Method for determining dynamic errors of a main spindle of a numerically controlled machine tool and computer program product

Abstract

Embodiments of the present application provide a method and computer program product for determining dynamic error of a main shaft of a numerical control machine tool. A vibration signal sequence of the main shaft of the numerical control machine tool is obtained; the vibration signal sequence is input into a pre-trained deep time convolution network, and a dynamic error category to which the dynamic error of the main shaft of the numerical control machine tool belongs is predicted by the deep time convolution network, wherein the deep time convolution network is trained by a plurality of sets of sample data, and the plurality of sets of sample data are constructed based on data collected by an experimental platform pre-built for simulating an actual working environment of the main shaft of the numerical control machine tool. Since the experimental platform can simulate the real working environment of the main shaft of the machine tool, the sample data collected can truly reflect the actual working state of the main shaft of the machine tool, and thus the dynamic error model trained by the sample data has better performance, and the predicted dynamic error has higher precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of spindle precision control of a numerical control machine tool, in particular to a method for determining dynamic error of a spindle of a numerical control machine tool and a computer program product. BACKGROUND

[0002] With the rapid development of manufacturing industry, as a key equipment for precision machining, the performance of a numerical control machine tool directly affects the quality and production efficiency of machined parts. As a core component of the machine tool, the dynamic error of the spindle of the numerical control machine tool is directly related to the machining precision and stability of the machine tool. Therefore, accurate measurement and control of the dynamic error of the spindle of the numerical control machine tool are of great significance to improve the machining precision and reliability of the machine tool. In related technologies, when determining the dynamic error of the spindle of the numerical control machine tool, the dynamic error of the whole numerical control machine tool (such as the movement of the whole machine tool in the axial or radial direction) is usually directly measured, and then the dynamic error of the spindle is estimated based on the dynamic error of the whole numerical control machine tool. This way of determining the dynamic error of the spindle of the machine tool is not direct and accurate enough, and therefore it is necessary to provide a more accurate scheme for determining the dynamic error of the spindle of the machine tool. SUMMARY

[0003] Therefore, the present application provides a method for determining dynamic error of a spindle of a numerical control machine tool and a computer program product.

[0004] According to a first aspect of the present application, a method for determining dynamic error of a spindle of a numerical control machine tool is provided, the method comprising:

[0005] obtaining a vibration signal sequence for characterizing vibration of the spindle of the numerical control machine tool;

[0006] inputting the vibration signal sequence into a pre-trained deep time convolution network, and predicting, by the deep time convolution network, an error category to which the dynamic error of the spindle of the numerical control machine tool belongs, wherein a dynamic error range covered by the spindle of the numerical control machine tool in a working process is divided into a plurality of error categories, and each error category corresponds to a sub-range in the dynamic error range;

[0007] The deep time convolution network is trained by a plurality of sets of sample data, and the plurality of sets of sample data are constructed based on data collected by an experimental platform pre-built for simulating an actual working environment of the spindle of the numerical control machine tool, the experimental platform comprising a spindle load simulation system, an experimental spindle, a spindle drive system, a control system, a dynamic error collection system, and a vibration signal collection system;

[0008] The spindle load simulation system is configured to receive loading force information set by a user from the control system, and apply a loading force to the experimental spindle based on the loading force information;

[0009] The main shaft driving system is configured to receive user-set rotation speed information from the control system and drive the experimental main shaft to rotate based on the rotation speed information.

[0010] The vibration signal acquisition system comprises one or more sensors configured to acquire vibration signals of the experimental main shaft and send the vibration signals to the control system.

[0011] The dynamic error acquisition system is configured to acquire dynamic errors of the experimental main shaft and send the dynamic errors to the control system.

[0012] According to a second aspect of the present application, a computer program product is provided, the computer program product comprising a computer program which, when executed, implements the method mentioned in the first aspect.

[0013] According to a third aspect of the present application, an electronic device is provided, the electronic device comprising a processor, a memory, a computer program stored in the memory and executable by the processor, and the processor implements the method mentioned in the first aspect when executing the computer program.

[0014] According to a fourth aspect of the present application, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, and the computer program, when executed, implements the method mentioned in the first aspect.

[0015] By using the scheme provided in the present application, an experimental platform simulating the actual working environment of the main shaft of a numerical control machine tool can be constructed, the experimental platform can simulate the working state of the main shaft of the machine tool under different operating conditions and acquire vibration signal sequences and dynamic errors of the main shaft under different working states to construct sample data, which can be used to train a dynamic error prediction model. Since the experimental platform can simulate the real working environment of the main shaft of the machine tool, the sample data acquired can truly reflect the actual working state of the main shaft of the machine tool and can cover the actual working state of the main shaft of the machine tool under different operating conditions, and thus the dynamic error model trained by using the sample data has better performance and the predicted dynamic error of the main shaft of the machine tool is more accurate.

[0016] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and are not limiting to the present application. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 is a structural schematic diagram of an experimental platform of an embodiment of the present application.

[0019] Figure 2 is a schematic diagram of a vertical spindle simulation loading experimental platform of an embodiment of the present application.

[0020] Figure 3 is a schematic diagram of a horizontal spindle simulation loading experimental platform of an embodiment of the present application.

[0021] Figure 4 is a flowchart of a method for determining dynamic error of a spindle of a numerical control machine tool of an embodiment of the present application.

[0022] Figure 5 is a flowchart of a method for determining dynamic error of a spindle of a numerical control machine tool of another embodiment of the present application.

[0023] Figure 6 is a structural schematic diagram of a deep time convolution network of an embodiment of the present application.

[0024] Figure 7 is a trend graph of model training accuracy and loss value under four RPM conditions.

[0025] Figure 8 is a vertical spindle error value distribution graph.

[0026] Figure 9 is a corresponding relationship graph of vibration signal sequence and dynamic error category of a training set of a vertical spindle.

[0027] Figure 10 is a chaotic matrix graph of dynamic error prediction of a vertical spindle under each rotational speed.

[0028] Figure 11 is a structural schematic diagram of a vertical spindle simulation loading experimental platform.

[0029] Figure 12 is a structural schematic diagram of a horizontal spindle simulation loading experimental platform.

[0030] Figure 13 is a schematic diagram of a logic structure of an electronic device of an embodiment of the present application. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0032] With the rapid development of manufacturing industry, as the key equipment for precision machining, the performance of numerical control machine tools directly affects the quality and production efficiency of machined parts. As the core component of the machine tool, the dynamic error of the spindle of the numerical control machine tool is directly related to the machining accuracy and stability of the machine tool. The dynamic error of the spindle of the numerical control machine tool refers to the error that the rotating axis of the spindle deviates from the ideal position due to various factors such as mechanical vibration, bearing wear, thermal deformation, etc. during the operation of the machine tool spindle. This error will affect the dimensional accuracy, shape accuracy and position accuracy of the machined parts, thereby affecting the overall machining quality. The dynamic error of the spindle of the numerical control machine tool mainly includes radial error and axial error, wherein the radial error refers to the error formed by the movement of the spindle rotation center line in the radial direction, and the axial error refers to the error formed by the movement of the spindle rotation center line in the axial direction.

[0033] In related technologies, when determining the dynamic error of the spindle of the numerical control machine tool, the dynamic error of the whole numerical control machine tool (such as the movement of the whole machine tool in the axial or radial direction) is usually directly measured, and then the dynamic error of the spindle is estimated based on the dynamic error of the whole numerical control machine tool. This way of determining the dynamic error of the spindle of the machine tool is not accurate enough, and the precision needs to be improved.

[0034] Considering that it is difficult to directly measure the dynamic error of the spindle of the numerical control machine tool by using a sensor during the operation of the numerical control machine tool, and the dynamic error of the spindle of the numerical control machine tool is usually related to the operating state parameters (such as vibration data, temperature, current, etc.) of the spindle, such as the operating parameters of the vibration data of the spindle, which are also relatively easy to measure. Therefore, the internal correlation between the operating state parameters of the spindle and the dynamic error of the spindle can be learned by using a model, and then the trained model can be used to predict the dynamic error of the spindle based on the collected operating parameters (such as vibration signals) of the spindle of the numerical control machine tool.

