A high-precision spindle performance testing method and system

Abstract

The present invention relates to the technical field of machine tool spindles, and specifically to a high-precision spindle performance testing method and system. The present invention collects parameter data related to the performance of high-precision spindles under different working conditions, and respectively integrates multiple features and methods to process and analyze various types of data. The thermal expansion data is processed in combination with the time series characteristics to obtain the first test index; the vibration and dynamic balance data are analyzed in combination with the dynamic balance correction to obtain the second test index; the rigidity and dynamic load data are modeled and processed to capture the global dependency relationship to obtain the third test index; a multi-layer neural network is used to analyze the rotation accuracy and state data to obtain the fourth test index. Finally, all test indicators are integrated to perform nonlinear fusion analysis to generate a comprehensive test index. By analyzing and processing multi-dimensional performance data, the accuracy and reliability of the test are improved, and a comprehensive evaluation of the performance of the high-precision spindle is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of machine tool spindles, and in particular to a high-precision spindle performance testing method and system. Background Art

[0002] High-precision spindles are one of the key core components in industrial production. They ensure that mechanical equipment can achieve high-quality processing results by providing high-precision, high-stability rotational motion. In order to ensure high precision and stability under actual production conditions, it is necessary to perform performance tests on high-precision spindles before leaving the factory. By verifying key indicators such as dynamic balance, thermal stability and vibration control, the long-term reliable operation and processing quality of the equipment can be guaranteed. Through performance testing, potential defects can be identified and eliminated, the life of high-precision spindles can be extended, and production efficiency and product quality can be improved.

[0003] In current research, the performance test of high-precision spindles is mainly focused on the detection of anti-deformation ability. The parameters of static and dynamic test parameters can be used to evaluate the deformation characteristics of the spindle under various loads and operating conditions. For example, by applying external force when the spindle is stationary to measure the spindle deformation, the deformation data of the spindle without the influence of external power can be obtained; or by applying load when the spindle rotates at high speed, and using high-precision sensors or detection devices to monitor the spindle deformation in real time. For another example, virtual modeling and simulation technology is used to generate a three-dimensional spindle structure model to simulate the spindle deformation under different loads and operating conditions, providing a comprehensive high-precision spindle deformation prediction.

[0004] The existing technology has improved the performance test effect of high-precision spindles, but there are still certain defects. The existing technology only considers the detection of deformation, and it is difficult to provide a comprehensive evaluation of the comprehensive performance of high-precision spindles.

[0005] Therefore, a high-precision spindle performance testing method and system are proposed. Summary of the invention

[0006] The purpose of the present invention is to provide a high-precision spindle performance test method and system, by collecting parameter data related to the high-precision spindle performance under different working conditions, and processing various types of data by integrating multiple features and methods. The thermal expansion data is processed in combination with the time series characteristics to obtain the first test index; the vibration and dynamic balance data are analyzed in combination with the dynamic balance correction to obtain the second test index; the rigidity and dynamic load data are modeled and processed to capture the global dependency to obtain the third test index; the rotation accuracy and state data are analyzed using a multi-layer neural network to obtain the fourth test index. Finally, all test indicators are integrated for nonlinear fusion analysis to generate comprehensive test indicators. By analyzing and processing multi-dimensional performance data, the accuracy and reliability of the test are improved, and a comprehensive evaluation of the performance of the high-precision spindle is achieved.

[0007] To achieve the above object, the present invention provides a high-precision spindle performance testing method and system, comprising:

[0008] Design different speed and load conditions, and continuously collect performance-related parameter data of high-precision spindles under different working conditions, including thermal expansion data, vibration and dynamic balance data, rigidity and dynamic load data, and rotation accuracy and status data.

[0009] Furthermore, the process of obtaining the first test indicator includes:

[0010] After preprocessing the thermal expansion data to form time series features, for each time step sequence feature, the initial dependency features between the sequence features and the sequence features of adjacent time steps are calculated to form a sequence dependency feature;

[0011] The sequence dependency feature is input into the fully connected layer to calculate the thermal expansion prediction value, and the first test index is obtained after denormalization.

[0012] ;

[0013] in, Represents the first test indicator, represents the number of time steps, Represents the time step The predicted value of thermal expansion is and They represent the maximum and minimum values ​​of thermal expansion prediction, represents the weight matrix of the fully connected layer, Represents the time step The dependent features of represents the bias term; the formula for obtaining the sequence feature dependency is:

[0014] ;

[0015] in, represents the sequence dependency feature, Represents the time step The sequence characteristics of and represents the time step difference calculation weight matrix, Indicates bias.

[0016] Further, the vibration and dynamic balance data are pre-processed and input into the first convolution layer to obtain local features, and then the data is subjected to dimensionality reduction in the first pooling layer to obtain a preliminary feature map, wherein the preliminary feature map includes a plurality of preliminary feature sub-maps;

[0017] ;

[0018] represents the preliminary feature map, represents the local feature, Represents input data, represents the weight matrix of the first convolutional layer, represents the bias of the first convolutional layer; represents the maximum pooling operation, represents the activation function, Represents a convolution operation.

