Machine tool spindle rotation precision online real-time prediction method and system

Abstract

The invention discloses an online real-time prediction method and system for the rotation precision of a machine tool spindle, and relates to the technical field of mechanical machine tool machining, and the method comprises the following steps: collecting vibration acceleration signals of the spindle under different working conditions through a vibration acceleration sensor installed at the shaft end of the spindle; synchronously acquiring rotation precision data of the main shaft through a main shaft error analyzer; feature extraction is carried out on the vibration acceleration signals under different working conditions, time domain features and frequency domain features of the vibration acceleration signals are used as input, corresponding rotation precision data are used as output, the deep learning regression model is trained, and a rotation precision prediction model is obtained; and performing feature extraction on the vibration acceleration signal acquired in real time, and inputting real-time time domain features and frequency domain features into the rotation precision prediction model to obtain predicted main shaft rotation precision. The regression prediction model is applied in the actual cutting process, the position of a cutter is not occupied, and online real-time prediction of the rotation precision of the machine tool spindle is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of machining of machine tools, and particularly relates to a method and system for online real-time prediction of the rotational accuracy of a machine tool spindle. Background Art

[0002] The high-end equipment manufacturing industry focuses on the production of advanced industrial equipment with high technological content and high added value. Driven by advanced technologies, it is the key to promoting industrial modernization and the forefront of innovation. Developing and strengthening the high-end equipment manufacturing industry is of great significance for enhancing industrial competitiveness and winning the initiative in future economic and technological competitions. In this field, numerically controlled machine tools play a crucial role. Their machining level is a reflection of a country's manufacturing level and economic strength, and is the foundation for ensuring the development of the manufacturing industry.

[0003] As the core functional and machining execution component of a numerically controlled machine tool, the performance of the spindle directly affects the level of high-grade numerically controlled machine tools. The rotational accuracy of the spindle is one of the important indicators for measuring the comprehensive performance of the spindle. Research indicates that 30% - 70% of the circular machining errors in precision turning are caused by rotational errors, and the higher the accuracy of the machine tool, the greater the proportion. Online measurement of the rotational accuracy of the spindle is of great significance for extending the accuracy retention of the machine tool.

[0004] The measurement methods for the rotational accuracy of the spindle include dial indicator measurement method, single-direction measurement method, two-direction measurement method, and measurement methods based on error separation technology, etc. These methods all require installing a standard ball or a standard bar at the tool fixture end, which occupies the tool position and causes the machine tool unable to complete normal machining. Therefore, the existing methods cannot measure the rotational accuracy of the spindle in the actual cutting process in real time. Summary of the Invention

[0005] Based on the above defects existing in the prior art, the present invention provides a method and system for online real-time prediction of the rotational accuracy of a machine tool spindle, which solves the problem that the existing methods cannot measure the rotational accuracy of the spindle in the actual cutting process in real time.

[0006] The present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for online real-time prediction of the rotational accuracy of a machine tool spindle, including the following steps:

[0008] Collect vibration acceleration signals of the spindle under different working conditions by using vibration acceleration sensors installed at the spindle end; simultaneously collect rotational accuracy data of the spindle synchronously through a spindle error analyzer;

[0009] Extract features from the vibration acceleration signals under different working conditions to obtain the time-domain features and frequency-domain features of the vibration acceleration signals;

[0010] Taking the time-domain characteristics and frequency-domain characteristics of the vibration acceleration signal as the input and the corresponding rotational accuracy data as the output, training the deep learning regression model to obtain the rotational accuracy prediction model;

[0011] Performing feature extraction on the vibration acceleration signal collected in real time to obtain the real-time time-domain characteristics and frequency-domain characteristics, and inputting the real-time time-domain characteristics and frequency-domain characteristics into the rotational accuracy prediction model to obtain the predicted spindle rotational accuracy.

[0012] Preferably, collecting the vibration acceleration signal of the spindle under different working conditions by using the vibration acceleration sensor installed at the spindle end includes the following steps:

[0013] Obtaining the offsets of the front bearing and the rear bearing of the machine tool spindle, and performing geometric analysis on the offsets of the front bearing and the rear bearing of the machine tool spindle and the rotational accuracy;

[0014] Determining the specific installation position of the sensor at the spindle end based on the geometric analysis result;

[0015] Installing the sensor at the specific installation position to collect the vibration acceleration signal of the spindle under different working conditions.