[0035] A significant feature of the dynamic error of the spindle of the numerical control machine tool is that it has strong time correlation characteristics, that is, as time goes on and the cumulative number of wear or use increases, the dynamic error may also gradually increase. Therefore, when predicting the dynamic error by using a model, the time correlation characteristics of the dynamic error can be considered to obtain more accurate prediction results.

[0036] In recent years, with the development of deep learning technology, especially the application of neural networks such as Deep Temporal Convolutional Network (TCN) and Bidirectional Long Short-Term Memory (BiLSTM) in sequence data processing, a new idea has been provided for the measurement of dynamic errors of CNC machine tool spindles. These neural networks can effectively process time series data and capture long-term dependencies in the data, which has potential application value for analyzing the dynamic errors of machine tool spindles under different working conditions.

[0037] However, existing deep learning-based methods still face some challenges in practical applications, such as how to construct an effective training data set to obtain a large number of valid sample data for training the neural network; how to improve the richness and coverage of the samples to cover different application scenarios of the CNC machine tool spindle to improve the generalization ability of the trained neural network; how to extract more useful features from the vibration signals of the spindle for the training of the neural network to improve the prediction accuracy of the neural network, etc. The solutions provided by the embodiments of the present application aim to solve one or more of the above technical problems, so that the dynamic errors of the CNC machine tool spindle can be measured more accurately.

[0038] Based on this, the embodiments of the present application provide a model training method, which can train a dynamic error prediction model. The dynamic error prediction model learns the internal relationship between the vibration signal sequence of the CNC machine tool spindle and the dynamic error, and then, during the working process of the CNC machine tool, the vibration signal sequence of the CNC machine tool spindle can be collected, so that the model can predict the dynamic error of the spindle based on the vibration signal sequence. In addition, considering that the dynamic error is a continuous numerical value, in order to simplify the model prediction difficulty and improve the model prediction accuracy, the dynamic error range covered in the actual working process of the CNC machine tool spindle can be divided into multiple error categories, and each error category corresponds to a sub-range in the dynamic error range. For example, assuming that the dynamic error of the CNC machine tool spindle does not exceed 6μm, the dynamic error can be divided into 6 categories, and each interval of 1μm is taken as a category, i.e. 0-1μm, 1-2μm, 2-3μm, 3-4μm, 4-5μm, 5-6μm, etc. Of course, the division method of the error categories and the corresponding dynamic error range of each error category (e.g. 1μm or 0.5μm) can be flexibly set based on actual requirements, and the present application examples are not limited.

[0039] In some embodiments, the dynamic error prediction model can be a deep temporal convolutional network. In some embodiments, the dynamic error prediction model can be a bidirectional long short-term memory network. Deep temporal convolutional networks and bidirectional long short-term memory networks have better performance in processing sequence data, can effectively process time series data, capture long-term dependencies in the data, and are more suitable for analyzing data such as dynamic errors of machine tool spindles under different working conditions, which have time correlation characteristics.

[0040] In some embodiments, an experimental platform can be built in advance to simulate the actual working environment of the numerical control machine tool spindle, and collect corresponding sample data for training the dynamic error prediction model. For example, the sample data can include a vibration signal sequence (hereinafter referred to as a sample vibration signal sequence) reflecting the vibration of the experimental spindle in the experimental platform, and a label corresponding to the sample vibration signal sequence, which is used to indicate the error category corresponding to the sample vibration signal sequence.

[0041] In some embodiments, in order to obtain a large amount of training data that truly reflects the actual working conditions of the numerical control machine tool spindle and improve the richness and coverage of the training data, the applicant designs an experimental platform for simulating the actual working environment of the numerical control machine tool spindle to collect a large amount of training data through the experimental platform. The structure of the experimental platform is shown in Figure 1 The experimental platform includes a spindle load simulation system, an experimental spindle, a spindle drive system, a control system, a dynamic error collection system, and a vibration signal collection system. The control system is in communication connection with the spindle load simulation system, the spindle drive system, the control system, the dynamic error collection system, respectively. The spindle load simulation system is used to receive the user-set load force information from the control system and apply a load force to the experimental spindle based on the load force information. The spindle drive system is used to receive the user-set speed information from the control system and drive the spindle to rotate based on the speed information. The vibration signal collection system includes one or more vibration sensors for collecting vibration signals of the experimental spindle during operation and sending them to the control system. The dynamic error collection system is used to collect the dynamic error of the experimental spindle and send it to the control system. The control system can include a user interface, and the user can set the operating conditions of the experimental spindle through the control interface, such as the size of the load force, the mode of the load force (constant force loading or variable force loading), the direction of the load force, the speed, the running time, etc., so as to simulate the running state of the numerical control machine tool spindle under different operating conditions.

[0042] The vibration signal collection system can collect vibration conditions of the experimental spindle under different operating conditions, such as axial vibration conditions, radial vibration conditions, and the like. The dynamic error collection system can collect dynamic errors of the experimental spindle under different operating conditions, and then sample data can be constructed based on the collected data of the two to enable the neural network to learn the internal correlation between the two.

[0043] By building the above experimental platform, the load simulation system in the experimental platform can accurately simulate the loading force applied to the experimental spindle in different directions and modes, which enables the experimental platform to truly reflect the stress process of the machine tool spindle when machining parts in actual processing, thereby more truly simulating the actual working state of the machine tool spindle. The spindle driving system can flexibly drive the spindle to rotate according to the rotational speed information set by the user received by the control system, which helps to simulate the operating state of the machine tool under different machining conditions. The vibration data collection system can accurately capture the vibration signal of the experimental spindle, and the dynamic error collection system can accurately collect the dynamic error of the experimental spindle, and then sample data can be constructed based on the vibration signal data and dynamic error data collected by the two.

[0044] Through the above experimental platform, the real state of the numerical control machine tool spindle under different operating conditions can be simulated, and sample data that can truly reflect the actual working state of the spindle can be collected for training the model and improving the prediction accuracy of the model. Moreover, the sample data can cover various operating conditions, which is more abundant and has a wider coverage, and can further improve the generalization ability of the trained model.

[0045] Of course, in addition to the above structure, the experimental platform can also include other components based on actual needs, such as a rotational speed monitoring device for monitoring the rotational speed of the experimental spindle, a temperature sensor for monitoring the temperature of the experimental spindle, and a current sensor for monitoring the current of the experimental spindle, and then the change trend of the dynamic error can be analyzed in combination with the monitored rotational speed, current, temperature, and the like of the experimental spindle. Of course, the experimental platform can also include an error instrument for error correction of the data collected by the sensors, and the like. The specific structure can be flexibly set based on actual needs, and the embodiments of the present application do not limit it.

[0046] In some embodiments, the main shaft load simulation system comprises: an air pump, an air pressure servo valve, an air cylinder, a reversing relay, a loading mechanism, and a tension sensor. The air pump is used to provide an air source, the air pressure servo valve is located between the air pump and the air cylinder, and is used to adjust the size of the air pressure output by the air pump to the air cylinder, so as to exert different loading forces on the experimental main shaft, so as to simulate the loading forces in the axial or radial direction that the main shaft of the numerical control machine tool is subjected to when machining parts. The reversing relay is located between the air pump and the air cylinder, and is used to adjust the flow direction of the air pressure output by the air pump to the air cylinder, so as to realize the bidirectional operation of the air cylinder, so that different directions of loading force, such as radial loading force and axial loading force, can be simulated. The air cylinder is used to drive the loading mechanism to move, so that the loading mechanism exerts a loading force on the experimental main shaft. The tension sensor is arranged at the connection between the loading mechanism and the air cylinder, and is used to detect the current loading force and feed back to the control system, so that the control system can adjust the air pressure servo valve based on the current loading force and the user-set loading force, so that the loading mechanism finally exerts a loading force on the experimental main shaft. The user-set loading force.

[0047] In some embodiments, the dynamic error acquisition system can comprise a standard rod arranged at the tool position of the experimental main shaft, and at least three eddy current sensors, wherein one eddy current sensor is arranged in each of the X, Y and Z directions of the standard rod, and is used to acquire the dynamic error of the experimental main shaft. The standard rod can be used in cooperation with the eddy current sensor to measure the distance information in the X, Y and Z directions at the tool position of the experimental main shaft, i.e. the distance of the axial and tangential runout of the experimental main shaft. The eddy current sensor can non-contact measure the distance between the standard rod and the tool position of the main shaft, thereby acquiring dynamic error data with high precision, and by arranging the eddy current sensor in the X, Y and Z directions, the dynamic error of the main shaft in three spatial dimensions can be comprehensively evaluated.