[0019] The preliminary feature map is input into the second convolutional layer to extract complex features; and then the complex feature map is obtained by dimension reduction in the second pooling layer, wherein the complex feature map includes multiple complex feature sub-maps;

[0020] ;

[0021] in, Represents a complex feature map, Represents the complex features, represents the weight matrix of the second convolutional layer, Represents the bias of the second convolutional layer;

[0022] According to the complex characteristic diagram, combined with the current spindle speed, a correction amount is calculated using a dynamic balancing algorithm; and then the correction amount is combined with the complex characteristic diagram to generate a second test index.

[0023] ;

[0024] in, represents the second test index, Represents the number of convolutional layer channels, Indicates The vibration amplitude extracted from the complex feature sub-graph, Indicates the weight coefficient of dynamic balance correction, Indicates the dynamic balance correction amount, Represents an unbalanced force, represents the angular velocity, Indicates the radius of the correction weight.

[0025] Furthermore, the rigidity and dynamic load data are normalized and organized in time series to form a sequence data structure and then mapped to generate a corresponding sequence feature vector;

[0026] The self-attention mechanism is used to calculate the correlation between each time step data and other time step data, and extract time-dependent features;

[0027] A multi-head attention mechanism is used to capture multi-level features from the time-dependent features;

[0028] ;

[0029] in, Represents the time step and The characteristic vector time series relationship between Represents the multi-head attention mechanism calculation function.

[0030] The multi-level features are nonlinearly transformed through a feedforward neural network, and high-level features are extracted and weighted combined to generate the third test index.

[0031] ;

[0032] in, represents the third test index, represents the number of time steps, represents the number of features in the sequence data, represents the outer layer weight, represents the inner layer weight, and Represent the nonlinear activation functions of the inner and outer layers respectively, and Represent the scaling parameter and the offset parameter respectively; Represents high-level features, Represents the feedforward neural network calculation function; is the interaction term.

[0033] Furthermore, the rotation accuracy and state data are preprocessed and converted into a data feature vector; the data feature vector is in the form of , represents the initial transformation matrix, Represents input data, represents the initial bias, represents a non-linear activation function.

[0034] The data feature vector is input into the first hidden layer, linearly transformed through the first fully connected layer, and a preliminary feature vector is extracted after ReLU processing;

[0035] The preliminary feature vector is input into the second hidden layer, further linearly transformed through the second fully connected layer, and a high-level feature vector is extracted after ReLU processing;

[0036] The high-level feature vectors are linearly combined to obtain a fourth test index.

[0037] ;

[0038] in, represents the fourth test indicator, represents the number of features in the data feature vector, Indicates The high-level feature vector of features, Represents the weight parameter corresponding to the feature, Indicates The activation function of the feature, represents the interaction weight coefficient between high-level feature vectors, Represents the bias term.

[0039] Further, the first test index, the second test index, the third test index and the fourth test index are subjected to nonlinear fusion analysis to obtain a comprehensive test index;

[0040] ;

[0041] Among them, the Indicates the comprehensive test index, represents the nonlinear transfer function, , , and represent the first test index, the second test index, the third test index and the fourth test index respectively, , , and Respectively , , and A nonlinear function of represents the interaction effect term; , , , and Represents the weight coefficient, satisfying .

[0042] A high-precision spindle performance testing system, characterized by comprising:

[0043] Performance data acquisition unit: used to design different working conditions and collect parameter data related to the performance of high-precision spindles under different working conditions; the parameter data include thermal expansion data, vibration and dynamic balance data, rigidity and dynamic load data, and rotation accuracy and state data;

[0044] Thermal expansion analysis unit: used for analyzing and processing the thermal expansion data using a long short-term memory network to obtain a first test index;

[0045] A vibration and dynamic balance analysis unit: used for analyzing and processing the vibration and dynamic balance data by using a convolutional neural network combined with a dynamic balance algorithm to obtain a second test index;

[0046] Rigidity and dynamic load analysis unit: used for modeling and processing the rigidity and dynamic load data using a Transformer model to obtain a third test index;

[0047] A rotation accuracy and state analysis unit: used for analyzing and processing the rotation accuracy and state data using a multi-layer neural network model to obtain a fourth test index;

[0048] Non-linear fusion analysis unit: used for performing non-linear fusion analysis on the first test index, the second test index, the third test index and the fourth test index to obtain a comprehensive test index.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The present invention processes the thermal expansion data into time series features, uses the multi-layer network structure of LSTM to extract the initial time-dependent features and hidden state vectors, and then generates thermal expansion prediction values ​​through the fully connected layer, and obtains the first test index after denormalization; the method captures the prediction of the thermal expansion behavior of the spindle during operation through in-depth analysis and processing of the thermal expansion data. At the same time, the present invention also extracts local features and complex features in the vibration and dynamic balancing data through a convolutional neural network, calculates the correction amount after pooling and dimensionality reduction combined with the dynamic balancing correction algorithm, and finally generates the second test index. The method reflects the vibration of the spindle during operation with the degree of pre-correction by finely processing the vibration characteristics and applying the dynamic balancing algorithm. The method provides data support for the comprehensive evaluation of the performance of high-precision spindles by improving the accuracy of thermal expansion tests and spindle vibration feature tests.

