A life detection method and system for bearing retainer for CNC machine tools

Abstract

The present invention provides a life detection method and system for a bearing retainer for a numerically controlled machine tool, and relates to the technical field of data processing. The method comprises: obtaining historical sensor data of the bearing retainer for the numerically controlled machine tool; screening the sensor data at each moment through a safety area recognition model based on a clustering algorithm, and removing abnormally distributed data; interpolating the missing data after removal; constructing a health index curve; normalizing the remaining service life of the bearing retainer as a remaining service life label value; recording historical sensor data and corresponding remaining service life label values ​​as training samples; training a remaining service life detection model based on a long short-term memory neural network through the training samples; obtaining real-time sensor data of the bearing retainer for the numerically controlled machine tool; and detecting the remaining service life of the bearing retainer for the numerically controlled machine tool through the remaining service life detection model based on the long short-term memory neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a life detection method and system for a bearing retainer for a numerically controlled machine tool. Background Art

[0002] The bearing cage is mainly used to support the rolling elements (such as balls or rollers) in the bearing to ensure that they run in the correct position. The design of the cage can prevent the rolling elements from contacting each other, reduce friction and wear, and extend the service life of the bearing. In CNC machine tools, the bearing cage plays a vital role. The design and function of the bearing cage directly affect the performance and service life of the bearing, and then affect the working efficiency and processing accuracy of the entire machine tool.

[0003] By predicting and detecting the life of the cage, maintenance can be arranged in advance to avoid equipment downtime and emergency repairs caused by unexpected bearing failures. This can not only reduce the high maintenance costs caused by sudden failures, but also reduce the impact of equipment downtime on production and improve production efficiency.

[0004] With the rapid development of technologies such as pattern recognition, machine vision, and deep learning, more and more modern technologies (such as support vector machines, random forests, etc.) are being applied to bearing cage life detection.

[0005] However, the current bearing cage life detection model requires a large amount of training data during the training process. In actual application, it is extremely difficult to accurately calibrate the life of the bearing cage, and it is mainly calibrated manually by experienced engineers. Manual calibration is time-consuming and labor-intensive. Due to the different experience levels of engineers, it may lead to inconsistent calibration standards and is easily affected by subjective factors, resulting in deviations in data labels, which in turn affects the training effect and prediction accuracy of the model. Summary of the invention

[0006] In order to solve the technical problems that the current manual calibration is time-consuming and labor-intensive, the calibration standards may be inconsistent due to the different experience levels of engineers, and it is easy to be affected by subjective factors, resulting in deviations in data labels, which in turn affects the training effect and prediction accuracy of the model, the present invention provides a life detection method and system for bearing retainers for CNC machine tools.

[0007] The technical solution provided by the embodiment of the present invention is as follows:

[0008] First aspect:

[0009] An embodiment of the present invention provides a life detection method for a bearing retainer for a CNC machine tool, comprising:

[0010] S1: Acquire historical sensor data of bearing cages for CNC machine tools;

[0011] S2: extracting data features of the historical sensor data;

[0012] S3: According to the data features of the historical sensor data, the sensor data at each moment is screened by a safety zone recognition model based on a clustering algorithm to remove abnormally distributed data;

[0013] S4: Impute missing data after removal;

[0014] S5: construct a health index curve based on the interpolated sensor data;

[0015] S6: normalizing the remaining service life of the bearing retainer according to the health index curve to obtain a remaining service life label value;

[0016] S7: Record historical sensor data and corresponding remaining service life label values ​​as training samples;

[0017] S8: training a remaining useful life detection model based on a long short-term memory neural network through the training samples;

[0018] S9: Acquire real-time sensor data of bearing cages for CNC machine tools;

[0019] S10: The remaining service life of the bearing retainer for CNC machine tools is detected by using the trained remaining service life detection model based on the long short-term memory neural network.

[0020] Second aspect:

[0021] An embodiment of the present invention provides a life detection system for a bearing retainer for a CNC machine tool, comprising:

[0022] processor;

[0023] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the life detection method for the bearing retainer for a CNC machine tool as described in the first aspect is implemented.