[0048] Considering that the main shaft of the numerical control machine tool includes two types of horizontal main shaft and vertical main shaft, in order to cover different types of main shafts and collect data in the actual working process of different types of main shafts, the embodiments of the present application design two experimental platforms, a vertical main shaft simulation loading experimental platform and a horizontal main shaft simulation loading experimental platform, as shown in Figure 2 , which is a schematic diagram of a vertical main shaft simulation loading experimental platform, and Figure 3 , which is a schematic diagram of a horizontal main shaft simulation loading experimental platform. By providing two experimental platform designs of vertical and horizontal, the working state of the main shaft under different installation modes can be simulated respectively, thereby enhancing the adaptability of the experiment and the universality of the results.

[0049] In some embodiments, the experimental platform is a vertical spindle simulation loading experimental platform, and the experimental spindle is a vertical spindle. Among them, the air cylinders in the load simulation system can include three, two of which are arranged at an angle of 90° to realize the radial application of loading force to the vertical spindle, and the other is used to realize the axial application of loading force to the vertical spindle. By applying radial loading force through two air cylinders arranged at an angle of 90° and axial loading force through one air cylinder, various load conditions that the vertical spindle may encounter in actual machining process can be accurately simulated, so that the experimental platform can adapt to different test requirements.

[0050] In some embodiments, the experimental platform is a horizontal spindle simulation loading experimental platform, and the experimental spindle is a horizontal spindle. Among them, the air cylinders in the load simulation system can include three, two of which are arranged at an angle of 45° to realize the radial application of loading force to the horizontal spindle, and the other is used to realize the axial application of loading force to the horizontal spindle. By applying radial loading force through two air cylinders arranged at an angle of 45° and axial loading force through one air cylinder, various load conditions that the horizontal spindle may encounter in actual machining process can be accurately simulated, so that the experimental platform can adapt to different test requirements.

[0051] In some embodiments, the experimental platform is a vertical spindle simulation loading experimental platform, and the experimental spindle is a vertical spindle. For the vertical spindle simulation loading experimental platform, the experimental platform further includes a ground platform for supporting and fixing the vertical spindle. The vibration signal acquisition system includes three vibration sensors respectively installed at the front bearing of the vertical spindle, the loading mechanism and the ground platform. Considering that the vertical spindle has higher precision requirements, for the vertical spindle, three vibration sensors can be arranged to collect vibration signals in different directions for predicting dynamic error, so as to obtain more accurate dynamic error. By installing vibration sensors at different key positions of the vertical spindle, the vibration characteristics of the spindle under different working conditions can be accurately captured, so that the dynamic error of the spindle can be more accurately evaluated.

[0052] In some embodiments, the experimental platform is a horizontal spindle simulation loading experimental platform, and the experimental spindle is a horizontal spindle. The horizontal spindle simulation loading experimental platform is built above the ground iron with standard T-shaped slot, and the total weight of the ground iron is more than 2 tons. By building the experimental platform above the ground iron with standard T-shaped slot, and setting the total weight of the ground iron to be relatively large, additional stability can be provided for the experimental platform, reducing the vibration and displacement that may occur during the experiment, thereby ensuring the accuracy of the measurement results.

[0053] In some embodiments, a plurality of sets of sample data can be collected using the experimental platform provided in the above embodiments, and then a dynamic error measurement model can be trained using the plurality of sets of sample data. Each set of sample data includes a sample vibration signal sequence and a label corresponding to the sample vibration signal sequence, and the label is used to indicate the error category corresponding to the sample vibration signal sequence. Each set of sample data can be determined based on the following manner: during the operation of the experimental spindle in the experimental platform, the original vibration signal sequence collected by the vibration signal collection system in a target time period is obtained, and the original dynamic error sequence synchronously collected by the dynamic error collection system in the target time period is obtained, then the sample vibration signal sequence in each set of sample data can be determined based on the original vibration signal, such as directly using the original vibration signal as the sample vibration signal sequence, or performing some preprocessing on the original vibration signal, such as normalization processing, feature extraction processing, etc., and then using the processed vibration signal sequence as the sample vibration signal sequence. Then the mean of the original dynamic error sequence can be determined, and the error category corresponding to the mean can be determined as the label of the sample vibration signal sequence.

[0054] In some embodiments, in order to make the collected sample data cover various working states of the spindle of the numerical control machine tool as much as possible, and improve the richness and comprehensiveness of the sample data, the plurality of sets of sample data can be constructed based on the data collected by the vibration signal collection system and the dynamic error collection system under different running conditions of the experimental spindle, wherein the different running conditions include a plurality of rotating speeds and a plurality of load force combinations. For example, data of the experimental spindle under a plurality of rotating speeds (such as 1000, 2000, 3000, 4000, 5000) and different axial forces (such as 0N, 500N, 1000N, 1500N, 2000N), different radial forces (such as 0N, 700N, 1400N, 2100N, 2800N) can be collected to obtain a large amount of sample data. By collecting data under different running conditions, including a plurality of rotating speeds and a plurality of load force combinations, it is ensured that the data set can comprehensively cover various working conditions that the spindle of the numerical control machine tool may encounter.

[0055] In addition, in order to collect data that truly reflects the working environment of the spindle of the numerical control machine tool, in some embodiments, the process of applying a load force to the experimental spindle at each rotational speed includes a constant force loading stage and a variable force loading stage. In the constant force loading stage, a first load force is used to load the experimental spindle at an early stage, and a second load force is used to load the experimental spindle at a later stage. The first load force is smaller than the second load force. The design of the constant force loading stage and the variable force loading stage simulates different load conditions that the spindle of the machine tool may encounter during actual machining, making the experimental results closer to the real application scenario. By controlling the load force in the constant force loading stage to be smaller at the initial stage and larger at the later stage, the stress condition of the spindle during actual machining of the part can be more realistically simulated.

[0056] In some embodiments, considering that there may be some differences in the vibration signals collected by different vibration sensors or under different test conditions, in order to eliminate the differences caused by the equipment and test conditions, the sample vibration signal sequence can be a vibration signal sequence obtained by normalizing the original vibration signal of the experimental spindle. The process of normalization is as follows: the original vibration signal sequence collected is divided into multiple vibration signal sequence blocks, the mean and variance of each vibration signal sequence block are determined, and the minimum mean and maximum variance are determined from the mean and variance of each vibration signal sequence block. For each original vibration signal in the original vibration signal sequence, the difference between the original vibration signal and the minimum mean, and the ratio of the maximum variance, are used as the normalized vibration signal of the original vibration signal.

[0057] The normalization process ensures that different vibration signal sequences have a unified scale, which helps to eliminate the effects of different test conditions or different sensors, making the data more standardized. Through normalization, the model can perform consistently on different data sets, improving the model's generalization ability to unseen data. Moreover, normalized data can enhance the discriminability between different features, helping the model better learn and identify features related to dynamic error. The normalization process can reduce the impact of noise by adjusting the variance of the signal, thereby improving the quality of the signal. In addition, normalized data can simplify the model training process, as the model does not need to learn the scale information of the data during training, which helps to improve the accuracy of the model in predicting dynamic error.

[0058] In the model training stage, if the sample vibration signal sequence in the sample data is normalized, then in the model application stage, the vibration signal sequence collected from the numerical control machine tool spindle can also be normalized and input into the trained model to predict the dynamic error.

