model that modal analysis relies on is sensitive to initial conditions and boundary conditions. A slight change may cause the simulation results to be unstable, which further reduces the reliability of the test method.

[0003] In the prior art, the dynamic balance test of the grinding wheel assembly for the sand washing machine relies on modal analysis, which has the technical problems of low test efficiency, low accuracy of test results, and difficulty in ensuring the stability of data simulation. Summary of the invention

[0004] The present application provides a dynamic balancing test method for a grinding wheel assembly for a sand washing machine, which is used to solve the technical problems in the prior art that the dynamic balancing test of a grinding wheel assembly for a sand washing machine relies on modal analysis, has low test efficiency, low test result accuracy, and difficulty in ensuring data simulation stability.

[0005] In view of the above problems, the present application provides a dynamic balancing test method for a grinding wheel assembly for a sand washing machine, the method comprising: matching a test rotating shaft model through a grinding wheel assembly model, wherein the test rotating shaft model corresponds one-to-one to the grinding wheel assembly model; performing a correlation analysis on a vibration state index set based on the test rotating shaft model and the grinding wheel assembly model to obtain a correlated vibration state index; training an unbalanced state prediction network based on the correlated vibration state index and in combination with the test rotating shaft model and the grinding wheel assembly model; installing the grinding wheel assembly to the test rotating shaft, and rotating the grinding wheel assembly based on the rated speed. , collect the associated vibration state indicators through a laser vibration measuring instrument to obtain a vibration state monitoring matrix of a grinding wheel assembly; fit the vibration state monitoring matrix of a grinding wheel assembly through the unbalanced state prediction network to generate an unbalanced point mass list, an unbalanced point eccentricity list, an unbalanced point detection angle list and an unbalanced point detection time list; locate according to the unbalanced point eccentricity list, the unbalanced point detection angle list and the unbalanced point detection time list to obtain an unbalanced point positioning list; add the unbalanced point positioning list and the unbalanced point mass list to the precession balancing test result.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The method provided in the embodiment of the present application matches the test rotating shaft model with the grinding wheel assembly model, performs correlation analysis on the vibration state index set based on the test rotating shaft model and the grinding wheel assembly model, and obtains the associated vibration state index; trains the unbalanced state prediction network according to the associated vibration state index in combination with the test rotating shaft model and the grinding wheel assembly model; installs the grinding wheel assembly to the test rotating shaft, rotates the grinding wheel assembly based on the rated speed, collects the associated vibration state index through a laser vibration measuring instrument, and obtains the vibration state monitoring matrix of the grinding wheel assembly; fits the vibration state monitoring matrix of the grinding wheel assembly through the unbalanced state prediction network, generates an unbalanced point mass list, an unbalanced point eccentric distance list, an unbalanced point detection angle list, and an unbalanced point detection time list; locates according to the unbalanced point eccentric distance list, the unbalanced point detection angle list, and the unbalanced point detection time list, and obtains an unbalanced point location list; adds the unbalanced point location list and the unbalanced point mass list to the precession balance test result. The technical effect of improving the efficiency and accuracy of dynamic balancing tests and ensuring the stability and reliability of test results is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A schematic flow chart of a dynamic balancing test method for a grinding wheel assembly for a sand washing machine provided in this application.

[0010] Figure 2 A schematic diagram of a flow chart for obtaining associated vibration state indicators in a dynamic balancing test method for a grinding wheel assembly for a sand washing machine provided in the present application. DETAILED DESCRIPTION

[0011] The present application provides a dynamic balancing test method for a grinding wheel assembly for a sand washer, which is used to solve the technical problems that the dynamic balancing test of a grinding wheel assembly for a sand washer in the prior art relies on modal analysis, has low test efficiency, low test result accuracy, and difficulty in ensuring data simulation stability. The method achieves the technical effect of improving the efficiency and test accuracy of the dynamic balancing test and ensuring the stability and reliability of the test results.

[0012] Below, the technical solutions in the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the accompanying drawings, rather than all of them.

[0013] like Figure 1 As shown, the present application provides a dynamic balancing test method for a grinding wheel assembly for a sand washing machine, the method comprising:

[0014] The test rotating axis model is matched by the grinding wheel assembly model, wherein the test rotating axis model corresponds one to one with the grinding wheel assembly model.

[0015] Specifically, the grinding wheel assembly for sand washing machine refers to the key component used to clean and separate sand and gravel in the sand washing equipment. The grinding wheel assembly model is used to fully reflect the physical characteristics and performance parameters of the grinding wheel assembly, including specific specifications and types, such as diameter, thickness, weight, material, etc. Each model of grinding wheel has its specific size, weight and usage requirements. Then, through the model of the grinding wheel assembly, the matching test rotating shaft model can be determined. The test rotating shaft model refers to the specifications and type of the specific shaft used to install and rotate the grinding wheel assembly, including parameters such as shaft diameter, shaft length, material and load-bearing capacity. The test rotating shaft model corresponds to the grinding wheel assembly model one by one. Matching the test rotating shaft model through the grinding wheel assembly model can ensure that the grinding wheel assembly can be firmly and stably supported during rotation, and can accurately reflect its vibration characteristics during the test process.