[0024] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0025] In the present invention, the health status changes of the bearing retainer are accurately tracked through the health index curve, and the remaining service life of the bearing retainer is normalized according to the health index curve as the remaining service life label value, which reduces the reliance on manual calibration, saves time and effort, reduces the influence of subjective factors, improves the accuracy of data label calibration, and improves the training effect and prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 A schematic flow chart of a life detection method for a bearing retainer for a CNC machine tool provided by an embodiment of the present invention;

[0028] Figure 2 A schematic structural diagram of a life detection system for a bearing retainer for a CNC machine tool provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0030] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0031] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0032] Reference Manual Attached Figure 1 , shows a flow chart of a life detection method for a bearing retainer for a CNC machine tool provided by an embodiment of the present invention.

[0033] The embodiment of the present invention provides a life detection method for a bearing cage for a CNC machine tool, which can be implemented by a life detection device for a bearing cage for a CNC machine tool, and the life detection device for a bearing cage for a CNC machine tool can be a terminal or a server. The processing flow of the life detection method for a bearing cage for a CNC machine tool can include the following steps:

[0034] S1: Acquire historical sensor data of bearing cages for CNC machine tools.

[0035] The sensor data may specifically be vibration sensor data, temperature sensor data, etc.

[0036] S2: Extract data features of historical sensor data.

[0037] The data features include time domain features and frequency domain features.

[0038] Furthermore, the time domain features include: mean, variance, standard deviation, kurtosis, root mean square value, kurtosis, etc.

[0039] Furthermore, frequency domain features include: spectrum centroid, spectrum bandwidth, spectrum peak, frequency center, total frequency energy, etc.

[0040] S3: According to the data characteristics of historical sensor data, the sensor data at each moment is screened through the safe area recognition model based on the clustering algorithm to remove abnormally distributed data.

[0041] In a possible implementation, S3 specifically includes sub-steps S301 to S305:

[0042] S301: Construct a feature data set based on the data features of historical sensor data:

[0043]

[0044] in, P represents the feature dataset, p ij Indicates i The moment j The feature value of feature dimension, , T Indicates the total duration, , n Indicates the total number of feature dimensions.

[0045] S302: Clustering the historical sensor data based on the K-Means clustering algorithm.

[0046] Specifically, the K-Means clustering algorithm is a classic unsupervised learning algorithm, which is often used to divide a data set into multiple independent clusters (Clusters) in order to discover patterns and groupings in the data. Its main idea is to assign data points to the nearest cluster center (Centroid) through iterative optimization, and continuously update the position of the cluster center until a convergence state is reached. The K-Means clustering algorithm is already a very mature prior art, and the present invention will not repeat it in detail.

[0047] S303: Calculate the safety boundary value of each category in each feature dimension:

[0048]

[0049] in, r kj Indicates k Category inj The safety boundary value under the feature dimension, Indicates j The average value of the feature values ​​of feature dimensions, T k Indicates k The total duration of data contained in the categories.

[0050] It should be noted that the safety boundary value is formulated using the triple standard deviation rule and is automatically calculated based on historical data. Each clustered category has a different safety area. This allows the method to be applied to different scenarios and application environments, and can flexibly adapt and identify normal operating conditions, slightly abnormal operating conditions, or extremely abnormal operating conditions.

[0051] S304: Determine the safety area of ​​each category according to the safety boundary value of each feature dimension:

[0052]

[0053] in, x j Indicates j The variable value of the dimension, c kj Indicates k The center of the cluster j The feature value of the feature dimension.

[0054] It should be noted that by calculating the cluster center through the clustering algorithm and establishing different safety boundaries for each category, it is possible to ensure targeted safety area definitions for different categories of data. This makes the method adaptable to data under various working conditions or operating states and has better generalization capabilities.

[0055] S305: Determine whether the sensor data at each moment is within the safe area. If so, keep it. Otherwise, remove it and remove the abnormally distributed data.

[0056] It should be noted that the presence of noise and abnormal data may have a negative impact on the model training process, causing the model to overfit or biased prediction results. By screening and removing these abnormal data, the data input to the model can be made more stable and consistent, thereby improving the training effect of the machine learning model and improving the accuracy of the prediction.

[0057] In the present invention, by identifying safe areas and removing abnormal data based on the K-means clustering algorithm, the quality of sensor data can be significantly improved, the impact of noise and outliers on model training can be reduced, and the stability and prediction accuracy of the model can be enhanced. The method has a high degree of automation, can dynamically adapt to a variety of working conditions and multi-dimensional data, reduces human intervention and subjective judgment, and is suitable for equipment health monitoring and life prediction in complex industrial scenarios.

[0058] S4: Impute missing data after removal.

[0059] In a possible implementation, S4 specifically includes: interpolating the missing data after removal by linear interpolation.