[0051] 2. The present invention organizes the rigidity and dynamic load data into a sequence structure, uses a self-attention mechanism to calculate the correlation between data at different time steps, and extracts time-dependent features; then uses a multi-head attention mechanism to capture multi-level features, performs nonlinear transformation on the multi-level features through a feedforward neural network, and finally generates a third test index through weighted combination. At the same time, the present invention also converts the rotation accuracy and state data into data feature vectors, inputs them into a multi-layer neural network to gradually extract preliminary features and advanced features, and finally generates a fourth detection index through linear combination. This method provides a scientific basis for the comprehensive evaluation of high-precision spindle performance by accurately evaluating the rigidity performance and rotation state of the spindle under dynamic load.

[0052] 3. The present invention collects parameter data related to the performance of high-precision spindles under different working conditions, integrates multiple features such as time dependence and data interaction dependence with deep learning methods, performs targeted processing and analysis on various types of data, and obtains corresponding test indicators respectively. Finally, according to the interaction relationship between various types of data, a nonlinear fusion analysis method is used to generate comprehensive test indicators. This method systematically captures and analyzes the key performance characteristics of high-precision spindles under actual working conditions through the combined application of multiple processing methods, ensuring a comprehensive and accurate comprehensive evaluation of each performance dimension of the high-precision spindle. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a high-precision spindle performance testing method provided by an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of the structure of a high-precision spindle performance testing system provided by an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of a thermal expansion analysis unit flow chart provided in an embodiment of the present invention;

[0056] Figure 4 A schematic diagram of a vibration and dynamic balance analysis unit flow chart provided in an embodiment of the present invention;

[0057] Figure 5 A schematic diagram of a rigidity and dynamic load analysis unit flow chart provided by an embodiment of the present invention;

[0058] Figure 6 A schematic diagram of the flow of the rotation accuracy and state analysis unit provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0060] The present invention provides a high-precision spindle performance test method, which is applied to a high-precision spindle performance test system. The specific method and system flowchart refer to Figure 1 and Figure 2 .

[0061] Embodiment 1

[0062] As an implementation mode of the present invention, this embodiment describes a specific implementation mode for high-precision spindles produced by high-precision spindle manufacturing workshop A.

[0063] Reference Figure 1 S10 in the embodiment is applied to a performance data acquisition unit of a high-precision spindle performance test system. This embodiment first presets different speeds and load conditions, and then collects thermal expansion data, vibration and dynamic balance data, rigidity and dynamic load data, and rotation accuracy and state data under different working conditions. The thermal expansion data includes temperature, spindle length change, spindle diameter change, and load; the vibration and dynamic balance data includes speed, vibration amplitude, unbalance, and acceleration; the rigidity and dynamic load data includes dynamic load, stress, strain, and spindle deformation; the rotation accuracy and state data includes radial runout, axial runout, coaxiality error, and spindle temperature.

[0064] Reference Figure 1 The S20 in the figure is used in the thermal expansion analysis unit of a high-precision spindle performance test system.

[0065] Further, refer to Figure 3 The first test index acquisition step includes: preprocessing the thermal expansion data to form a time series feature , where each Represents the time step The sequence characteristics of Represents the number of time steps; for each time step sequence feature, calculate the initial dependency feature between it and the adjacent time step sequence features to form a sequence dependency feature;

[0066] ;

[0067] in, represents the sequence dependency feature, Represents the time step The dependent features of Represents the time step The sequence characteristics of and represents the time step difference calculation weight matrix, Indicates bias.

[0068] Inputting the sequence dependency feature into a fully connected layer to calculate a thermal expansion prediction value, and obtaining a first test index after denormalization;

[0069] ;

[0070] in, Represents the first test indicator, and They represent the maximum and minimum values ​​of thermal expansion prediction, Represents the time step The predicted value of thermal expansion is represents the weight matrix of the fully connected layer, Represents the bias term.

[0071] This embodiment collects the thermal expansion data of the high-precision spindles manufactured by the high-precision spindle manufacturing workshop A under normal working conditions, and performs processing and analysis of the above implementation method. Table 1 is an example of some data.

[0072] Table 1 Some examples of thermal expansion data

[0073] Data timestamp Temperature (℃) Change in main axis length (μm) Spindle diameter change (μm) Load(N) 8：00 25.0 0.0 0.0 1500 9：00 28.5 1.2 1.8 1500 10：00 32.6 2.4 2.6 1500 11：00 35.3 3.1 2.8 1500 12：00 38.0 5.3 3.2 1500 13：00 40.5 6.8 4.3 1500

[0074] In Table 1, temperature indicates the actual temperature of the spindle during operation, spindle length change indicates the length change of the spindle as the temperature rises, spindle diameter change indicates the diameter change of the spindle as the temperature rises, and load indicates the mechanical load borne by the spindle during operation. According to the analysis and calculation of the thermal expansion analysis unit of the high-precision spindle performance test system provided in this embodiment, the first test index of the high-precision spindle manufactured by workshop A is 1.83.