[0016] A correlation analysis is performed on a vibration state index set based on the test rotation axis model and the grinding wheel assembly model to obtain a correlated vibration state index.

[0017] Specifically, after determining the grinding wheel assembly model and the matching test rotating axis model, a vibration state index set with correlation with the test rotating axis model and the grinding wheel assembly model is identified and extracted from the dynamic balance test database of the grinding wheel assembly for sand washing machine, and the dynamic balance test database of the grinding wheel assembly for sand washing machine contains a combination of multiple rotating axis models and corresponding grinding wheel assembly models, historical or simulated dynamic balance test data of the grinding wheel assembly for sand washing machine, such as vibration state index data, dynamic balance sample experimental data, etc. And using the gray correlation analysis method, the collected vibration state index set data is analyzed for correlation with the grinding wheel assembly model and the test rotating axis model, and the vibration state index data with significant correlation with the dynamic balance state of the grinding wheel assembly is determined to form associated vibration state indicators, such as vibration frequency and vibration amplitude. The associated vibration state indicators can accurately reflect the dynamic characteristics of the grinding wheel assembly in the dynamic balance state, and provide important data basis for the subsequent unbalanced state prediction network training.

[0018] According to the associated vibration state index, in combination with the test rotating shaft model and the grinding wheel assembly model, an imbalance state prediction network is trained.

[0019] Specifically, the highly correlated associated vibration state index is used in combination with the test rotating axis model and the grinding wheel assembly model to construct a training data set. Then, a neural network is used to define an imbalance state prediction network, and the imbalance state prediction network is trained according to the training data set, so that the trained imbalance state prediction network can accurately predict the imbalance state of the grinding wheel assembly, thereby improving the test effect and the accuracy and stability of the test results.

[0020] The grinding wheel assembly is mounted on a test rotating shaft, the grinding wheel assembly is rotated based on a rated speed, the associated vibration state index is collected by a laser vibration measuring instrument, and a vibration state monitoring matrix of the grinding wheel assembly is obtained.

[0021] Specifically, first, the determined grinding wheel assembly is installed on the matching test rotating shaft. During the installation process, ensure that the connection between the grinding wheel assembly and the test rotating shaft is stable and not loose, so as to ensure that the grinding wheel can remain stable during the rotation process. Then, according to the specifications of the grinding wheel assembly and the operating requirements of the test rotating shaft, the rated speed of the test rotating shaft is set, and the rated speed refers to the standard rotation speed of the grinding wheel assembly under normal working conditions. The grinding wheel assembly is rotated based on the set rated speed. During the rotation of the grinding wheel assembly, the associated vibration index data of the grinding wheel is collected in real time by a laser vibration measuring instrument. The laser vibration measuring instrument is a high-precision non-contact measuring device. It uses the principle of laser beam reflection to accurately detect the vibration frequency, vibration amplitude, vibration direction and phase of the grinding wheel assembly. Finally, the collected associated vibration state data are summarized to form a vibration state monitoring matrix of the grinding wheel assembly. The vibration state monitoring matrix is ​​a multidimensional array that contains vibration data at different time points and different measurement directions. By collecting detailed vibration state data of the grinding wheel assembly at rated speed, a complete vibration state monitoring matrix is ​​formed, ensuring the accuracy and comprehensiveness of the data, and providing an important basis for subsequent imbalance state analysis and prediction.

[0022] The vibration state monitoring matrix of the grinding wheel assembly is fitted through the imbalance state prediction network to generate an imbalance point mass list, an imbalance point eccentricity distance list, an imbalance point detection angle list and an imbalance point detection time list.

[0023] Specifically, the obtained grinding wheel assembly vibration state monitoring matrix is ​​input into the unbalanced state prediction network, and the unbalanced state prediction network performs in-depth analysis and pattern recognition on the input associated vibration data. During the fitting process, the prediction network will automatically learn and associate the mapping relationship between the characteristic mode in the vibration data and the unbalanced state of the grinding wheel assembly. Based on the mapping relationship, a series of key parameter lists are predicted and generated, including an unbalanced point mass list, an unbalanced point eccentric distance list, an unbalanced point detection angle list, and an unbalanced point detection time list, wherein the unbalanced point mass list includes the mass size of the unbalanced point detected at different times, and these mass values ​​reflect the degree of imbalance. The unbalanced point eccentric distance list includes the eccentric distance information of the unbalanced point, that is, the distance of the unbalanced mass center from the rotation center, which affects the vibration amplitude and direction. The unbalanced point detection angle list includes the rotation angle position of the unbalanced point detected, which helps to accurately locate the specific location where the imbalance occurs. The unbalanced point detection time list includes the exact time when the unbalanced point is detected, which helps to analyze the time characteristics of the vibration and understand the dynamic changes of the unbalanced state. These lists provide a detailed description of the imbalance state of the grinding wheel assembly, which can improve the accuracy and comprehensiveness of subsequent balance test results.

[0024] Positioning is performed according to the unbalance point eccentricity distance list, the unbalance point detection angle list and the unbalance point detection time list to obtain an unbalance point positioning list.