[0060] Among them, linear interpolation is a common numerical interpolation method, which is used to interpolate between known data points and estimate the value of unknown data points. It assumes that the data between two known points changes linearly, and estimates the intermediate value by constructing a straight line passing through the two known points. Linear interpolation is a very mature prior art, and the present invention will not repeat it in detail.

[0061] In the present invention, after removing the abnormal data, if interpolation is not performed, the data set will have gaps or blanks, especially in time series data, where continuity is very important. Through linear interpolation, these missing points can be smoothly filled to ensure the continuity of the data, which is beneficial to subsequent analysis and model training.

[0062] S5: Construct a health index curve based on the interpolated sensor data.

[0063] In a possible implementation, S5 specifically includes sub-steps S501 to S503:

[0064] S501: Determine the target intersection point of the line between the sensor data and the cluster center at each moment in the safety area:

[0065]

[0066] in, I i Indicates i The target intersection of the line between the sensor data at the moment and the cluster center in the safe area, I ij Indicates i The intersection of the line between the sensor data at time instant and the cluster center on the safe area is at j The value of the feature dimension, , n Indicates the total number of feature dimensions.

[0067] S502: Calculate the distance between the sensor data at each moment and the target intersection point, and use it as the health index of the sensor data at each moment:

[0068]

[0069] in, H i Indicates i The health index at all times, p ij Indicates i The moment j The feature value of the feature dimension.

[0070] It should be noted that by calculating the Euclidean distance between the intersection of the sensor data and the cluster center in the safe area, the degree to which the data deviates from the normal working state can be accurately quantified. The greater the deviation, the higher the health index, indicating that the health status of the device may be abnormal or deteriorated. This method can sensitively capture small changes in the operating status of the device.

[0071] S503: Construct a health index curve according to the health index of the sensor data at each moment.

[0072] S6: According to the health index curve, the remaining service life of the bearing cage is normalized as the remaining service life label value.

[0073] In a possible implementation, S6 specifically includes sub-steps S601 to S603:

[0074] S601: Determine a turning point in the health index curve according to the health index curve.

[0075] Optionally, the time when the device enters the degradation stage can be determined by presetting a threshold of the health index. The turning point is the time when the health index exceeds a certain threshold. Define a health index threshold, for example, you can choose an empirical value, such as when the health index reaches 70% or 80%, it is considered to enter the degradation stage.

[0076] Optionally, the health index curve usually shows a significant change in slope at the turning point, and the turning point can be identified by calculating the slope of the curve. The slope of the health index curve at each time point is calculated, and the time point at which the slope changes significantly is identified as the turning point.

[0077] S602: Divide the health index curve into a healthy stage and a deterioration stage according to the turning point.

[0078] S603: According to the turning point, the remaining service life of the bearing retainer is normalized:

[0079]

[0080] in, Y t express t The remaining useful life tag value at the moment, t Indicates the time, T h Indicates the turning point. T s Indicates the failure time point.

[0081] It should be noted that the health index curve is divided into a healthy stage and a deterioration stage, which makes the prediction of equipment life more targeted. In the healthy stage, the remaining service life of the equipment is maintained at a high level; in the deterioration stage, the equipment gradually approaches the failure point. Staged processing allows the model to more effectively distinguish between the normal operation state and the state of impending failure of the equipment, and improve the accuracy of the prediction.

[0082] In the present invention, the normalization process compresses the remaining useful life (RUL) label values ​​at different time points into a unified range (between 0 and 1), making it easier for the model to process and understand the data. The standardized labels reduce the impact of data at different time scales and improve the training efficiency and stability of the model.

[0083] S7: Record historical sensor data and corresponding remaining useful life label values ​​as training samples.

[0084] S8: Train the remaining useful life detection model based on the long short-term memory neural network through training samples.

[0085] Among them, Long Short-Term Memory (LSTM) is a special recurrent neural network (RNN). The core idea of ​​LSTM is to use forget gate, input gate and output gate to control the flow of information when processing time series. Each LSTM unit uses these gating mechanisms to filter out important information from the input data and the hidden state of the previous moment, and update the hidden state and cell state of the current moment.

[0086] In a possible implementation, the remaining useful life detection method of the remaining useful life detection model based on the long short-term memory neural network is specifically as follows:

[0087] Input sensor data to the LSTM neural network.