[0075] This embodiment captures the dynamic effects of factors such as temperature and load on the thermal expansion of the spindle by combining the time series method and calculates the first test index. This method can more accurately evaluate the performance changes of the spindle over time in actual operation and improve the accuracy of the thermal expansion performance test of the spindle.

[0076] This embodiment identifies important factors affecting thermal expansion by analyzing the correlation between different time steps. It is not limited to the characteristics of a single time point, but takes into account the dynamic changes of the entire time series, making the evaluation of thermal expansion more accurate and reliable, and ensuring that the test results under different working conditions are more stable and consistent.

[0077] Reference Figure 1 The S30 in the figure is used in the vibration and dynamic balance analysis unit of a high-precision spindle performance test system.

[0078] Further, refer to Figure 4 After preprocessing the vibration and dynamic balance data, an input feature map with a size of 64×64×1 is formed.

[0079] The input feature map is input into the first convolution layer with a convolution kernel size of 3×3 and a number of 16 to obtain local features; then, the input feature map is reduced in dimension in the first pooling layer with a size of 2×2 and a stride of 2×2 to obtain a preliminary feature map, which contains multiple preliminary feature sub-maps;

[0080] ;

[0081] represents the preliminary feature map, represents the local feature, Represents input data, represents the weight matrix of the first convolutional layer, represents the bias of the first convolutional layer; represents the maximum pooling operation, represents the activation function, Represents a convolution operation.

[0082] The preliminary feature map is input into a second convolutional layer with a size of 3×3 and a number of 32 to extract complex features; and then a second pooling layer with a size of 2×2 and a stride of 2×2 is used for dimensionality reduction to obtain a complex feature map, wherein the complex feature map includes multiple complex feature sub-maps;

[0083] ;

[0084] in, Represents a complex feature map, Represents the complex features, represents the weight matrix of the second convolutional layer, Represents the bias of the second convolutional layer;

[0085] According to the complex characteristic diagram, combined with the current spindle speed, a correction amount is calculated using a dynamic balancing algorithm; and the correction amount is combined with the complex characteristic diagram to generate a second test index;

[0086] ;

[0087] in, represents the second test index, Represents the number of convolutional layer channels, Indicates The vibration amplitude extracted from the complex feature sub-graph, Indicates the weight coefficient of dynamic balance correction, Indicates the dynamic balance correction amount, Represents an unbalanced force, represents the angular velocity, Indicates the radius of the correction weight.

[0088] This embodiment collects vibration and dynamic balance data of high-precision spindles manufactured by high-precision spindle manufacturing workshop A under different preset working conditions to perform processing and analysis of the above implementation method. Table 2 is an example of some data.

[0089] Table 2 Some examples of vibration and dynamic balance data

[0090]

[0091] In Table 2, the speed represents the preset speed of the spindle at different time points, the vibration amplitude represents the vibration amplitude of the spindle at the corresponding speed, the unbalance represents the unbalance of the spindle at the corresponding speed, and the acceleration represents the acceleration caused by vibration. According to the analysis and calculation of the vibration and dynamic balance analysis unit of the high-precision spindle performance test system provided in this embodiment, the second test index of the high-precision spindle manufactured by workshop A is 1.43.

[0092] This embodiment uses a convolutional neural network to process vibration and imbalance data, extracts preliminary features and complex features through multi-layer convolution and pooling, and then combines the dynamic balancing algorithm to correct the imbalance amount and generate a second test index. This method reflects the vibration of the spindle during operation with the pre-correction degree, which improves the test accuracy of the spindle vibration characteristics.

[0093] This embodiment uses a multi-layer convolutional neural network to process vibration and dynamic balancing data, making the extraction of complex features and the dynamic balancing correction process more automated and accurate. By extracting features in layers and combining them with a dynamic balancing algorithm, this method can systematically capture the vibration changes of the spindle at different speeds, making the test indicators more comprehensive and reliable, and helping to more finely evaluate and optimize the spindle performance.

[0094] Reference Figure 1 The S40 in the test system is used as a rigidity and dynamic load analysis unit for a high-precision spindle performance test system.

[0095] Further, refer to Figure 5 , normalizing the rigidity and dynamic load data and organizing them in time series to form a sequence data structure and then mapping to generate a corresponding sequence feature vector;

[0096] The self-attention mechanism is used to calculate the correlation between each time step data and other time step data, and extract time-dependent features;

[0097] A multi-head attention mechanism is used to capture multi-level features from the time-dependent features;

[0098] ;

[0099] in, Represents the time step and The characteristic vector time series relationship between Represents the calculation function of the multi-head attention mechanism;

[0100] Performing nonlinear transformation on the multi-level features through a feedforward neural network, extracting high-level features and performing weighted combination to generate the third test index;

[0101] ;

[0102] in, represents the third test index, represents the number of time steps, represents the number of features in the sequence data, represents the outer layer weight, represents the inner layer weight, and Represent the nonlinear activation functions of the inner and outer layers respectively, and Represent the scaling parameter and the offset parameter respectively; Represents high-level features, Represents the feedforward neural network calculation function; is the interaction term.