[0059] In some embodiments, the sample vibration signal sequence can be a new vibration signal sequence obtained after feature extraction and fusion of the original vibration signal sequence of the collected experimental spindle. For example, after obtaining the original vibration signal sequence of the experimental spindle collected by the experimental platform, the original vibration signal sequence can be processed in blocks to obtain a plurality of vibration signal sequence blocks, then each vibration signal sequence block can be processed by local mean decomposition to obtain the first sample signal feature and the second sample signal feature corresponding to the vibration signal sequence block, and the first sample signal feature sequence can be constituted by the first sample signal feature corresponding to each vibration signal sequence block in the plurality of vibration signal sequence blocks, and the second sample signal feature sequence can be constituted by the second sample signal feature corresponding to each vibration signal sequence block in the plurality of vibration signal sequence blocks. Then the first sample signal feature sequence and the second sample signal feature sequence can be fused to obtain a fused vibration signal sequence, wherein the i-th vibration signal in the fused vibration signal sequence is obtained by weighted fusion of the i-th signal feature in the first sample signal feature sequence and the i-th signal feature in the second sample signal feature sequence, and as i increases, the fusion weight of the first sample signal feature in the first sample signal feature sequence also increases, wherein i is a positive integer, then the fused vibration signal sequence can be used as the sample vibration signal sequence in each group of sample data as the input of the model for training the model. The fusion weight can increase step by step, for example, the fusion weight of the first sample signal feature in the next group of signal features to be fused can be increased by a certain value based on the fusion weight of the first sample signal feature in the previous group of signal features to be fused, so that the model can learn the influence of different features on dynamic error. By block processing and local mean decomposition of the original vibration signal sequence, more detailed features can be extracted from the vibration signal, which helps to understand the vibration characteristics of the machine tool spindle, and by constructing the first sample signal feature sequence and the second sample signal feature sequence, the information related to dynamic error in the vibration signal can be better expressed, and by fusing the two sequences, the advantages of the two feature sequences can be combined, the feature combination can be optimized by weighted fusion, and the prediction ability of the model for dynamic error can be improved. And by dynamically adjusting the fusion weight of each group of signal features, the model can automatically "learn" which features are more useful for the current task, and the model can adapt to different signal features, which helps the model to maintain good performance under different conditions. The fused vibration signal sequence can be used as the input of the deep learning model, which helps to improve the accuracy of dynamic error prediction and enhances the generalization ability of the model to data under different working conditions and conditions.

[0060] In the model training stage, if the sample vibration signal sequence in the sample data is subjected to local mean decomposition and fusion processing, then in the model application stage, the vibration signal sequence collected from the spindle of the numerical control machine tool can also be subjected to local mean decomposition and fusion processing, and then input into the trained model to predict the dynamic error.

[0061] In some embodiments, the collected original vibration signal can be subjected to normalization processing first, and then the vibration signal sequence subjected to normalization processing can be subjected to the above-mentioned local mean decomposition and fusion processing to obtain a new fusion vibration signal sequence for training the model.

[0062] In some embodiments, each set of sample data includes a sample vibration signal sequence and a label corresponding to the sample vibration signal sequence, the label being used to indicate an error category corresponding to the sample vibration signal sequence. The dynamic error prediction model can be trained based on the following manner: the sample vibration signal sequence corresponding to each set of sample data can be input into a preset model, and the model can output a prediction probability of each set of sample vibration signal sequence for each error category in a plurality of error categories, respectively. Then, a target loss can be constructed based on the prediction probability of each set of sample vibration signal sequence for each error category in the plurality of error categories, the actual probability of each set of sample vibration signal sequence for each error category in the plurality of error categories, and the weight value of each set of sample vibration signal sequence, and the model parameters of the model can be adjusted based on the target loss to train the model, thereby obtaining the dynamic error prediction model, wherein the weight value of each set of sample vibration signal sequence is negatively correlated with the proportion of noise signal in the vibration signal sequence.

[0063] In some embodiments, the target loss is determined by the following formula:

[0064]

[0065] wherein N is the number of sample vibration sequence signals input into the model, K is the total number of error categories into which the dynamic error range covered by the spindle of the numerical control machine tool in the working process is divided, zi is the weight value of the i th sample vibration sequence signal, aij is the prediction probability of the model predicting that the i th sample vibration sequence signal belongs to the j th error category, and yij is the actual probability that the i th sample vibration sequence signal belongs to the j th error category.

[0066] Considering that the proportion of noise in different samples (i.e., different sample vibration signal sequences) is different, when constructing the loss function, the weight value of each sample can be adjusted based on the noise adaptability in each sample, and the loss function can balance the contribution of different samples to the overall loss, especially for samples with more noise, giving smaller weight, so as to reduce the interference of noise signal to the model and improve the accuracy of the trained model.

[0067] After the dynamic error prediction model is trained, the trained dynamic error prediction model can be used to predict the dynamic error of the spindle of the numerical control machine tool. For example, during the operation of the numerical control machine tool, the vibration signal sequence of the spindle can be collected by the vibration sensor, and the vibration signal sequence is input into the model to predict the error category to which the current dynamic error belongs, such as whether the dynamic error is 1-2 pm or 2-3 pm, and the like.

[0068] Further, the embodiment of the present application also provides a method for determining the dynamic error of the spindle of the numerical control machine tool, which can be executed by an electronic device in which a pre-trained dynamic error prediction model is deployed. The electronic device can be a mobile phone, a computer, a server, or the like. The pre-trained dynamic error prediction model can be a deep time convolution network or a bidirectional long short-term memory network.

[0069] As shown in the method shown in Figure 4 , the method can include the following steps:

[0070] S402, obtaining a vibration signal sequence for representing the vibration condition of the spindle of the numerical control machine tool;

[0071] In step S402, the vibration signal sequence of the spindle of the numerical control machine tool during the operation process can be obtained by the vibration sensor.

[0072] S404, inputting the vibration signal sequence into a pre-trained dynamic error prediction model, and predicting the error category to which the dynamic error of the spindle of the numerical control machine tool belongs by the dynamic error prediction model, wherein the dynamic error range covered by the spindle of the numerical control machine tool during the operation process is divided into a plurality of error categories, and each error category corresponds to a sub-range in the dynamic error range; the dynamic error prediction model is trained by a plurality of sample data, and the plurality of sample data is constructed based on the data collected by the experimental platform pre-built for simulating the actual working environment of the spindle of the numerical control machine tool.

[0073] In step S404, the vibration signal sequence can be input into the pre-trained dynamic error prediction model, and the error category to which the dynamic error of the numerical control machine tool belongs is predicted by the model.

[0074] The specific structure of the experimental platform can refer to the description in the above embodiments, and will not be repeated here. The training method of the dynamic error prediction model can also refer to the description in the above embodiments, and will not be repeated here.

[0075] In some embodiments, in order to ensure that different vibration signal sequences have a unified scale, so as to eliminate the influence of different test conditions or different sensors, so that the data is more standardized, the vibration signal sequence input into the dynamic error prediction model can be a vibration signal sequence obtained by normalizing the collected original vibration signal of the main shaft of the numerical control machine tool, wherein the process of normalization is as follows: the collected original vibration signal sequence is divided into a plurality of vibration signal sequence blocks, the mean and variance of each vibration signal sequence block are determined, and the minimum mean and maximum variance are determined from the mean and variance corresponding to each of the plurality of vibration signal sequence blocks, for each original vibration signal in the original vibration signal sequence, the difference between the original vibration signal and the minimum mean, and the ratio of the maximum variance, are taken as the normalized vibration signal of the original vibration signal. For example, the following formula can be used to determine the normalized vibration signal:

[0076]

[0077] wherein x* (t) is the normalized vibration signal, xs(t) represents the original vibration signal, Emin represents the minimum mean of the means of each vibration sequence block, and Dmax represents the maximum variance of the variances of each vibration sequence block.