[0025] Specifically, the rotation angle position of the unbalanced point detected is obtained from the unbalanced point detection angle list, and the rotation angle position of each unbalanced point at a certain moment is determined in combination with the time information in the unbalanced point detection time list. For example, based on the grinding wheel assembly at a certain moment, the rotation angle of a certain unbalanced point reaches a specific degree, and the rotation angle position of the unbalanced point at that time point can be plotted, and these unbalanced points are distributed on the same radius. Then, the unbalanced point eccentricity distance list is used to determine the specific position of each unbalanced point around the rotation axis, and by locating all unbalanced points one by one, an unbalanced point positioning list is generated, and the unbalanced point positioning list contains the specific time and specific position of each unbalanced point around the rotation axis.

[0026] The unbalance point location list and the unbalance point mass list are added to the precession balancing test result.

[0027] Specifically, after completing the precise positioning of the unbalanced points to obtain the unbalanced point positioning list, the unbalanced point positioning list and the unbalanced point mass list are integrated into the dynamic balancing test results to form a detailed and comprehensive dynamic balancing test report. The dynamic balancing test results clearly record the specific position and quality of each unbalanced point, thereby realizing efficient and precise dynamic balancing testing of the grinding wheel assembly for the sand washing machine, and improving the test efficiency as well as the accuracy and stability of the test results.

[0028] In one embodiment, Figure 2 As shown, a correlation analysis is performed on a vibration state index set based on the test rotating axis model and the grinding wheel assembly model to obtain a correlated vibration state index, including: collecting a dynamic balance unqualified sample experimental data set according to the test rotating axis model and the grinding wheel assembly model, wherein the dynamic balance unqualified sample experimental data set includes several groups of vibration state experimental monitoring data and several mass imbalance experimental record data; normalizing the several mass imbalance experimental record data to construct a reference sequence; traversing the several groups of vibration state experimental monitoring data and performing normalization processing respectively to construct several feature sequences; performing grey correlation analysis based on the reference sequence and the several feature sequences to generate an index correlation set; based on the index correlation set, extracting an index greater than or equal to a correlation threshold from the vibration state index set, and setting it as the correlated vibration state index.

[0029] Specifically, first, according to the determined test rotating shaft model and the grinding wheel assembly model, a matching dynamic balance unqualified sample experimental data set is collected from the dynamic balance test database of the grinding wheel assembly for the sand washing machine. The dynamic balance unqualified sample experimental data set refers to a set of experimental data in which the grinding wheel assembly fails to meet the dynamic balance standard during the test. The dynamic balance unqualified sample experimental data set includes several groups of vibration state experimental monitoring data and several mass imbalance experimental recording data. The vibration state experimental monitoring data includes various vibration parameters of the grinding wheel assembly in an unbalanced state, such as vibration frequency, vibration amplitude, etc. The mass imbalance experimental recording data is the measurement experimental data of the dynamic imbalance caused by the uneven mass distribution of the grinding wheel assembly during the rotation process. The unbalanced mass is calculated by measuring the centrifugal force of the grinding wheel assembly when it is stationary or rotating at a low speed, and then the ratio of the unbalanced mass to the total mass is used as the mass imbalance.

[0030] Next, the plurality of mass imbalance experimental record data are normalized, and the data of different dimensions are converted to the same dimension to eliminate the magnitude difference between the data, and a reference sequence is constructed. The reference sequence is a standard sequence for grey correlation analysis, representing the standard characteristics of mass imbalance. Then, the plurality of vibration state experimental monitoring data are traversed, and each vibration state experimental monitoring data is normalized to construct a plurality of feature sequences. The feature sequence refers to the vibration state data that has been normalized, and each group of vibration state data corresponds to a feature sequence.

[0031] After constructing the reference sequence and the characteristic sequence, a grey correlation analysis is performed on the reference sequence and the several characteristic sequences. Grey correlation analysis is a method for measuring the degree of correlation between different data sequences. It determines whether the connection is close by determining the degree of similarity of the shapes of the reference data column and the several comparison data columns, and reflects the degree of correlation between the curves. The closer the curves are, the greater the correlation between the corresponding sequences, and vice versa. The correlation between the reference sequence and each characteristic sequence is calculated by grey correlation analysis, and the calculated correlation values ​​are summarized to form an indicator correlation set, which refers to the set of correlation values ​​between each vibration state indicator and the mass imbalance. A correlation threshold is set in advance. The correlation threshold is a pre-set standard value used to screen vibration state indicators with significant correlation. By traversing the indicator correlation set, all vibration state indicators with correlation values ​​greater than or equal to the threshold in the vibration state indicator set are extracted to obtain the vibration state indicators with significant correlation with the mass imbalance. The extracted vibration state indicators are set as the associated vibration state indicators, which can accurately reflect the dynamic balance state of the grinding wheel assembly, improve the accuracy and efficiency of data analysis, and provide reliable data support for the subsequent imbalance state prediction network training.