[0103] This embodiment collects the rigidity and dynamic load data of the high-precision spindles manufactured by the high-precision spindle manufacturing workshop A to perform processing and analysis of the above implementation. Table 2 is an example of some data.

[0104] Table 3 Some examples of rigidity and dynamic load data

[0105] Data timestamp Dynamic load (N) Stress (MPa) Unbalance (μm / m) Spindle deformation (μm) 8：00 500 45.2 225 0.18 9：00 1000 92.7 463 0.35 10：00 1500 138.9 697 0.53 11：00 2000 185.4 928 0.71 12：00 2500 231.1 1153 0.89 13：00 3000 276.8 1382 1.07

[0106] In Table 3, dynamic load indicates the preset load change applied to the spindle, stress indicates the stress value of the spindle under different loads, strain indicates the strain amount of the spindle under different loads, and spindle deformation indicates the actual deformation amount of the spindle under different loads. According to the analysis and calculation of the rigidity and dynamic load analysis unit of the high-precision spindle performance test system provided in this embodiment, the third test index of the high-precision spindle manufactured by workshop A is 0.95.

[0107] This embodiment normalizes the rigidity and dynamic load data and organizes them in time series, extracts multi-level features by combining self-attention mechanism and multi-head attention mechanism, and then performs nonlinear transformation and weighted combination through feedforward neural network to generate the third test index. This method can accurately capture the complex relationship in time series data and extract key features, thereby improving the accuracy of rigidity and dynamic load testing.

[0108] This embodiment uses self-attention and multi-head attention mechanisms to automatically learn important time-dependent features from time series data, and then extracts and combines high-level features through a feedforward neural network. This method not only enhances the expressiveness of the model, but also ensures the comprehensive capture of multi-dimensional and nonlinear features, thereby improving the evaluation accuracy of high-precision spindle rigidity and dynamic load.

[0109] Reference Figure 1The S50 in the test system is used in the rotation accuracy and state analysis unit of a high-precision spindle performance test system.

[0110] Further, refer to Figure 6 , converting the rotation accuracy and state data into a data feature vector after preprocessing;

[0111] The data feature vector is in the form of , represents the initial transformation matrix, Represents input data, represents the initial bias, represents a non-linear activation function.

[0112] The data feature vector is input into the first hidden layer, linearly transformed through the first fully connected layer, and a preliminary feature vector is extracted after ReLU processing;

[0113] The preliminary feature vector is input into the second hidden layer, further linearly transformed through the second fully connected layer, and a high-level feature vector is extracted after ReLU processing;

[0114] Performing linear combination on the high-level feature vectors to obtain a fourth test index;

[0115] ;

[0116] in, represents the fourth test indicator, represents the number of features in the data feature vector, Indicates The high-level feature vector of features, Represents the weight parameter corresponding to the feature, Indicates The activation function of the feature, represents the interaction weight coefficient between high-level feature vectors, Represents the bias term.

[0117] This embodiment collects the rotation accuracy and state data of the high-precision spindles manufactured by the high-precision spindle manufacturing workshop A under different preset working conditions to perform processing and analysis of the above implementation. Table 4 is an example of some data.

[0118] Table 4 Some examples of rotation accuracy and status data

[0119]

[0120] In Table 4, radial runout reflects radial stability of the spindle during operation, and axial runout reflects axial stability of the spindle. According to the analysis and calculation of the rotation accuracy and state analysis unit of the high-precision spindle performance test system provided in this embodiment, the fourth test index of the high-precision spindle manufactured by workshop A is 2.16.

[0121] This embodiment pre-processes the rotation accuracy and state data, converts them into data feature vectors, and then gradually extracts and combines high-level features through linear transformation and nonlinear activation function processing of a multi-layer fully connected network, and finally generates the fourth test index. This method can effectively extract and enhance the relationship between complex features, ensure that the results of rotation accuracy and state analysis are more accurate and stable, and improve the reliability of the spindle rotation performance test.

[0122] This embodiment uses a multi-layer neural network to extract data features. By extracting and combining features layer by layer, the potential relationship between rotation accuracy and state data is fully explored. In particular, the high-level feature combination after nonlinear processing can better reflect the complexity of the spindle rotation state. This method improves the processing capability of multi-dimensional data, making the final test indicators more comprehensive and targeted, and provides theoretical support for the performance evaluation of high-precision spindles.

[0123] Reference Figure 1 The S60 in the figure is used in a nonlinear fusion analysis unit of a high-precision spindle performance test system.

[0124] Further, the first test index, the second test index, the third test index and the fourth test index are subjected to nonlinear fusion analysis to obtain a comprehensive test index;

[0125] ;

[0126] Among them, the Represents the comprehensive test index, represents the nonlinear transfer function, , , and represent the first test index, the second test index, the third test index and the fourth test index respectively, , , and Respectively , , and A nonlinear function of represents the interaction effect term; , , , and Represents the weight coefficient, satisfying .