[0078] In some embodiments, in order to extract more effective features from the vibration signal sequence, so as to improve the accuracy of the prediction results of the model, the vibration signal sequence input into the model can be a new vibration signal sequence obtained by feature extraction and fusion on the original vibration signal sequence of the collected main shaft of the numerical control machine tool. For example, as shown in FIG. 4, the vibration signal sequence input into the model can be a new vibration signal sequence obtained by feature extraction and fusion on the original vibration signal sequence of the collected main shaft of the numerical control machine tool. Figure 5As shown, the original vibration signal sequence 9 of the spindle of the numerical control machine tool collected by the vibration sensor can be obtained (S502), and then the original vibration signal sequence can be processed in blocks to obtain a plurality of vibration signal sequence blocks (S504), and then each vibration signal sequence block can be processed by local mean decomposition to obtain a first signal feature and a second signal feature corresponding to the vibration signal sequence block, and a first signal feature sequence is constituted by the first signal features corresponding to each vibration signal sequence block in the plurality of vibration signal sequence blocks, and a second signal feature sequence is constituted by the second signal features corresponding to each vibration signal sequence block in the plurality of vibration signal sequence blocks (S506). Then, the first signal feature sequence and the second signal feature sequence can be fused to obtain a fused vibration signal sequence (S508), wherein the i-th vibration signal in the fused vibration signal sequence is obtained by weighted fusion of the i-th signal feature in the first signal feature sequence and the i-th signal feature in the second signal feature sequence, and as i increases, the fusion weight of the first signal feature in the first signal feature sequence also increases, wherein i is a positive integer. Then, the fused vibration signal sequence can be input into the trained dynamic error prediction model (such as a bidirectional long short-term memory network) to predict the current dynamic error of the numerical control machine tool spindle (S510). The fusion weight can be increased step by step, for example, the fusion weight of the first sample signal feature in the next group of signal features to be fused can be increased by a certain value based on the fusion weight of the first sample signal feature in the previous group of signal features to be fused, so that the model can learn the influence of different features on the dynamic error. The first signal feature and the second signal feature can represent different types of features of the vibration signal, for example, the high frequency part and the low frequency part of the vibration signal.

[0079] In some embodiments, the original vibration signal collected can be normalized first, and then the vibration signal sequence after normalization can be processed by the above local mean decomposition and fusion to obtain a new fused vibration signal sequence, which is input into the model, so that the model predicts the dynamic error of the numerical control machine tool spindle based on the fused vibration signal sequence.

[0080] In some embodiments, the dynamic error prediction model can be a deep time convolutional network, wherein the deep time convolutional network includes an input layer, a feature extraction layer, a full connection layer, and an output layer, wherein the feature extraction layer includes 4 residual blocks connected in series, the input layer is used to obtain the input vibration signal sequence and input into the feature extraction layer, the feature extraction layer is used to extract features from the vibration signal sequence, the full connection layer is used to perform full connection processing on the features output by the feature extraction layer, and the output layer is used to determine the error category corresponding to the vibration signal sequence based on the features output by the full connection layer.

[0081] The schemes of the above embodiments can be freely combined to obtain new schemes in the absence of conflicts. For the sake of brevity, they will not be listed one by one.

[0082] The method for determining the dynamic error of a main shaft of a numerical control machine tool provided in the application is described below in combination with a specific embodiment.

[0083] In this embodiment, a deep time convolution network can be trained to predict the dynamic error of the main shaft of the numerical control machine tool, and specifically can include the following steps:

[0084] S1, a main shaft simulation loading experiment platform is built, and a vibration sensor for detecting radial and / or axial vibration of the main shaft and a dynamic error acquisition system are arranged on the main shaft.

[0085] The main shaft simulation loading experiment platform is divided into a vertical main shaft simulation loading experiment platform and a horizontal main shaft simulation loading experiment platform (the specific construction of the experiment platform is written at the end of the embodiment). The diameter of the cylinder used in the experiment platform reaches 100 mm, the output contact surface of the top rod is not more than 20 mm, and the theoretical air pressure loading force can reach 4KN. The vertical main shaft simulation loading experiment platform and the horizontal main shaft simulation loading experiment platform have the same loading principle and method. According to the input table loading spectrum, the functions of loading force setting, speed control and data acquisition are automatically realized; the radial / axial loading force is 4000N at most, the maximum loading frequency is 10Hz (sine signal, 50 output points per cycle), the main shaft dynamic error (rotary error, axial runout, etc.) can be monitored in real time, and the minimum resolution is 0.25μm.

[0086] S2, start the main shaft, and the main shaft rotates for not less than 10 hours per day. The data of the main shaft running for 100 days on the experiment platform is collected. According to the 100-day wear experiment of the experiment platform, the cumulative wear time is more than 1000 hours, and the vibration signal data set and the dynamic error data set of the main shaft achieving the rotation speeds of 1000, 2000, 3000 and 4000 are mainly collected;

[0087] After the spindle start-up experiment, the experiment process includes two parts of wear loading process and data collection process, and the overall process is carried out according to the mode of initial constant force loading, then increasing the axial and lateral constant force loading, and integrating the variable force loading. The time series of each sensor are collected to form a data set. Through the analysis of the data collection, the influence of the rotation speed, stress and running time on the spindle error is found. The influence of the rotation speed on the spindle error is that the higher the rotation speed, the greater the spindle error. When RPM<1500, although the stress is different, the spindle error tends to be the same broken line. When RPM>1500, the change rate of the spindle error is small, but when the rotation speed is 4000, the error value is greatly affected by the stress. The influence of the stress on the spindle error is that the stress has little effect on the spindle error fluctuation at low rotation speed, and has great effect on the spindle error fluctuation at high rotation speed. When 1500<RPM<3500, the error values of the experimental spindle are very close, and the stress has little effect. The influence of the spindle error on the use time is that the spindle precision retention is good when RPM=1000, and the error value gradually increases with the use time when RPM=2000, 3000 and 4000.

[0088] Therefore, the spindle simulation loading experiment platform combines and verifies the stress conditions. The axial force of the spindle is collected as 0N, 500N, 1000N, 1500N and 2000N, the radial stress is 0N, 700N, 1400N, 2100N and 2800N, and the rotation speed is 1000, 2000, 3000, 4000 and 5000 under each variable constant force.

[0089] S3, a deep time convolution network is constructed, the whole network architecture is as shown in Figure 6 The input vibration signal sequence is normalized in the input layer, and then the four residual blocks are used to extract the features of the input vibration signal sequence. Then, the extracted features are fully connected by the fully connected layer to flatten the vibration signal sequence, and finally the error category is identified by the output layer (i.e. Softmax layer). The dynamic error of the spindle can be divided into multiple error categories (Target a1~Target aK), and the interval between different error categories is 1pm.

[0090] When training the deep time convolution network, the following loss function can be constructed:

[0091]

[0092] wherein, loss is the target function, N is the number of sample vibration sequence signals input into the model, K is the total number of error categories into which the dynamic error range covered by the numerical control machine tool spindle in the working process is divided, zi is the weight value of the i th sample vibration sequence signal, aij is the prediction probability of the model that the i th sample vibration sequence signal belongs to the j th error category, and yij is the actual probability that the i th sample vibration sequence signal belongs to the j th error category.

[0093] S4, using the data collected in the S2 step, the data is divided into a training set and a test set, and in 100 days of data, the ratio of the training set and the test set is 9:1; for the entire sample, the proportion is equivalent to 100 days of samples: 180,000 samples in the first 90 days of the 100 days are used as the training set, and 20,000 samples in the last 10 days are used as the test set. It is equivalent to the first 90 days of data for training the model, and the last 10 days for verifying the performance of the model.

[0094] S5, the training set in S4 is imported into the deep time convolution network in S3 for model training, and then the test set can be imported into the trained model for verifying the performance of the model, and the indirect measurement accuracy under the conditions of RPM=1000, RPM=2000, RPM=3000 and RPM=4000 can be verified.

[0095] (1) The data collected by the horizontal spindle simulation loading experiment platform is used to train and verify the model

[0096] The indirect measurement accuracy under the conditions of RPM=1000, RPM=2000, RPM=3000 and RPM=4000 is verified, and the training process draws a waveform graph for the training process of Matlab at each speed of 1000, and by configuring the output during training, the corresponding training accuracy and loss value are output according to each iteration, and finally the results of each output are plotted into a graph, and the accuracy and loss value changes of the sample training process under each speed are as shown in Figure 7 With the increase of the number of training iterations, the training accuracy is constantly improved, and the training loss value is constantly tending to 0, indicating that the network architecture under the four speeds can maintain high convergence, and the convergence of speed 1000 is better than that of other speeds, mainly because the spindle can introduce less noise under lower speed, and it is easy to realize feature extraction and learning.

[0097] After the model is trained, the performance of the trained model can be verified by using the test set. In order to observe the relationship between the true error value and the actual error value, the difference between the predicted result and the actual result can be presented in the form of a chaotic matrix.

[0098] The verification method of this part adopts TCN-based prediction measurement to actually perform equal-length expansion according to the vibration sequence collection frequency (20 KSa). Then, the single error value can be expanded by 20000 data, thereby causing obvious differences in error values in the chaos matrix.