[0032] Furthermore, according to the associated vibration state index, in combination with the test rotating shaft model and the grinding wheel assembly model, an unbalanced state prediction network is trained, including: taking the test rotating shaft model and the grinding wheel assembly model as constraints, and taking the associated vibration state index, unbalance point mass, and unbalance point mass eccentric distance as retrieval targets to perform data monitoring, and obtain an associated vibration index retrieval data set, an unbalance point mass retrieval data set, and an unbalance point mass eccentric distance retrieval data set; traversing the associated vibration index retrieval data set, and constructing a vibration state training input matrix data set, wherein the index attribute and the matrix position in any vibration state training input matrix data correspond one to one; constructing a binary data identification data set according to the unbalance point mass retrieval data set and the unbalance point mass eccentric distance retrieval data set; configuring a dual-constraint loss function, wherein the dual-constraint loss function includes a data loss function and a physical loss function; according to the data loss function and the physical loss function, retrieving the vibration state training input matrix data set and the binary data identification data set to train the unbalanced state prediction network.

[0033] Specifically, because different models of grinding wheel assemblies may have significant differences in vibration characteristics and unbalanced performance, the test rotating axis model and the grinding wheel assembly model are used as constraints. Based on the determined constraints, the associated vibration state index, unbalance point mass and unbalance point mass eccentric distance are used as retrieval targets, and the corresponding vibration state data, unbalance point mass data and eccentric distance data are retrieved in the dynamic balancing test database of the grinding wheel assembly for the sand washing machine, wherein the unbalance point mass is the specific mass size that causes the unbalance phenomenon, and the unbalance point mass eccentric distance is the degree of deviation of the unbalance mass relative to the rotation center. According to the monitoring results, the corresponding data set is formed, that is, the associated vibration index retrieval data set, the unbalance point mass retrieval data set and the unbalance point mass eccentric distance retrieval data set are obtained, and the associated vibration index retrieval data set includes all the vibration state index data retrieved within the constraints, the unbalance point mass retrieval data set includes all the unbalance point mass data retrieved within the constraints, and the unbalance point mass eccentric distance retrieval data set includes all the unbalance point eccentric distance data retrieved within the constraints.

[0034] Next, traverse each group of vibration state data in the associated vibration index retrieval data set, and input the vibration state index values ​​therein into the two-dimensional matrix in sequence, ensuring that the index attributes and matrix positions in any vibration state training input matrix data correspond one to one. Based on this, construct a vibration state training input matrix data set, in which each row corresponds to a vibration state record, and each column corresponds to a vibration state index. Read the unbalance point mass retrieval data set and the unbalance point mass eccentric distance retrieval data set, wherein the unbalance point mass retrieval data set and the unbalance point mass eccentric distance retrieval data set are one-to-one corresponding, that is, each unbalance point mass record corresponds to an unbalance point mass eccentric distance record. For each pair of matching unbalance point mass and unbalance point mass eccentric distance records, combine them into a binary group (or binary vector) to construct a binary data identification data set, which is used to identify whether the grinding wheel component is unbalanced or the degree of its unbalance.

[0035] Then, select or design a data loss function suitable for regression or classification tasks to measure the difference between the imbalance state predicted by the model and the true value. Based on physical principles and prior knowledge, design a physical loss function to constrain the prediction results of the model. The physical loss function can capture situations that may violate physical laws in the prediction results and give corresponding penalties. For example, design a penalty term to limit the predicted value of the imbalance point mass to a reasonable range, or limit the predicted value of the eccentric distance of the imbalance point mass to not exceed the radius of the rotation axis, etc. Combine the data loss function and the physical loss function to form a dual-constraint loss function.

[0036] Finally, a neural network architecture is selected as a network model for predicting the imbalance state. The vibration state training input matrix data set is used as the input data of the model, the binary data identification data set is used as the supervision signal of the model, and the input data is used to train the imbalance state prediction network. In each iteration, the input data is input into the network model to obtain the predicted value, and then the data loss function value and the physical loss function value between the predicted value and the true value are calculated, and the parameters of the model are updated using the back propagation algorithm to minimize the loss function value. Until the imbalance state prediction network reaches a convergence state, the convergence state is such as the loss function value of the model, including the data loss function and the physical loss function, which tend to be stable after multiple iterations and no longer decrease significantly, or the performance indicators of the model such as accuracy and recall rate reach a stable level on the validation set and no longer have a significant improvement, and a trained imbalance state prediction network is obtained. By configuring the dual-constraint loss function, calling the vibration state training input matrix data set and the binary data identification data set, and training the imbalance state prediction network for state prediction, accurate and comprehensive prediction of the imbalance state can be achieved.

[0037] Furthermore, according to the data loss function and the physical loss function, the vibration state training input matrix data set and the binary data identification data set are retrieved to train the imbalance state prediction network, including: building an angle recorder, wherein the angle recorder is used to record the rotation angle of any position of the grinding wheel assembly relative to the position at time 0; building a timer, wherein the timer is used to record the rotation time of the grinding wheel assembly; connecting the angle recorder, the timer, and the PINN neural network in parallel, and configuring a fully connected layer to generate an imbalance state prediction network topology; wherein the fully connected layer is used to fit the outputs of the angle recorder, the timer, and the PINN neural network, fitting the detection time of the PINN neural network output value through the timer output value, and fitting the detection angle of the PINN neural network output value through the angle recorder; according to the data loss function and the physical loss function, the vibration state training input matrix data set and the binary data identification data set are retrieved to train the imbalance state prediction network topology and generate the imbalance state prediction network.