[0088] Through the long short-term memory neural network, the hidden state at each moment is extracted:

[0089]

[0090] in, X t express t The sensor data sequence at time, I t express t The activation output vector of the input gate at the moment, Sigmoid () represents the Sigmoid activation function, W XI Represents the weight matrix between the word segmentation sequence and the input gate, W HI represents the weight matrix between the hidden state and the input gate, b I represents the bias term of the input gate, F t express t Always forget the activation output vector of the gate, W XF Represents the weight matrix between the word segmentation sequence and the forget gate, W HF represents the weight matrix between the hidden state and the forget gate, b F represents the bias term of the forget gate, O t express t The activation output vector of the output gate at time t, W XO Represents the weight matrix between the word segmentation sequence and the output gate, W HO represents the weight matrix between the hidden state and the output gate, b O represents the bias term of the output gate, C t express t The activation output vector of the cell storage unit at time instant, express t The moment cell stores the temporary cell state of the cell, C t-1 express t -1 is the activation output vector of the cell storage unit, tanh() represents the tanh activation function, W XC Represents the weight matrix between the word segmentation sequence and the cell storage unit, W HC represents the weight matrix between the hidden state and the cell storage unit, b C represents the bias term of the cell storage unit, h t express t The hidden state of the moment, h t-1 expresst -1 time hidden state, t Indicates time.

[0091] Aggregate the hidden states at each moment and predict the remaining service life at the next moment:

[0092]

[0093] in, express t The remaining useful life prediction value at time +1, W Y represents the prediction weight matrix, b Y represents the prediction bias term, T Indicates the total duration.

[0094] In the present invention, the forget gate, input gate and output gate in the LSTM structure can effectively handle the long-term dependency problem in the time series. The operating status of the equipment often deteriorates over time. LSTM can remember the remote historical information through its design, which helps to more accurately predict the remaining life of the equipment in the future.

[0095] Furthermore, LSTM can be updated in real time in the continuously acquired data stream and continuously adjust the prediction of remaining life according to the current status. Therefore, after the sensor data is input, the model can continuously monitor the equipment status and make dynamic predictions, which helps to take maintenance measures in time and prevent sudden failures of equipment.

[0096] In a possible implementation, S8 specifically includes:

[0097] S801: Construct a mean square error loss function based on the remaining useful life prediction value and the remaining useful life label value.

[0098] Optionally, the mean square error loss function is specifically:

[0099]

[0100] in, L represents the mean square error loss function, i represents the parameters of the long short-term memory neural network, Y i Indicates i The remaining useful life label value of samples, Indicates i The remaining useful life prediction value of samples, N Represents the total number of samples.

[0101] S802: Based on the mean square error loss function, using the Adam optimizer, the model parameters of the remaining useful life detection model based on the long short-term memory neural network are optimized.

[0102] Specifically, the present invention uses an improved Adam optimizer to optimize the model parameters of the remaining useful life detection model based on the long short-term memory neural network:

[0103]

[0104] in, i t Indicates t The LSTM neural network parameters at the iteration, i t-1 Indicates t -1 LSTM neural network parameters at iteration 1, α represents the basic learning rate, Indicates t The second-order moment estimate after bias correction at the iteration, e represents a hyperparameter for numerical stability, Indicates t The first-order moment estimate after bias correction at the iteration, v t Indicates t The second-order moment estimate at the iteration, β 2 represents the second-order moment estimation attenuation coefficient, v t-1 Indicates t -The second moment estimate at iteration 1, g t Indicates t The gradient at iteration , represents the gradient operation, L represents the loss function, Indicates t At the iteration, the loss function performs gradient calculation on the convolutional neural network parameters. m t Indicates t The first-order moment estimate at the iteration, β 1 represents the first-order moment estimation attenuation coefficient, m t-1 Indicates t -1st moment estimate at iteration 1.

[0105] In the present invention, the Adam optimizer introduces an adaptive learning rate, designs an independent learning rate for each parameter, and dynamically adjusts the learning rate based on the estimated values ​​of the first-order moment (average) and second-order moment (variance) of the gradient. This adaptive mechanism ensures that the learning rate under different parameters or different time steps can be automatically adjusted, avoiding the problem of oscillation caused by too large a learning rate or slow convergence caused by too small a learning rate.

[0106] Furthermore, Adam combines the momentum mechanism to reduce the oscillation in gradient updates by taking a weighted average of past gradients (i.e., first-order moment estimation), especially to smooth parameter updates on steep loss surfaces.