[0099] Although there are correct and incorrect predictions under different rotational speed conditions, the accuracy of measuring the dynamic error of the main shaft under each RPM condition is 91.23%, 91.78%, 94.12%, and 91.65% respectively. Through the measurement results: it is shown that when RPM=3000, the measurement result of the dynamic error of the main shaft is optimal. In fact, the rated rotational speed of the experimental main shaft is RPM=2800, and the test rotational speed RPM=3000 is the closest to the rated rotational speed. Under the condition of RPM=3000, the types of error values are the least, and then under the conditions of RPM=1000, RPM=2000, and RPM=4000, the number of required classifications of RPM=3000 is the least, the number of samples that can be trained for each error is relatively more than that of other rotational speeds, thereby the final prediction accuracy is the highest.

[0100] (2) Use the data collected by the vertical main shaft simulation loading experimental platform to train and verify the model

[0101] The error values are segmented according to 1 um intervals, and the number of occurrences is counted. The statistical distribution results are shown in Figure 8 . Except for the 4-class error value distribution at the rotational speed RPM=3000, all others only have 3-class error values. With the increase of rotational speed, the central distribution of the main shaft error value is also different. When the rotational speed RPM=1000, the most common error value is 3 um; when the rotational speed RPM=2000, the most common error value is 3 um; when the rotational speed RPM=3000, the most common error value is 4 um; when the rotational speed RPM=4000, the most common error value is 4 um; when the rotational speed RPM=5000, the most common error value is 4 um or 5 um.

[0102] At the same time, it is found that the error value of the vertically installed main shaft (i.e. vertical main shaft) experimental platform is low, and the horizontally installed main shaft (i.e. horizontal main shaft) is mainly used for rough machining, and the vertically installed main shaft is mainly used for five-axis CNC machine tools, which can be said to be a finishing main shaft. The data collected by the experimental platform can be preprocessed to construct the corresponding relationship between a single error value and a sequence as shown in Figure 10 .

[0103] Figure 9 ​​​​​​​​​​​​​​In the five small graphs, the small graphs are arranged according to the corresponding relationship of the vibration sequence and the error class from left to right, which represents RPM = 1000, RPM = 2000, RPM = 3000, RPM = 4000 and RPM = 5000. In addition, the vibration amplitude can be seen to have some differences, such as when the rotation speed RPM = 1000, the vibration amplitude of the error value of 4um is obviously lower than that of 2um and 3um; when the rotation speed RPM = 2000, the waveform amplitude fluctuation of the error value of 4um is faster than that of 3um and 5um; for the rotation speed RPM = 3000, the vibration amplitude of the error class of 3um is higher than that of other error classes; similarly, when RPM = 5000, the error class of 5um is also significantly different from other error classes. The data verification of the vertical spindle simulation loading experiment platform uses the same way as the data verification of the horizontal spindle simulation loading experiment platform, except that the maximum training round is modified. The training round is modified from the original 100 rounds to 200 rounds.

[0104] For the setting of test samples and prediction samples, the essence is that the input sequence processing has been completed, and the ratio of input and test samples is 120000:30000 = 4:1, which changes relative to the horizontal sample ratio, which is equivalent to using fewer samples for model training. Through the data, it can be known that during the entire training process, the training accuracy is gradually improved, and finally each rotation speed is close to 100% straight line, which indicates that the training accuracy convergence is good; the loss value is also gradually reduced during the entire training process, and finally the loss value under each rotation speed is close to 0, which indicates that the training loss value convergence is good.

[0105] The model verification result is as follows: the final chaotic matrix under each rotation speed is as shown in Figure 10 Figure 10 The small graphs in each small graph are arranged in order from top to bottom and from left to right, which are the chaotic matrices representing the difference between the prediction result and the true result when the rotation speed is RPM = 1000, RPM = 2000, RPM = 3000, RPM = 4000 and RPM = 5000, respectively. Figure 10 It can be easily seen from

[0106] ​Take the chaotic matrix with RPM=1000 as an example: when the rotation speed RPM=1000, the true value is 4um, the number of times of predicting 4um is 3421, the number of times of predicting 2um is 9, and the number of times of predicting 3um is 71; when the true error value is 2um, the number of times of predicting correctly is 1499, and there is no error value predicted as other categories; for the error value of 3um, the number of times of predicting error is 48, and the number of times of predicting correctly is 34952. In addition, the error prediction accuracy at each rotation speed is calculated by using the mean error, and finally the error prediction accuracy at each rotation speed is obtained as: 96.68%, 94.69%, 96.85%, 95.89% and 96.90%.

[0107] The data results show that TCN can achieve more than 90% prediction accuracy for the training model, TCN is a selection of time memory association, and completes the indirect measurement verification of horizontal spindle dynamic error, realizes indirect measurement at rotation speeds of 1000, 2000, 3000 and 4000, and the measurement accuracies are 96.68%, 96.69%, 96.85% and 95.89% respectively. Since the vertical experimental platform is generated from the horizontal experimental platform, in addition to the related verification discussion similar to the horizontal experimental platform, the structure optimization of the vertical experimental platform is increased, and finally the indirect measurement at rotation speeds of 1000, 2000, 3000, 4000 and 5000 is realized, and the measurement accuracies are 96.68%, 94.69%, 96.85%, 95.89% and 96.90% respectively.

[0108] In this embodiment, the specific structure of the horizontal spindle simulation loading experimental platform involved in the S1 step is as shown in Figure 11 The specific structure of the horizontal experimental platform is as follows:

[0109] The pneumatic loading hardware part is mainly composed of an air pump providing air source, three air pressure servo valves for controlling air pressure, and two reversing relays. Among the three air pressure servo valves and two reversing relays, one air pressure servo valve is used for axial loading of the main shaft, and two air pressure servo valves are used for radial loading of the main shaft corresponding to one reversing relay. The horizontal main shaft experimental platform part is mainly composed of a programmed horizontal main shaft, a vibration sensor I, three tension sensors I, three air cylinders I, and a main shaft simulation loading mechanism. The programmed horizontal main shaft is mainly connected with a variable frequency power supply of a frequency converter and an encoder. Two of the three air cylinders are connected with the loading mechanism to realize radial loading of the horizontal main shaft and form a 45° angle. One of the three air cylinders realizes axial loading of the horizontal main shaft. A tension sensor is added at the connection between each air cylinder and the horizontal main shaft simulation loading mechanism. The radial or axial loading of the horizontal main shaft is completed through the air cylinder I of the horizontal main shaft experimental platform. The diameter of the air cylinder I used in the experimental platform reaches 100 mm, the output contact surface of the top rod is not more than 20 mm, and the theoretical air pressure loading force can reach 4 KN.

[0110] During the simulation loading process, the rotation of the horizontal main shaft drives the high-speed rotation of the inner ring of the ball bearing, and the outer ring and the fixed sleeve can be fixed. Through the design of the simulation loading mechanism, the loading process during the simulation machining of the main shaft is realized through the control of the air cylinder. Not only the waste of actual machining materials can be saved, but also the full automatic loading process can be realized through the loading spectrum.

[0111] The acquisition and control part is mainly composed of a data acquisition device, a main shaft driver, and a computer display. The vibration sensor I is connected with the data acquisition device through a BNC joint, which is used to realize the closed-loop control of force and the closed-loop control of the rotation speed of the main shaft, and the acquisition of data of the tension sensor. The acquisition and control unit is a PXIe-1082 which controls the pneumatic loading unit through an analog output (AO) channel to control the air cylinder through the pneumatic servo valve, so as to realize the force loading of the loading mechanism. The force sensor feeds back to the analog input (AI) channel of the PXIe-1082 through a 4-wire differential signal, so that the PXIe-108 acquisition device can realize the closed-loop monitoring of the loading force. The PXIe-1082 communicates with the main shaft driver through an RS485-USB interface to realize the control of the rotation speed, torque, acceleration and deceleration of the horizontal main shaft, the output of the three-phase power (U V W) of the driver causes the main shaft to run, and the feedback of the encoder information facilitates the main shaft to obtain accurate rotation speed information.