[0038] Specifically, the angle recorder is used to accurately record the rotation angle of the grinding wheel assembly at any time during operation relative to the position at time 0, that is, the initial position. The angle recorder is implemented by measuring the angle change through a built-in encoder or sensor, and can capture and feedback the rotation angle information of the grinding wheel in real time. For example, the angle recorder updates the record every time the grinding wheel rotates one degree to ensure accurate tracking of the grinding wheel position. The timer works in parallel with the angle recorder to record the rotation time of the grinding wheel assembly. The timer is implemented based on a high-precision clock chip, which can ensure the accuracy of time recording and provide basic data for analyzing the time correlation during the operation of the grinding wheel.

[0039] The PINN neural network is a physical information neural network that combines physical constraints and data-driven methods to predict the imbalance state. The angle recorder, the timer, and the PINN neural network are connected in parallel to ensure that the outputs of the angle recorder and the timer can be synchronously input into the PINN neural network, while maintaining information flow and collaborative work among the three. Several fully connected layers are configured after the PINN neural network to further integrate and process the angle, time information, and the output of the PINN network, fit the detection time of the PINN neural network output value through the timer output value, and fit the detection angle of the PINN neural network output value through the angle recorder. The fully connected layer learns and fits the complex mapping relationship between the input data and the target output, that is, the imbalance state identifier, through the weight matrix and activation function. Furthermore, the imbalance state prediction network topology is generated based on the angle recorder, the timer, the PINN neural network, and the fully connected layer.

[0040] The vibration state training input matrix data set and the binary data identification data set are retrieved, and the unbalanced state prediction network topology is trained according to the data loss function and the physical loss function. The network parameters are adjusted by the back propagation algorithm until the loss function value converges to an acceptable range, and the final unbalanced state prediction network model is generated. The unbalanced state prediction network can improve the prediction efficiency and the accuracy of the prediction results.

[0041] Further, a dual-constraint loss function is configured, wherein the dual-constraint loss function includes a data loss function and a physical loss function, including: the data loss function is:

[0042] ,in, Represents the data loss value for each M training, Characterizes the quality prediction value of the i-th imbalance point predicted by the j-th training, Characterizes the quality supervision value of the i-th imbalance point predicted by the j-th training, Characterizes the eccentricity distance of the i-th imbalance point predicted by the j-th training, represents the eccentricity distance supervision value of the i-th imbalance point predicted in the j-th training, N represents the number of imbalance points analyzed, and Characterization weights;

[0043] The physical loss function is:

[0044] ,in, Characterize the physical loss value, Characterizes the turning angle of the i-th imbalance point predicted by the j-th training, Characterizes the rated speed, Characterizes the moment of inertia, It represents the deviation parameter of the ith imbalance point in the time domain from the theoretical physical state, and M represents the number of training times of the fitting loss.

[0045] Specifically, the dual-constraint loss function is configured to consider both the prediction accuracy at the data level and the rationality of physical laws when training the imbalance state prediction network, thereby improving the prediction accuracy of the model and the reliability in practical applications. The data loss function is used to quantify the difference between the imbalance point mass and eccentricity distance predicted by the model and the true supervision value. The specific formula is as follows:

[0046] ,

[0047] in, Represents the data loss value for each M training, Characterizes the quality prediction value of the i-th imbalance point predicted by the j-th training, Characterizes the quality supervision value of the i-th imbalance point predicted by the j-th training, Characterizes the eccentricity distance of the i-th imbalance point predicted by the j-th training, represents the eccentricity distance supervision value of the i-th imbalance point predicted in the j-th training, N represents the number of imbalance points analyzed, and Characterize the weight.

[0048] The data loss function calculates the average of the sum of squares of the differences between the predicted values ​​and the true values ​​of the mass and eccentricity distance of each imbalance point in all training samples. and The value of can control the weight of quality and eccentricity distance in the loss function and adjust the learning focus of the model.

[0049] The physical loss function is used to ensure that the prediction results of the model are not only accurate at the data level, but also conform to the laws of physics. The specific expression is as follows:

[0050] ,

[0051] in, Characterize the physical loss value, Characterizes the turning angle of the i-th imbalance point predicted by the j-th training, Characterizes the rated speed, Characterizes the moment of inertia, It represents the deviation parameter of the ith imbalance point in the time domain from the theoretical physical state, and M represents the number of training times of the fitting loss.

[0052] The physical loss function calculates the maximum value of the difference between the angular acceleration caused by each imbalance point and the theoretical calculated value in all training samples. This maximum value is used to reflect the maximum deviation between the model prediction results and the laws of physics. By minimizing this value, the model's prediction results can be made more consistent with the laws of physics.