[0127] According to the analysis and calculation by the nonlinear fusion analysis unit of the high-precision spindle performance test system provided in this embodiment, it is obtained that the comprehensive test index of the high-precision spindle manufactured by workshop A is 1.23.

[0128] This embodiment generates a comprehensive test index by performing nonlinear fusion analysis on the first test index, the second test index, the third test index and the fourth test index. This method comprehensively integrates multi-dimensional performance indicators and combines nonlinear conversion functions and interaction influence terms. It not only considers the independent contribution of each indicator, but also effectively captures the interaction between them, and improves the collaborative analysis effect between different indicators. The final comprehensive test index more accurately reflects the overall performance of the high-precision spindle, enhances the processing capability and analysis effect of complex performance data, and ensures that the evaluation results are more representative and have reference value for decision-making.

[0129] Embodiment 2

[0130] As an implementation mode of the present invention, this embodiment describes a specific implementation mode for high-precision spindles produced by high-precision spindle manufacturing workshop B.

[0131] Reference Figure 1 S10 in the embodiment is applied to a performance data acquisition unit of a high-precision spindle performance test system. This embodiment first presets different speeds and load conditions, and then collects thermal expansion data, vibration and dynamic balance data, rigidity and dynamic load data, and rotation accuracy and state data under different working conditions. The thermal expansion data includes temperature, spindle length change, spindle diameter change, and load; the vibration and dynamic balance data includes speed, vibration amplitude, unbalance, and acceleration; the rigidity and dynamic load data includes dynamic load, stress, strain, and spindle deformation; the rotation accuracy and state data includes radial runout, axial runout, coaxiality error, and spindle temperature.

[0132] Reference Figure 1 The S20 in the figure is used in the thermal expansion analysis unit of a high-precision spindle performance test system.

[0133] Further, refer to Figure 3 , the step of obtaining the first test indicator includes:

[0134] Preprocessing the thermal expansion data to form time series features;

[0135] For each time step sequence feature, calculate the initial dependency feature between it and the adjacent time step sequence features to form a sequence dependency feature;

[0136] Inputting the sequence dependency feature into a fully connected layer to calculate a thermal expansion prediction value, and obtaining a first test index after denormalization;

[0137] This embodiment collects the thermal expansion data of the high-precision spindles manufactured by the high-precision spindle manufacturing workshop B under normal working conditions, and performs processing and analysis of the above implementation method. Table 5 is an example of some data.

[0138] Table 5 Some examples of thermal expansion data

[0139]

[0140] According to the analysis and calculation of the thermal expansion analysis unit of the high-precision spindle performance test system provided in this embodiment, it is obtained that the first test index of the high-precision spindle manufactured by the workshop B is 1.21.

[0141] Reference Figure 1 The S30 in the figure is used in the vibration and dynamic balance analysis unit of a high-precision spindle performance test system.

[0142] Further, refer to Figure 4 After preprocessing the vibration and dynamic balance data, an input feature map with a size of 64×64×1 is formed.

[0143] The input feature map is input into the first convolution layer with a convolution kernel size of 3×3 and a number of 16 to obtain local features; then, the input feature map is reduced in dimension in the first pooling layer with a size of 2×2 and a stride of 2×2 to obtain a preliminary feature map, which contains multiple preliminary feature sub-maps;

[0144] The preliminary feature map is input into a second convolutional layer with a size of 3×3 and a number of 32 to extract complex features; and then a second pooling layer with a size of 2×2 and a stride of 2×2 is used for dimensionality reduction to obtain a complex feature map, wherein the complex feature map includes multiple complex feature sub-maps;

[0145] According to the complex characteristic diagram, combined with the current spindle speed, a correction amount is calculated using a dynamic balancing algorithm; and the correction amount is combined with the complex characteristic diagram to generate a second test index;

[0146] This embodiment collects the vibration and dynamic balance data of the high-precision spindles manufactured by the high-precision spindle manufacturing workshop B under different preset working conditions to perform processing and analysis of the above implementation method. Table 6 is an example of some data.

[0147] Table 6 Some examples of vibration and dynamic balance data

[0148]

[0149] According to the analysis and calculation of the vibration and dynamic balance analysis unit of the high-precision spindle performance test system provided in this embodiment, it is obtained that the second test index of the high-precision spindle manufactured by workshop B is 1.16.

[0150] Reference Figure 1 The S40 in the test system is used as a rigidity and dynamic load analysis unit for a high-precision spindle performance test system.

[0151] Further, refer to Figure 5 , normalize the rigidity and dynamic load data and organize them in time series to form a sequence data structure and then map them to generate corresponding sequence feature vectors; calculate the correlation between each time step data and other time step data through a self-attention mechanism to extract time-dependent features; use a multi-head attention mechanism to capture multi-level features from the time-dependent features; perform nonlinear transformation on the multi-level features through a feedforward neural network, extract high-level features and perform weighted combination to generate the third test index;

[0152] This embodiment collects the rigidity and dynamic load data of the high-precision spindles manufactured by the high-precision spindle manufacturing workshop B to perform processing and analysis of the above implementation. Table 7 is an example of some data.