[0016] Preferably, performing geometric analysis on the offsets of the front bearing and the rear bearing of the machine tool spindle and the rotational accuracy includes the following steps:

[0017] When the rear bearing has no offset and the front bearing has an offset, the first offset at the front end of the spindle is as follows:

[0018]

[0019] In the formula, δ1 is the first offset, a is the overhang, L is the span between the supports of the front bearing and the rear bearing, and δ a is the offset of the front bearing;

[0020] When the front bearing has no offset and the rear bearing has an offset, the second offset at the front end of the spindle is as follows:

[0021]

[0022] In the formula, δ2 is the second offset, and δ b is the offset of the rear bearing;

[0023] Let δ a =δ b , then δ1>δ2, and the influence of the front bearing on the rotational accuracy of the spindle is greater than that of the rear bearing. Install the vibration acceleration sensor at the front spindle end.

[0024] The influence of the front bearing on the accuracy is greater than that of the rear bearing, and the sensor needs to be installed at the front bearing position.

[0025] Preferably, before extracting the characteristics of the vibration acceleration signals under different working conditions, it is necessary to preprocess the vibration acceleration signals under different working conditions by filtering out noise and electromagnetic interference.

[0026] Preferably, the time-domain characteristics include mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value, peak factor, skewness, kurtosis, impulse factor, shape factor, clearance factor, average absolute value, root mean square value, and zero crossing rate; the frequency-domain characteristics include signal energy, signal power, dominant frequency, spectral center, bandwidth, spectral kurtosis, and spectral skewness.

[0027] Preferably, the deep learning regression model includes 3 LSTM layers and 1 fully connected layer. The LSTM layers are used to extract the time-series characteristics of the features of the vibration signals to obtain high-dimensional feature vectors, and the fully connected layer is used to map the high-dimensional feature vectors to obtain the rotational accuracy.

[0028] Preferably, when training the deep learning regression model, the Adam optimizer is used to optimize the deep learning regression model, and the loss function is the mean squared error loss function.

[0029] Preferably, before training the deep learning regression model, it is necessary to round the measured rotational accuracy data.

[0030] In a second aspect, the present invention provides an on-line real-time prediction system for the rotational accuracy of a machine tool spindle, including:

[0031] An analysis module, configured to collect vibration acceleration signals of the spindle under different working conditions by using a vibration acceleration sensor installed at the spindle end; and simultaneously collect the rotational accuracy data of the spindle through a spindle error analyzer.

[0032] An extraction module, configured to extract the characteristics of the vibration acceleration signals under different working conditions to obtain the time-domain characteristics and frequency-domain characteristics of the vibration acceleration signals.

[0033] A training module, configured to use the time-domain characteristics and frequency-domain characteristics of the vibration acceleration signals as inputs and the corresponding rotational accuracy data as outputs to train a deep learning regression model to obtain a rotational accuracy prediction model.

[0034] A prediction module, configured to extract the characteristics of the vibration acceleration signals collected in real time to obtain the real-time time-domain characteristics and frequency-domain characteristics, and input the real-time time-domain characteristics and frequency-domain characteristics into the rotational accuracy prediction model to obtain the predicted rotational accuracy of the spindle.

[0035] Compared with the prior art, at least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0036] The present invention uses a vibration acceleration sensor installed at the end of the main shaft to measure the vibration acceleration signal at the end of the shaft reflecting the vibration of the main shaft, without occupying the tool position. At the same time, a high-precision main shaft error analyzer is used to synchronously measure the rotational accuracy of the main shaft, and signal processing methods are used to process the vibration acceleration signal to obtain time-domain characteristics and frequency-domain characteristics. Taking the time-domain characteristics and frequency-domain characteristics of the vibration acceleration signal as inputs and the corresponding rotational accuracy data as outputs, a deep learning regression model is trained to obtain a rotational accuracy prediction model. During the actual cutting process, applying this regression prediction model, the machine tool can complete normal machining and achieve online real-time prediction of the rotational accuracy of the machine tool spindle. Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0038] Figure 1 Schematic diagram of the influence of the bearing on the rotational accuracy of the main shaft of the present invention;

[0039] Among them, Figure 1 (a) of : Schematic diagram of the influence of the front bearing on the rotational accuracy of the main shaft, Figure 1 (b) of : Schematic diagram of the influence of the rear bearing on the rotational accuracy of the main shaft;

[0040] Figure 2 Schematic diagram of the placement position of the vibration acceleration sensor of the present invention;

[0041] Figure 3 Schematic diagram of the layout of the rotational accuracy acquisition sensors of the present invention;

[0042] Among them, Figure 3 (a) of : Schematic diagram of the layout of the rotational accuracy acquisition sensors, Figure 3 (b) of : Measured diagram of the rotational accuracy acquisition sensors;

[0043] Figure 4 Schematic diagram of the deep learning model of the present invention;

[0044] Figure 5 Schematic diagram of the predicted rotational accuracy result of the present invention;

[0045] Figure 6 Flowchart of an online real-time prediction method for the rotational accuracy of a machine tool spindle of the present invention. Detailed Embodiments

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] Explanation of the embodiment. This part is an explanatory embodiment that expands and explains the technical solution of the claim in order to enable those skilled in the art to fully understand how the present invention is specifically implemented.