[0053] By defining and combining the data loss function and the physical loss function and configuring the double-constraint loss function, the prediction accuracy of the imbalance state prediction network can be ensured and the efficiency and reliability of the prediction data can be improved.

[0054] Furthermore, according to the data loss function and the physical loss function, the vibration state training input matrix data set and the binary data identification data set are retrieved, the imbalance state prediction network topology is trained, and the imbalance state prediction network is generated, including: configuring a data loss threshold and a physical loss threshold; when the data loss function satisfies the data loss threshold for a preset number of consecutive times, and the physical loss function satisfies the physical loss threshold for a preset number of consecutive times, the imbalance state prediction network is generated.

[0055] Specifically, a data loss threshold and a physical loss threshold are set. The data loss threshold represents the minimum error tolerance of the model in predicting the imbalance point and eccentricity distance, and is used to measure the prediction accuracy of the model at the data level. The physical loss threshold represents the minimum deviation tolerance between the model prediction result and the physical law, and is used to measure the deviation between the model prediction result and the physical law. Using the vibration state training input matrix data set as input and the binary data identification data set as a label, the vibration state training input matrix data set and the binary data identification data set are input into the imbalance state prediction network, and the network is iteratively trained by the gradient descent method or other optimization algorithms. In each iteration, the data loss function is used to calculate the error between the imbalance point mass and eccentricity distance predicted by the model and the true value to obtain the data loss value. At the same time, the physical loss function is used to calculate the deviation between the rotation angle predicted by the model and the physical model to obtain the physical loss value, and the error is back-propagated to update the network weight. During the training process, the values ​​of the data loss function and the physical loss function are continuously monitored and compared with the set thresholds, and the number of consecutive times that the output value of the data loss function is less than or equal to the data loss threshold is recorded. When the data loss value is less than or equal to the set data loss threshold for a preset number of times (e.g., 10 times), and the physical loss value is also less than or equal to the set physical loss threshold for a preset number of times (e.g., 10 times), the model is considered to have achieved satisfactory training results, and the training can be terminated to generate the final imbalance state prediction network. In this way, it can be ensured that the generated imbalance state prediction network has good performance at both the data and physical levels, improving the accuracy and reliability of the model.

[0056] The method comprises: performing positioning according to the unbalanced point eccentricity distance list, the unbalanced point detection angle list and the unbalanced point detection time list to obtain an unbalanced point positioning list, including: performing positioning according to the unbalanced point eccentricity distance list, the unbalanced point detection angle list and the unbalanced point detection time list to obtain an unbalanced point set at a first time, an unbalanced point set at a second time, and an unbalanced point set at a T-th time; clustering the unbalanced point set at the first time, the unbalanced point set at the second time, and the unbalanced point set at the T-th time to obtain a first balance point triggering time proportion, a second balance point triggering time proportion, and an N-th balance point triggering time proportion; traversing the first balance point triggering time proportion, the second balance point triggering time proportion, and the N-th balance point triggering time proportion, extracting points greater than or equal to a threshold of a triggering time proportion from the first balance point set, the second balance point set, and the T-th balance point set to generate the unbalanced point positioning list.

[0057] Specifically, first, according to the unbalanced point eccentricity list, the unbalanced point detection angle list and the unbalanced point detection time list, at each moment, the unbalanced point detection angle and eccentricity are used to locate, obtain the specific position of the unbalanced point of the grinding wheel assembly, and obtain the unbalanced point set at each moment, including the unbalanced point set at the first moment, the unbalanced point set at the second moment, and the unbalanced point set at the Tth moment. Next, the unbalanced point sets at these moments are clustered based on the spatial position of the unbalanced point using the K-means clustering algorithm or the hierarchical clustering algorithm, and the spatial position is determined by the eccentricity distance and the angle. Through clustering, similar points are classified into one category, and possible repeated detections or unbalanced points at similar positions are identified. For each cluster cluster, its triggering time proportion is calculated, and the triggering time proportion refers to the frequency of occurrence of a certain unbalanced point in all detected unbalanced points, and the first balance point triggering time proportion, the second balance point triggering time proportion, and the Nth balance point triggering time proportion are obtained. By calculating the proportion of balance point triggering moments, it is possible to identify which imbalance points occur frequently, that is, which are more likely to be actual imbalance points.

[0058] Finally, set the trigger moment ratio threshold, traverse the first balance point trigger moment ratio, the second balance point trigger moment ratio, and all the trigger moment ratios in the Nth balance point trigger moment ratio, and extract those points whose ratio is greater than or equal to the preset trigger moment ratio threshold from the first moment unbalanced point set, the second moment unbalanced point set, and the Tth moment unbalanced point set. These points are considered to be stable and frequently occurring unbalanced points. Summarize the extracted points to generate the final unbalanced point positioning list. For example, if the trigger moment ratio threshold is set to 0.6, then only those points whose trigger moment ratio is greater than or equal to 0.6 will be included in the positioning list. Through cluster analysis and threshold comparison, the real unbalanced points detected at multiple times can be effectively identified to avoid the interference of accidental errors, thereby obtaining accurate unbalanced point positioning results. This process ensures the accuracy and reliability of unbalanced point detection of grinding wheel assemblies for sand washing machines.