[0112] The main shaft error acquisition part mainly includes three eddy current sensors for acquiring distance information of the standard rod in X, Y and Z directions, one rotation speed monitoring Hall sensor, one standard rod, and an error instrument required for sensor data conditioning. The error instrument outputs a digital signal which is directly connected with the data acquisition device and the control unit through a USB data line.

[0113] In addition, the whole horizontal spindle simulation loading experiment platform is built above the ground iron with a standard T-shaped slot. The overall weight of the ground iron is more than 2 tons. Such a high weight is to ensure that the spindle has good static support when running on it.

[0114] In this embodiment, the vertical spindle simulation loading experiment platform involved in the S1 step is shown in Figure 12 As shown in the figure, it includes a vertical spindle experiment platform, a pneumatic servo control unit, a vertical spindle state data acquisition unit, and a vertical spindle precision measurement unit. Before building the experiment platform, modal analysis is performed using the finite element method. Modal analysis can obtain the natural frequency and natural mode of the structure, which is the core of structural dynamic analysis. The natural frequency and natural mode can reflect the resonance and mode of the structure at a certain frequency. The specific structure of the vertical spindle is as follows:

[0115] The experiment platform is installed by four ground iron feet. After installation, the level is placed in the middle of the experiment platform. By adjusting the four ground iron feet, the entire platform is kept level. To prevent the spindle from damaging the ground and increase the static stiffness of the entire platform, a general platform is used as the installation base. The weight of the entire base reaches 1.5 tons.

[0116] The vertical spindle experiment platform includes a ground platform for placing the vertical spindle, a vertical spindle, a vertical spindle loading cylinder, and a vibration sensor II, a temperature sensor II, and a tension sensor II embedded in the vertical spindle. The vibration sensor II has three sensors installed at the front bearing, loading bearing, and ground platform of the vertical spindle. The temperature sensor II has three sensors, two of which are embedded in the front and rear bearings of the vertical spindle, and one is magnetically attracted to the outer ring of the loading bearing. The vertical spindle loading cylinder has three cylinders, two of which are used for radial loading, and one is used for axial loading. The two radial loading cylinders form a 90° angle. The servo cylinder can achieve a single-axis loading force of 4000N on the spindle. The tension sensor II is installed between the vertical spindle loading cylinder and the loading bearing, which can measure a range of up to 5000N. This forms a closed-loop control system for the relationship between the air pressure and force output controlled by the servo cylinder, ensuring the effectiveness between force control and air pressure output.

[0117] The pneumatic servo control unit mainly includes three pneumatic servo valves, three cylinders, and a reversing electromagnetic valve. Two pneumatic servo valves form a 90° angle for radial loading of the vertical spindle. The other pneumatic servo valve realizes the axial loading of the spindle. The reversing electromagnetic valve is connected to the pneumatic servo valve for realizing the reversing of the vertical spindle loading force, i.e., the loading process of tension and compression on the spindle.

[0118] As shown in Figure 12As shown, the vertical spindle state data acquisition unit includes a data acquisition case, and the spindle state is monitored and acquired through a sensor. The acquisition platform adopts a PXIe-1082 high-performance data acquisition case. The experimental platform mainly monitors vibration signals, current signals and temperature signals. Among them, there are three vibration sensors, which are installed at the loading bearing, the front bearing of the spindle and the position close to the ground platform. The vibration sensors are installed in a magnetic attraction mode. The experimental platform uses a 3-axis vibration sensor, and the directions of the three sensors are the same during installation. There are three temperature sensors, which measure the temperature at the loading bearing, the front support bearing of the spindle and the rear support bearing. The current sensor mainly acquires the current of the three-phase UVW output by the servo driver. The sensor uses a high-precision current transformer.

[0119] The vertical spindle precision measurement unit is used to measure the distance information of the standard rod at the tool position of the vertical spindle. It includes three eddy current sensors, which are connected to a precision conditioning unit. Two of the three eddy current sensors measure the radial X-axis and Y-axis distance information, and the other one measures the axial direction of the spindle. The rotational speed of the test rod is also measured. The eddy current sensor is connected to a corresponding precision conditioning unit. Finally, the corresponding calculation results are uploaded to the PC through a USB data line. The precision information of the spindle is finally displayed on the PC.

[0120] The numerical control machine tool spindle dynamic error measurement scheme provided in the embodiment has the following beneficial effects:

[0121] (1) Since the indirect measurement method only considers the time memory correlation characteristics and lacks multi-layer perception filtering capability, a classification model based on the time convolution characteristic sequence is constructed to realize higher precision indirect measurement of the spindle dynamic error and less time-consuming model training.

[0122] (2) The deep network method can be more effectively applied, and needs to be based on a complete spindle dynamic error data set. First, the installation attitude is diversified. The data set is not only from the horizontal installation spindle experimental platform, but also from the vertical installation spindle experimental platform. Second, the data set is from multiple wear states. Whether the spindle is horizontally or vertically installed, the phased wear experiment is carried out according to the existing loading spectrum model, and the corresponding dynamic error and state information is collected after each wear. Finally, the data set also has a certain amount. The entire data set is more than 3TB horizontally and more than 1.5TB vertically. A relatively complete spindle dynamic error data set is constructed, and a long-time spindle dynamic error tracking record is completed, laying a data foundation for online real-time measurement of the spindle dynamic error.

[0123] (3) By building a main shaft simulation loading experiment platform, not only the data of the main shaft under various speeds (1000, 2000, 3000, 4000 or even 5000) are collected, but also the cases under different axial force (0N, 500N, 1000N, 1500N, 2000N) and radial force (0N, 700N, 1400N, 2100N, 2800N) combinations are considered, a total of 180 combinations. This comprehensive data collection method can more comprehensively reflect the various working conditions of the main shaft in actual work, and provide rich and accurate data basis for subsequent model training and error evaluation. However, the prior art lacks similar comprehensive data collection research in the dynamic error measurement and evaluation of horizontal main shafts under different speeds.

[0124] (4) A deep time convolution network based on the dynamic error of the main shaft with an interval of 1 um is constructed. The network uses the input layer to construct the corresponding sequence relationship and performs sequence normalization processing, then connects 4 residual blocks, and then realizes data flattening through full connection, and finally completes category recognition through the Softmax layer. This model structure fully considers the time correlation characteristics of the main shaft error accumulated with time, can effectively process time series data, and thus realizes the indirect measurement and evaluation of the dynamic error of the main shaft, which is an effective model construction method for the dynamic error measurement of horizontal main shafts that is not available in the prior art.

[0125] (5) The structure design of the horizontal main shaft simulation loading experiment platform is detailed and reasonable, including the pneumatic loading hardware part, the horizontal main shaft experiment platform part, the collection and control part, and the main shaft dynamic error collection part. Each part has clear division of labor, such as the pneumatic loading hardware part realizes loading control in different directions through the air pump, air pressure servo valve and reversing relay; the horizontal main shaft experiment platform part realizes loading and data collection of the main shaft through the programmed horizontal main shaft, various sensors and loading mechanisms. This detailed structure design can accurately simulate the working state of the horizontal main shaft, which is helpful for accurate measurement and evaluation of its dynamic error, while the prior art may not be perfect in the design of the horizontal main shaft experiment platform.

[0126] (6) Cross-entropy is used as a quantitative calculation method of the output layer loss rate, and the loss rate of the corresponding classification output is calculated to improve the output layer identification ratio. This optimization measure helps to improve the accuracy of the model in classifying and identifying the dynamic error of the main shaft, so as to more accurately evaluate the dynamic error of the main shaft, which is not fully applied in the existing research on the dynamic error measurement and evaluation of horizontal main shafts.

[0127] Correspondingly, the embodiments of the present application also provide a computer program product, which comprises a computer program, and the computer program is executed to realize the method mentioned in any of the above embodiments.

[0128] Further, the embodiments of the present application also provide an electronic device, such as Figure 13 As shown in Fig. 13, the electronic device includes a processor 1301, a memory 1302, a computer program stored in the memory 1302 and executable by the processor 1301, and the processor 1301 implements the method mentioned in any of the above embodiments when executing the computer program.

[0129] Correspondingly, the embodiments of the present application also provide a computer storage medium, which stores a program, and the program is executed by a processor to implement the method in any of the above embodiments.