[0048] The present invention discloses an on-line real-time prediction method for the rotational accuracy of a machine tool spindle. Referring to Figure 6 , the method of the present invention uses a vibration acceleration sensor installed at the spindle end to collect the vibration signals of the spindle system. After signal processing, the time-domain and frequency-domain characteristics of the vibration signals are extracted. The rotational accuracy of the spindle is synchronously collected by a high-precision spindle error analyzer, a data set of the vibration acceleration signal characteristics and the rotational accuracy under multiple working conditions is constructed, and a deep learning regression prediction model for the spindle rotational accuracy is established using the constructed data set. In the actual cutting process, this regression prediction model is applied to realize the on-line real-time prediction of the rotational accuracy of the machine tool spindle. The present invention provides a method for real-time predicting the rotational accuracy of a spindle based on deep learning during the actual cutting process, which specifically includes the following steps:

[0049] S1: Obtain the offsets of the front bearing and the rear bearing of the machine tool spindle, perform geometric analysis on the offsets of the front bearing and the rear bearing of the machine tool spindle and the rotational accuracy, and determine the installation position of the sensor based on the geometric analysis result.

[0050] The main factors affecting the rotational accuracy of the spindle are bearing clearance and thermal deformation. The wear of the bearing is the main reason for the degradation of the spindle return accuracy. The bearing offset refers to the deviation distance of the actual rotation axis of the bearing relative to the ideal rotation axis. The larger the offset, the lower the rotational accuracy of the bearing.

[0051] Establish the geometric relationship between the axis offsets of the front and rear bearings of the spindle and the rotational accuracy of the spindle. Considering that the vibration at the spindle end reflects the vibration caused by spindle imbalance and misalignment, referring to Figure 1 , through geometric analysis, determine the installation position of the sensor. The sensor is a vibration acceleration sensor.

[0052] In Figure 1 (a), it is assumed that the rear bearing has no offset and only the front bearing has an offset δ a , the offset at the front end of the spindle is:

[0053]

[0054] In Figure 1In (b), it is assumed that there is no offset in the front bearing and only an offset δ in the rear bearing. b The offset at the front end of the main shaft is:

[0055]

[0056] Assume δ a = δ b , that is, the offset of the front bearing is equal to the offset of the rear bearing. It can be seen from the geometric relationship that δ1 > δ2, which means that the rotational accuracy of the main shaft is more affected by the front bearing. In addition, installing a sensor at the front end of the main shaft can capture the vibrations caused by imbalance and misalignment.

[0057] S2: Install a sensor at the installation position and collect the vibration acceleration signals at this position under different working conditions through the sensor. At the same time, synchronously measure the rotational accuracy data of the main shaft through the main shaft error analyzer.

[0058] Place the vibration acceleration sensor directly above the radial direction of the main shaft end. The placement position is as Figure 2 shown.

[0059] Use the main shaft error analyzer to measure the rotational accuracy of the main shaft. Clamp a high-precision standard ball at the front end of the main shaft through the tool holder, and place capacitive displacement sensors in three mutually perpendicular directions of the standard ball X, Y, and Z. The placement position is as Figure 3 shown. Use the software of the main shaft error analyzer to analyze and evaluate the sensor data in the two radial channels of X and Y, and round the obtained data. That is, the minimum resolution of the rotational accuracy is 0.1 μm, and the radial rotational accuracy of the main shaft can be obtained.

[0060] S3: Extract the characteristics of the vibration signals under different working conditions to obtain the characteristics of the vibration signals; use the characteristics of the vibration signals as the input and the corresponding rotational accuracy data as the output to train the deep learning regression model to obtain the rotational accuracy prediction model.

[0061] Conduct signal analysis and processing on the collected vibration acceleration signals, and use a low-pass filter to filter out components such as electromagnetic interference. Extract the time-domain characteristics of the vibration acceleration after signal processing, including 15 characteristics such as mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value, peak factor, skewness, kurtosis, impulse factor, shape factor, clearance factor, average absolute value, root mean square value, and zero crossing rate. Extract the frequency-domain characteristics of the vibration acceleration after signal processing, including 7 characteristics such as signal energy, signal power, dominant frequency, spectral center, bandwidth, spectral kurtosis, and spectral skewness rate. Generate a dataset of the time-domain and frequency-domain characteristics of the vibration acceleration signals and the rotational accuracy under multiple working conditions.