[0059] Furthermore, the unbalanced point positioning list and the unbalanced point mass list are added with the precession balancing test result, which also includes: extracting the kth unbalanced point mass set of the unbalanced point positioning list from the unbalanced point mass list, the number of unbalanced points in the unbalanced point positioning list ≥ k ≥ 1, and the initial value of k is equal to 1; performing a mode analysis on the kth unbalanced point mass set to obtain the kth unbalanced point characteristic mass; updating the unbalanced point mass list according to the kth unbalanced point characteristic mass, and judging whether k is equal to the number of unbalanced points in the unbalanced point positioning list; if so, adding the precession balancing test result to the unbalanced point positioning list and the unbalanced point mass list; if not, k is increased by one.

[0060] Specifically, first, according to the position of each unbalanced point in the unbalanced point positioning list, the kth unbalanced point mass set is extracted from the unbalanced point mass list, the number of unbalanced points in the unbalanced point positioning list is ≥k≥1, and the initial value of k is equal to 1. Next, the kth unbalanced point mass set is subjected to mode analysis. The mode refers to the value that appears the most times in a set of data. The characteristic quality of the unbalanced point can be determined by mode analysis, that is, the most likely quality value of the unbalanced point under different test conditions. By analyzing the mode of the kth unbalanced point mass set, the characteristic quality of the unbalanced point is obtained, and this value reflects the stable characteristics of the unbalanced point in multiple tests. Then, the characteristic quality of the kth unbalanced point obtained by mode analysis is updated to the unbalanced point mass list. In this way, it can be ensured that the quality value of each unbalanced point is representative and stable. Determine whether k is equal to the number of unbalanced points in the unbalanced point positioning list. If k is equal to the number of unbalanced points in the unbalanced point location list, it means that characteristic masses have been assigned to all unbalanced points, that is, the characteristic masses of all unbalanced points have been updated, and the updated unbalanced point mass list and unbalanced point location list are added to the dynamic balancing test results. If k is not equal to the number of unbalanced points in the unbalanced point location list, it means that there are still unbalanced points that have not been processed, then k is added by one, and the mass set of the next unbalanced point is extracted and analyzed. By judging the update, it is ensured that each unbalanced point has a representative characteristic mass, and this information is accurately recorded in the dynamic balancing test results, so that the dynamic balancing test results can comprehensively and accurately reflect the specific location and characteristic mass of each unbalanced point, and improve the efficiency and accuracy of the dynamic balancing test of the grinding wheel assembly for the sand washing machine.

[0061] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0062] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A dynamic balance test method for a grinding wheel assembly for a sand washing machine, characterized in that: include: Match the test rotating shaft model with the grinding wheel assembly model, wherein the test rotating shaft model corresponds one to one with the grinding wheel assembly model; Performing correlation analysis on a vibration state index set based on the test rotation axis model and the grinding wheel assembly model to obtain a correlated vibration state index; According to the associated vibration state index, combined with the test rotating shaft model and the grinding wheel assembly model, training an unbalanced state prediction network; The grinding wheel assembly is mounted on a test rotating shaft, the grinding wheel assembly is rotated based on a rated speed, and the associated vibration state index is collected by a laser vibration measuring instrument to obtain a vibration state monitoring matrix of the grinding wheel assembly; The vibration state monitoring matrix of the grinding wheel assembly is fitted through the imbalance state prediction network to generate an imbalance point mass list, an imbalance point eccentricity distance list, an imbalance point detection angle list and an imbalance point detection time list; Perform positioning according to the unbalance point eccentricity distance list, the unbalance point detection angle list and the unbalance point detection time list to obtain an unbalance point positioning list; Adding the unbalance point location list and the unbalance point mass list to the precession balance test result; According to the associated vibration state index, combined with the test rotating shaft model and the grinding wheel assembly model, training an unbalanced state prediction network includes: Taking the test rotating shaft model and the grinding wheel assembly model as constraints, and taking the associated vibration state index, unbalance point mass, and unbalance point mass eccentric distance as retrieval targets, data monitoring is performed to obtain an associated vibration index retrieval data set, an unbalance point mass retrieval data set, and an unbalance point mass eccentric distance retrieval data set; Traversing the associated vibration index retrieval data set to construct a vibration state training input matrix data set, wherein the index attribute and the matrix position in any vibration state training input matrix data are in one-to-one correspondence; constructing a binary data identification data set according to the unbalanced point mass retrieval data set and the unbalanced point mass eccentricity distance retrieval data set; Configure a dual-constraint loss function, wherein the dual-constraint loss function includes a data loss function and a physical loss function; According to the data loss function and the physical loss function, the vibration state training input matrix data set and the binary data identification data set are retrieved to train the imbalance state prediction network.