[0130] The embodiments of the present application can adopt the form of computer program products implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program codes. The computer usable storage media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, read only optical disc, read only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic disc or other magnetic storage device, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0131] For the device embodiment, since it basically corresponds to the method embodiment, the related part can refer to the part of the method embodiment. The device embodiment described above is only illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0132] The user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0133] It should be noted that, in this text, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0134] The above describes the method and device provided by the embodiments of the present application in detail. The principles and implementation modes of the present application are described by applying specific examples in this text. The above description of the embodiments is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed; in view of the above, the content of the present application should not be understood as a limitation of the present application.

Claims

1. A method of determining dynamic errors of a spindle of a numerically controlled machine tool, characterized in that, The method comprises: obtaining a vibration signal sequence for representing vibration conditions of a main shaft of a numerical control machine tool; inputting the vibration signal sequence into a pre-trained deep time convolution network to predict an error category to which a dynamic error of the main shaft of the numerical control machine tool belongs, wherein a dynamic error range covered by the main shaft of the numerical control machine tool in a working process is divided into a plurality of error categories, and each error category corresponds to a sub-range in the dynamic error range; the deep time convolution network is trained by a plurality of sets of sample data, the plurality of sets of sample data are constructed based on data collected by an experimental platform pre-built for simulating an actual working environment of the main shaft of the numerical control machine tool, and the experimental platform comprises a main shaft load simulation system, an experimental main shaft, a main shaft driving system, a control system, a dynamic error collection system, and a vibration signal collection system; the main shaft load simulation system comprises an air pump, an air pressure servo valve, an air cylinder, a reversing relay, a loading mechanism, and a tension sensor, the air pump is used to provide an air source, the air pressure servo valve is located between the air pump and the air cylinder and is used to adjust the size of air pressure output by the air pump to the air cylinder, the reversing relay is located between the air pump and the air cylinder and is used to adjust the flow direction of air pressure output by the air pump to the air cylinder to simulate the application of radial loading force and axial loading force to the experimental main shaft, the air cylinder is used to drive the loading mechanism to move so that the loading mechanism applies loading force to the experimental main shaft, and the tension sensor is arranged at the connection between the loading mechanism and the air cylinder and is used to detect current loading force information and feed back to the control system so that the control system adjusts the air pressure servo valve based on the current loading force information and user-set loading force information; the main shaft driving system is used to receive user-set rotating speed information from the control system and drive the experimental main shaft to rotate based on the rotating speed information; the vibration signal collection system comprises one or more sensors and is used to collect vibration signals of the experimental main shaft and send the vibration signals to the control system; the dynamic error collection system is used to collect dynamic errors of the experimental main shaft and send the dynamic errors to the control system; wherein the plurality of sets of sample data can be constructed based on data respectively collected by the vibration signal collection system and the dynamic error collection system under different running conditions of the experimental main shaft, wherein the different running conditions comprise combined conditions obtained by a plurality of rotating speeds and a plurality of loading forces, wherein the process of applying loading force to the experimental main shaft under each rotating speed comprises a constant force loading stage and a variable force loading stage, wherein a first loading force is used to apply constant force to the experimental main shaft in an early stage of the constant force loading stage, a second loading force is used to apply constant force to the experimental main shaft in a later stage of the constant force loading stage, and the first loading force is smaller than the second loading force.

2. The method of claim 1, wherein, The deep time convolution network comprises an input layer, a feature extraction layer, a full connection layer and an output layer, wherein the feature extraction layer comprises four residual blocks connected in series; the input layer is used to acquire the vibration signal sequence and input to the feature extraction layer; the feature extraction layer is used to extract features of the vibration signal sequence; the full connection layer is used to integrate the features output by the feature extraction layer; and the output layer is used to determine the error category corresponding to the vibration signal sequence based on the features output by the full connection layer. And / or Each set of sample data comprises a sample vibration signal sequence and a label corresponding to the sample vibration signal sequence, wherein the label is used to indicate the error category corresponding to the sample vibration signal sequence, and each set of sample data is determined based on the following manner: during the operation of the experimental spindle, the original vibration signal sequence collected by the vibration signal acquisition system in a target time period is acquired, and the original dynamic error sequence synchronously collected by the dynamic error acquisition system in the target time period is acquired; the sample vibration signal sequence is obtained based on the original vibration signal sequence, the mean of the original dynamic error sequence is determined, and the error category corresponding to the mean is determined as the label.

3. The method of claim 1, wherein, Each set of sample data comprises a sample vibration signal sequence and a label corresponding to the sample vibration signal sequence, wherein the label is used to indicate the error category corresponding to the sample vibration signal sequence, and the deep time convolution network is obtained based on the following manner: The sample vibration signal sequence corresponding to each set of sample data is input into a preset deep time convolution network, and the deep time convolution network outputs the prediction probability of each set of sample vibration signal sequence for each error category in the plurality of error categories; A target loss is constructed based on the prediction probability of each set of sample vibration signal sequence for each error category in the plurality of error categories, the actual probability of each set of sample vibration signal sequence for each error category in the plurality of error categories, and the weight value of each set of sample vibration signal sequence, and the network parameters of the deep time convolution network are adjusted based on the target loss to train the deep time convolution network; wherein the weight value of each set of sample vibration signal sequence is negatively correlated with the proportion of noise signals in the set of sample vibration signal sequence.

4. The method of claim 3, wherein, The target loss is determined by the following formula: wherein, loss is a target loss, N is a number of sample vibration signal sequences input into the deep time convolution network, K is a total number of error categories into which a dynamic error range covered by the spindle of the numerical control machine tool in a working process is divided, zi is a weight value of an i-th sample vibration signal sequence, a ij is a prediction probability of the i-th sample vibration signal sequence predicted by the deep time convolution network belonging to the j-th error category, y ij is an actual probability of the i-th sample vibration signal sequence belonging to the j-th error category.

5. The method of claim 1, wherein, The vibration signal sequence is obtained by normalizing the original vibration signal sequence collected by the numerical control machine tool spindle, wherein the normalization process is as follows: The original vibration signal sequence is divided into a plurality of vibration signal sequence blocks; The mean and variance of each vibration signal sequence block are determined, and the minimum mean and maximum variance are determined from the mean and variance corresponding to each of the plurality of vibration signal sequence blocks; For each original vibration signal in the original vibration signal sequence, the difference between the original vibration signal and the minimum mean, and the ratio of the maximum variance, are taken as the normalized vibration signal of the original vibration signal.

6. The method of claim 1, wherein, The dynamic error acquisition system comprises a standard bar arranged at a tool position of the experimental spindle, and used for assisting in measurement of dynamic error of the experimental spindle; and at least three eddy current sensors, wherein one eddy current sensor is arranged in each of X, Y and Z directions of the standard bar, and used for acquiring the dynamic error of the experimental spindle.

7. The method of claim 6, wherein, The experimental platform is a vertical spindle simulation loading experimental platform, the experimental spindle is a vertical spindle, the air cylinder includes three, two of which are arranged at an angle of 90°, used for realizing radial loading force applied to the vertical spindle, and the other air cylinder is used for realizing axial loading force applied to the vertical spindle; or The experimental platform is a horizontal spindle simulation loading experimental platform, the experimental spindle is a horizontal spindle, the air cylinder includes three, two of which are arranged at an angle of 45°, used for realizing radial loading force applied to the vertical spindle, and the other air cylinder is used for realizing axial loading force applied to the vertical spindle.

8. The method of claim 6, wherein, The experimental platform is a vertical spindle simulation loading experimental platform, the experimental spindle is a vertical spindle, the experimental platform further comprises a ground platform, used for supporting and fixing the vertical spindle, the vibration signal acquisition system comprises three vibration sensors, respectively installed at the front bearing of the vertical spindle, the loading mechanism and the ground platform; and / or The experimental platform is a horizontal spindle simulation loading experimental platform, the experimental spindle is a horizontal spindle, the horizontal spindle simulation loading experimental platform is built above a ground flat iron with a standard T-shaped groove, and the total weight of the ground flat iron is more than 2 tons.

9. A computer program product, characterised in that, The computer program product comprises a computer program, which, when executed by a processor, implements the method of any one of claims 1-8.

Images