[0153] Table 7 Some examples of rigidity and dynamic load data

[0154]

[0155] According to the analysis and calculation of the rigidity and dynamic load analysis unit of the high-precision spindle performance test system provided in this embodiment, it is obtained that the third test index of the high-precision spindle manufactured by workshop B is 2.51.

[0156] Reference Figure 1 The S50 in the test system is used in the rotation accuracy and state analysis unit of a high-precision spindle performance test system.

[0157] Further, refer to Figure 6 , converting the rotation accuracy and state data into a data feature vector after preprocessing;

[0158] The data feature vector is input into the first hidden layer, linearly transformed through the first fully connected layer, and a preliminary feature vector is extracted after ReLU processing;

[0159] The preliminary feature vector is input into the second hidden layer, further linearly transformed through the second fully connected layer, and a high-level feature vector is extracted after ReLU processing;

[0160] Performing linear combination on the high-level feature vectors to obtain a fourth test index;

[0161] This embodiment collects the rotation accuracy and state data of the high-precision spindles manufactured by the high-precision spindle manufacturing workshop B under different preset working conditions to perform processing and analysis of the above implementation. Table 8 is an example of some data.

[0162] Table 8 Partial example of rotation accuracy and status data

[0163]

[0164] According to the analysis and calculation of the rotation accuracy and state analysis unit of the high-precision spindle performance test system provided in this embodiment, it is obtained that the fourth test index of the high-precision spindle manufactured by workshop B is 1.67.

[0165] Reference Figure 1 The S60 in the figure is used in a nonlinear fusion analysis unit of a high-precision spindle performance test system.

[0166] Further, the first test index, the second test index, the third test index and the fourth test index are subjected to nonlinear fusion analysis to obtain a comprehensive test index;

[0167] According to the analysis and calculation by the nonlinear fusion analysis unit of the high-precision spindle performance test system provided in this embodiment, it is obtained that the comprehensive test index of the high-precision spindle manufactured by workshop A is 1.97.

[0168] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0169] This embodiment introduces a variety of data processing and analysis methods to perform detailed calculations and evaluations on various performance indicators of high-precision spindles. Each method conducts in-depth mining and processing on specific performance parameters, and finally generates comprehensive test indicators through nonlinear fusion analysis. This comprehensive method improves the accuracy and stability of various performance evaluations, making the overall performance analysis of high-precision spindles more comprehensive and reliable.

[0170] This embodiment integrates multiple data processing ideas, such as time series analysis, attention mechanism, multi-layer feature extraction and nonlinear fusion, to fully mine the data correlation and complex features of various test indicators, and perform fusion analysis in an expressive way. This method can systematically capture the interaction between different performance indicators, and provides a highly comprehensive, accurate and adaptable evaluation framework for the performance test of high-precision spindles, thereby improving the scientificity and practicality of spindle performance evaluation.

Claims

1. A high-precision spindle performance testing method, characterized in that: include: Collect parameter data related to high-precision spindle performance under different working conditions, including thermal expansion data, vibration and dynamic balance data, rigidity and dynamic load data, and rotation accuracy and status data; The thermal expansion data is analyzed and processed in combination with the time series characteristics to obtain a first test index; the vibration and dynamic balance data are analyzed and processed using a dynamic balance correction algorithm to obtain a second test index; the rigidity and dynamic load data are modeled and processed to capture global dependencies to obtain a third test index; the rotation accuracy and state data are analyzed and processed using a multi-layer neural network model to obtain a fourth test index; The first test index, the second test index, the third test index and the fourth test index are subjected to nonlinear fusion analysis to obtain a comprehensive test index; Wherein, S represents the comprehensive test index, f(·) represents a nonlinear conversion function, Y1, Y2, Y3 and Y4 represent the first test index, the second test index, the third test index and the fourth test index respectively, , γ and θ represent the nonlinear functions of Y1, Y2, Y3 and Y4 respectively, ξ represents the interaction term; ω1, ω2, ω3, ω4 and ω5 represent weight coefficients, satisfying 2. A high-precision spindle performance testing method according to claim 1, characterized in that: The process of obtaining the first test indicator includes: After preprocessing the thermal expansion data to form time series features, for each time step sequence feature, the dependency features between the sequence features and the sequence features of adjacent time steps are calculated to form a sequence dependency feature; The sequence dependency feature is input into the fully connected layer to calculate the thermal expansion prediction value, and the first test index is obtained after denormalization.

3. A high-precision spindle performance testing method according to claim 1, characterized in that: The calculation formula of the first test index is: Among them, Y1 represents the first test index, T represents the number of time steps, represents the predicted value of thermal expansion at time step t, Y max and Y min Respectively represent the maximum and minimum values ​​of thermal expansion prediction, W y represents the weight matrix of the fully connected layer, d t represents the dependent features at time step t, b y represents the bias term; the formula for obtaining the sequence feature dependency is: d t =W a ·(x' t -x' t-1 )+W b ·(x' t -x' t+1 )+b; Among them, D represents the sequence dependence feature, x' t represents the sequence features of time step t, W a and W b represents the time step difference calculation weight matrix, and b represents the bias.