2. A dynamic balance test method for a grinding wheel assembly for a sand washing machine according to claim 1, characterized in that: Based on the test rotating shaft model and the grinding wheel assembly model, a correlation analysis is performed on the vibration state index set to obtain the associated vibration state index, including: According to the test rotating shaft model and the grinding wheel assembly model, a dynamic balancing unqualified sample experimental data set is collected, wherein the dynamic balancing unqualified sample experimental data set includes several groups of vibration state experimental monitoring data and several mass imbalance experimental recording data; Normalizing the plurality of mass imbalance experimental record data to construct a reference sequence; Traversing the plurality of groups of vibration state experimental monitoring data, respectively performing normalization processing, and constructing a plurality of characteristic sequences; Performing grey correlation analysis on the reference sequence and the plurality of characteristic sequences to generate an indicator correlation set; Based on the indicator association degree set, an indicator greater than or equal to an association degree threshold is extracted from the vibration state indicator set and set as the associated vibration state indicator.

3. A dynamic balance test method for a grinding wheel assembly for a sand washing machine according to claim 1, characterized in that: According to the data loss function and the physical loss function, the vibration state training input matrix data set and the binary data identification data set are retrieved to train the imbalance state prediction network, including: Build an angle recorder, wherein the angle recorder is used to record the rotation angle of any position of the grinding wheel assembly relative to the position at time 0; Building a timer, wherein the timer is used to record the rotation time of the grinding wheel assembly; The turning angle counter, the timer, and the PINN neural network are connected in parallel, and a fully connected layer is configured to generate an imbalance state prediction network topology; The fully connected layer is used to fit the outputs of the angle recorder, the timer, and the PINN neural network, fit the detection time of the PINN neural network output value through the timer output value, and fit the detection angle of the PINN neural network output value through the angle recorder; According to the data loss function and the physical loss function, the vibration state training input matrix data set and the binary data identification data set are retrieved, the imbalance state prediction network topology is trained, and the imbalance state prediction network is generated.

4. A dynamic balance test method for a grinding wheel assembly for a sand washing machine as claimed in claim 1, characterized in that: A dual-constraint loss function is configured, wherein the dual-constraint loss function includes a data loss function and a physical loss function, including: The data loss function is: ， in, Represents the data loss value for each M training, Characterizes the quality prediction value of the i-th imbalance point predicted by the j-th training, Characterizes the quality supervision value of the i-th imbalance point predicted by the j-th training, Characterizes the eccentricity distance of the i-th imbalance point predicted by the j-th training, represents the eccentricity distance supervision value of the i-th imbalance point predicted in the j-th training, N represents the number of imbalance points analyzed, and Characterization weights; The physical loss function is: ， in, Characterize the physical loss value, Characterizes the turning angle of the i-th imbalance point predicted by the j-th training, Characterizes the rated speed, Characterizes the moment of inertia, It represents the deviation parameter of the ith imbalance point in the time domain from the theoretical physical state, and M represents the number of training times of the fitting loss.

5. A dynamic balance test method for a grinding wheel assembly for a sand washing machine as claimed in claim 3, characterized in that: According to the data loss function and the physical loss function, the vibration state training input matrix data set and the binary data identification data set are retrieved, the unbalanced state prediction network topology is trained, and the unbalanced state prediction network is generated, including: Configure data loss threshold and physical loss threshold; When the data loss function satisfies the data loss threshold for a preset number of consecutive times, and the physical loss function satisfies the physical loss threshold for a preset number of consecutive times, the imbalance state prediction network is generated.

6. A dynamic balance test method for a grinding wheel assembly for a sand washing machine as claimed in claim 1, characterized in that: Positioning is performed according to the unbalance point eccentricity distance list, the unbalance point detection angle list and the unbalance point detection time list to obtain an unbalance point positioning list, including: Positioning is performed according to the unbalanced point eccentricity distance list, the unbalanced point detection angle list and the unbalanced point detection time list to obtain the unbalanced point set at the first time, the unbalanced point set at the second time, and finally the unbalanced point set at the Tth time; Clustering the unbalanced point set at the first moment, the unbalanced point set at the second moment, and the unbalanced point set at the Tth moment to obtain the first balance point triggering moment proportion, the second balance point triggering moment proportion, and the Nth balance point triggering moment proportion; The first balance point triggering time ratio, the second balance point triggering time ratio, and the Nth balance point triggering time ratio are traversed, and points greater than or equal to the triggering time ratio threshold are extracted from the first time unbalanced point set, the second time unbalanced point set, and the Tth time unbalanced point set to generate the unbalanced point location list.

7. A dynamic balance test method for a grinding wheel assembly for a sand washing machine as claimed in claim 6, characterized in that: Adding the unbalance point location list and the unbalance point mass list to the precession balancing test result, the method also includes: Extracting the kth unbalanced point mass set of the unbalanced point positioning list from the unbalanced point mass list, the number of unbalanced points in the unbalanced point positioning list being ≥ k ≥ 1, and the initial value of k being equal to 1; Performing a mode analysis on the k-th unbalanced point mass set to obtain the k-th unbalanced point characteristic mass; Update the unbalanced point quality list according to the kth unbalanced point characteristic quality, and determine whether k is equal to the number of unbalanced points in the unbalanced point positioning list; If so, adding the precession balance test result to the unbalance point location list and the unbalance point mass list; If not, k increases by one.

Images