[0062] Establish a deep learning regression model, such as Figure 4As shown. Select appropriate network models, network parameters, optimizers, loss functions, and evaluation methods to establish a deep learning regression model.

[0063] The deep learning regression model is a long short-term memory regression prediction model, including 3 LSTM layers and 1 fully connected layer.

[0064] The optimizer is the Adam optimizer, the loss function is the mean squared error loss function, and the evaluation methods are mean absolute error, root mean squared error, and coefficient of determination.

[0065] Hyperparameters: The number of neurons in the hidden layer is 200, the number of iterations is 500, the batch size is 32, and the learning rate is 0.001.

[0066] Divide the generated dataset into a training set, a validation set, and a test set according to a certain ratio. The division ratio of the training set, the validation set, and the test set is: 0.8:0.1:0.1. After normalization, use the extracted vibration acceleration signal features as input data, and the processed spindle rotational accuracy as the prediction label. Train and test the deep learning regression prediction network, adjust the network hyperparameters according to the evaluation results and the loss function graph, and finally select the model with the best prediction effect on the test set as the final prediction model. The prediction results of the regression model are as Figure 5 shown. The mean absolute error predicted by the model is 0.1178, the root mean squared error is 0.0235, and the coefficient of determination is 0.9729.

[0067] S4: Extract features from the real-time collected vibration signals to obtain real-time features, and input the real-time features into the rotational accuracy prediction model to obtain the predicted spindle rotational accuracy.

[0068] The present invention uses a vibration acceleration sensor to measure the shaft end vibration acceleration signal reflecting the spindle vibration, uses a high-precision spindle error analyzer to synchronously measure the spindle rotational accuracy, extracts features after preprocessing the data by signal processing methods, trains and tests a deep learning regression model, and applies the regression prediction model in the actual cutting process to realize the online real-time prediction of the spindle rotational accuracy of the machine tool.

[0069] The present invention only uses a vibration acceleration sensor installed at the spindle shaft end to collect vibration acceleration sensors, with low data acquisition cost; uses a high-precision spindle error analyzer to measure the spindle rotational accuracy, with high measurement rotational accuracy; and uses a deep learning regression model to realize online real-time measurement, with low calculation cost.

[0070] The present invention provides a method for online real-time prediction and estimation of the rotational accuracy of a machine tool spindle, which solves the limitations of existing measurement methods that can only be used in static and no-load states, and has a low calculation cost compared to other online measurement methods. This method has the advantages of convenient measurement, wide application range, and low economic cost, and is of great significance for the regulation of the rotational accuracy of the machine tool spindle and the extension of the machine tool accuracy retention.

[0071] The present invention provides a method for real-time predicting the rotational accuracy of a spindle based on deep learning during the actual cutting process, which has the advantages of convenient measurement, wide application range, and low economic cost. On the one hand, it solves the problems of difficult online measurement and high calculation cost of traditional spindle rotational accuracy. On the other hand, it is of great significance for the regulation of the rotational accuracy of the machine tool spindle and the extension of the machine tool accuracy retention.

[0072] Based on the same concept, the present invention also provides an online real-time prediction system for the rotational accuracy of a machine tool spindle, including an analysis module, a determination module, a collection module, a training module, and a prediction module.

[0073] The analysis module is used to collect the vibration acceleration signals of the spindle under different working conditions by using the vibration acceleration sensors installed at the spindle end; at the same time, the rotational accuracy data of the spindle is synchronously collected through a spindle error analyzer.

[0074] The extraction module is used to extract the features of the vibration acceleration signals under different working conditions to obtain the time-domain features and frequency-domain features of the vibration acceleration signals.

[0075] The training module is used to train a deep learning regression model with the time-domain features and frequency-domain features of the vibration acceleration signals as the input and the corresponding rotational accuracy data as the output to obtain a rotational accuracy prediction model.

[0076] The prediction module is used to extract the features of the real-time collected vibration acceleration signals to obtain the real-time time-domain features and frequency-domain features, and input the real-time time-domain features and frequency-domain features into the rotational accuracy prediction model to obtain the predicted rotational accuracy of the spindle.

[0077] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0078] Obviously, those skilled in the art can make various changes and deformations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and deformations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and deformations.