4. A high-precision spindle performance testing method according to claim 1, characterized in that: The process of obtaining the second test indicator includes: The vibration and dynamic balance data are pre-processed and input into the first convolution layer to obtain local features, and then the dimensions are reduced in the first pooling layer to obtain a preliminary feature map; the preliminary feature map is input into the second convolution layer to extract complex features; and then the dimensions are reduced in the second pooling layer to obtain a complex feature map; According to the complex characteristic diagram, combined with the current spindle speed, a correction amount is calculated using a dynamic balancing algorithm; and then the correction amount is combined with the complex characteristic diagram to generate a second test index.

5. A high-precision spindle performance testing method according to claim 1, characterized in that: The formula for obtaining the second test index is: Among them, Y2 represents the second test index, M represents the number of convolutional layer channels, and A i represents the vibration amplitude extracted from the i-th complex feature subgraph, k represents the weight coefficient of dynamic balance correction, Δm represents the dynamic balance correction amount, F unbalance Represents an unbalanced force, represents the angular velocity, r represents the radius of the correction weight; the formula for obtaining the complex feature map is: Among them, P2 represents a complex feature map, F2 represents a complex feature, W2 represents the weight matrix of the second convolutional layer, and b2 represents the bias of the second convolutional layer; P1 represents a preliminary feature map, F1 represents a local feature, X represents input data, W1 represents the weight matrix of the first convolutional layer, and b1 represents the bias of the first convolutional layer; MaxPool(·) represents the maximum pooling operation, ReLU(·) represents the activation function, and Conv(·) represents the convolution operation.

6. A high-precision spindle performance testing method according to claim 1, characterized in that: The process of obtaining the third test indicator includes: The rigidity and dynamic load data are normalized and organized in time series to form a sequence data structure and then mapped to generate a corresponding sequence feature vector; The correlation between each time step data and other time step data is calculated through the self-attention mechanism to extract time-dependent features; a multi-head attention mechanism is used to capture multi-level features from the time-dependent features; The third test index is generated by extracting high-level features from the multi-level features through a feedforward neural network.

7. A high-precision spindle performance testing method according to claim 1, characterized in that: The formula for obtaining the third test index is: Among them, Y3 represents the third test index, U represents the number of time steps, V represents the number of features in the sequence data, α p represents the outer layer weight, β pq represents the inner layer weight, f int (·) and g ext (·) represent the nonlinear activation functions of the inner and outer layers, respectively, and γ pq and δ pq Respectively represent the scaling parameter and the offset parameter; F h represents high-level features, F(·) represents the feedforward neural network calculation function, and ε is the interaction term; Depend * t-series Represents multi-level features, and its acquisition formula is: in, represents the time step t i and t j The feature vector time series relationship between them, G(·) represents the multi-head attention mechanism calculation function.

8. A high-precision spindle performance testing method according to claim 1, characterized in that: The process of obtaining the fourth test indicator includes: The rotation accuracy and state data are preprocessed into data feature vectors and then input into the first hidden layer, linearly transformed through the first fully connected layer, and a preliminary feature vector is extracted after ReLU processing; The preliminary feature vector is input into the second hidden layer, further linearly transformed through the second fully connected layer, and a high-level feature vector is extracted after ReLU processing; The high-level feature vectors are linearly combined to obtain a fourth test index.

9. A high-precision spindle performance testing method according to claim 1, characterized in that: The acquisition formula of the fourth test index is: Wherein, Y4 represents the fourth test index, d represents the number of features in the data feature vector, Represents the high-level feature vector of the mth feature, ω m Represents the weight parameter corresponding to the feature, f m (·) represents the activation function of the mth feature, represents the interaction weight coefficient between high-level feature vectors, and b3 represents the bias term.

10. A high-precision spindle performance testing system, characterized in that: include: Performance data acquisition unit: used to design different working conditions and collect parameter data related to high-precision spindle performance under different working conditions; The parameter data include thermal expansion data, vibration and dynamic balance data, rigidity and dynamic load data and rotation accuracy and state data; Thermal expansion analysis unit: used for analyzing and processing the thermal expansion data using a long short-term memory network to obtain a first test index; A vibration and dynamic balance analysis unit: used for analyzing and processing the vibration and dynamic balance data by using a convolutional neural network combined with a dynamic balance algorithm to obtain a second test index; Rigidity and dynamic load analysis unit: used for modeling the rigidity and dynamic load data using a Transformer model to obtain a third test index; A rotation accuracy and state analysis unit: used for analyzing and processing the rotation accuracy and state data using a multi-layer neural network model to obtain a fourth test index; Non-linear fusion analysis unit: used for performing non-linear fusion analysis on the first test index, the second test index, the third test index and the fourth test index to obtain a comprehensive test index.

Images