Claims

1. A method for online real-time prediction of machine tool spindle rotation accuracy, characterized in that: The following steps are involved: The vibration acceleration sensor installed at the end of the spindle is used to collect the vibration acceleration signal of the spindle under different working conditions; at the same time, the spindle rotation accuracy data is synchronously collected through the spindle error analyzer; Extract the features of the vibration acceleration signals under different working conditions to obtain the time domain features and frequency domain features of the vibration acceleration signals; The time domain characteristics and frequency domain characteristics of the vibration acceleration signal are used as input, and the corresponding rotation accuracy data is used as output. The deep learning regression model is trained to obtain the rotation accuracy prediction model. The real-time collected vibration acceleration signal is feature extracted to obtain real-time time domain features and frequency domain features, which are then input into a rotation accuracy prediction model to obtain the predicted spindle rotation accuracy.

2. The online real-time prediction method for the machine tool spindle rotation accuracy according to claim 1, characterized in that: The method of collecting the vibration acceleration signal of the main shaft under different working conditions by using a vibration acceleration sensor installed at the shaft end of the main shaft comprises the following steps: Obtain the offset of the front bearing and the rear bearing of the machine tool spindle, and perform geometric analysis on the offset and rotation accuracy of the front bearing and the rear bearing of the machine tool spindle; Determine the specific installation position of the sensor at the spindle end based on the geometric analysis results; Install the sensor at the specific installation position to collect the vibration acceleration signal of the main shaft under different working conditions.

3. The online real-time prediction method for the machine tool spindle rotation accuracy according to claim 2, characterized in that: The geometric analysis of the offset and rotation accuracy of the front bearing and the rear bearing of the machine tool spindle includes the following steps: When the rear bearing has no offset and the front bearing has an offset, the first offset of the front end of the spindle is as follows: Where δ1 is the first offset, a is the overhang, L is the span between the front bearing and the rear bearing support, δ a is the front bearing offset; When the front bearing has no offset and the rear bearing has offset, the second offset of the front end of the spindle is as follows: Where δ2 is the second offset, δ b is the rear bearing offset; Let δ a =δ b , then δ1>δ2, the rotation accuracy of the spindle is affected by the front bearing more than the rear bearing, and the vibration acceleration sensor is installed at the front shaft end.

4. The online real-time prediction method for the machine tool spindle rotation accuracy according to claim 1, characterized in that: Before extracting the features of the vibration acceleration signals under different working conditions, the vibration acceleration signals under different working conditions need to be preprocessed to remove noise and electromagnetic interference.

5. The method for online real-time prediction of the machine tool spindle rotation accuracy according to claim 1, characterized in that: The time domain features include mean, variance, standard deviation, maximum value, minimum value, peak-to-peak value, peak factor, skewness, kurtosis, impulse factor, shape factor, gap factor, average absolute value, root mean square value and zero crossing rate; the frequency domain features include signal energy, signal power, dominant frequency, spectrum center, bandwidth, spectrum kurtosis and spectrum skewness.

6. The online real-time prediction method for the machine tool spindle rotation accuracy according to claim 1, characterized in that: The deep learning regression model includes 3 LSTM layers and 1 fully connected layer. The LSTM layer is used to extract the time series features of the vibration signal to obtain a high-dimensional feature vector. The fully connected layer is used to map the high-dimensional feature vector to obtain the rotation accuracy.

7. The method for online real-time prediction of the machine tool spindle rotation accuracy according to claim 1, characterized in that: The deep learning regression model is trained and optimized by the Adam optimizer, and the loss function is the mean square error loss function.

8. The method for online real-time prediction of the machine tool spindle rotation accuracy according to claim 1, characterized in that: Before training the deep learning regression model, the measured rotation accuracy data needs to be rounded off.

9. An online real-time prediction system for machine tool spindle rotation accuracy, characterized in that: include: The analysis module is used to collect the vibration acceleration signal of the spindle under different working conditions by using the vibration acceleration sensor installed at the end of the spindle; at the same time, the rotation accuracy data of the spindle is synchronously collected through the spindle error analyzer; An extraction module is used to extract features of vibration acceleration signals under different working conditions to obtain time domain features and frequency domain features of vibration acceleration signals; A training module is used to train a deep learning regression model using the time domain features and frequency domain features of the vibration acceleration signal as input and the corresponding rotation accuracy data as output to obtain a rotation accuracy prediction model; The prediction module is used to extract features from the real-time collected vibration acceleration signal to obtain real-time time domain features and frequency domain features, and input the real-time time domain features and frequency domain features into the rotation accuracy prediction model to obtain the predicted spindle rotation accuracy.

Images