[0107] At the same time, the Adam optimizer performs bias correction on the first-order and second-order moments, that is, at the beginning of the iteration, the moment estimate is corrected to offset the initial bias. Through this correction, Adam can more accurately evaluate the gradient in the early iterations and avoid inaccurate optimization caused by estimation bias.

[0108] In equipment failure prediction scenarios, historical sensor data may be insufficient or the cost of collecting data may be high. Generative adversarial networks can learn from the existing data distribution and generate synthetic data similar to real sensor data, thereby expanding the data set so that the LSTM model can learn on more diverse training data.

[0109] In a possible implementation, after S1, it further includes: based on the historical sensor data, generating similar sensor data by generating an adversarial network, and using the similar sensor data to train a remaining service life detection model based on a long short-term memory neural network.

[0110] Specifically, the generator and discriminator are constructed through convolutional neural networks, and some original convolutions in the generator are replaced with dilated convolutions to expand the receptive field of the model and better capture the sensor signal characteristics at multiple scales. The convolution kernel size of the dilated convolution is determined as follows:

[0111]

[0112] in, represents the height of the convolution kernel of the dilated convolution, f h represents the original convolution kernel height, represents the convolution kernel width of the dilated convolution, f w represents the original convolution kernel width, r Represents the expansion rate.

[0113] It should be noted that the dilated convolution can expand the receptive field without increasing the number of convolution kernel parameters by introducing intervals in the convolution kernel. This means that the model can obtain a wider range of input information in a single convolution operation, thereby analyzing data of a larger scale. Expanding the receptive field helps capture signal features over a long period of time or a large range, especially when processing time series data or sensor data, and can better capture patterns and changes in long-distance dependencies.

[0114] For the joint training of the generator and the discriminator, first fix the parameters of the generator, and train the discriminator separately by gradient descent with the goal of minimizing the training loss function of the discriminator. Then fix the parameters of the discriminator, and train the generator separately by gradient descent with the goal of minimizing the training loss function of the generator.

[0115] Optionally, the discriminator training loss function is specifically:

[0116]

[0117] in, L ( i D ) represents the discriminator training loss function, i D represents the discriminator parameters, z i represents the first i A real sample, The generator generates i pseudo samples, log represents the logarithmic function with base 10, m Represents the total number of samples.

[0118] Optionally, the generator training loss function is specifically:

[0119]

[0120] in, L ( i G ) represents the generator training loss function, i G represents the generator parameters, y i Indicates i The true label value of the sample, It means that the discriminator is i The discrimination result of samples, log represents the logarithmic function with base 10, m represents the total number of samples, α represents the first regularization coefficient, Represents the modulo operation.

[0121] In the present invention, similar sensor data is generated by generative adversarial networks (GANs), and these data are combined to train LSTM models, which can effectively make up for the lack of historical data, balance data sets, simulate rare or abnormal events, and improve model robustness and generalization capabilities. The combination of GAN and LSTM enhances the accuracy and stability of remaining service life prediction, especially under complex working conditions, and can help the model better cope with equipment degradation and fault prediction.

[0122] S9: Acquire real-time sensor data of bearing cages used in CNC machine tools.

[0123] S10: The remaining service life of the bearing retainer for CNC machine tools is detected by using the trained remaining service life detection model based on the long short-term memory neural network.

[0124] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0125] In the present invention, the health status changes of the bearing retainer are accurately tracked through the health index curve, and the remaining service life of the bearing retainer is normalized according to the health index curve as the remaining service life label value, which reduces the reliance on manual calibration, saves time and effort, reduces the influence of subjective factors, improves the accuracy of data label calibration, and improves the training effect and prediction accuracy of the model.

[0126] Reference Manual Attached Figure 2 , showing a structural schematic diagram of a life detection system for a bearing retainer for a CNC machine tool provided by the present invention.

[0127] The present invention also provides a life detection system 20 for a bearing retainer for a CNC machine tool, comprising:

[0128] Processor 201;

[0129] The memory 202 stores computer-readable instructions, and when the computer-readable instructions are executed by the processor 201, the life detection method for the bearing retainer for a CNC machine tool as described in the method embodiment is implemented.

[0130] The life detection system 20 for a bearing retainer for a CNC machine tool provided by the present invention can execute the above-mentioned life detection method for a bearing retainer for a CNC machine tool and achieve the same or similar technical effects. To avoid repetition, the present invention will not elaborate on it.

[0131] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0132] In the present invention, the health status changes of the bearing retainer are accurately tracked through the health index curve, and the remaining service life of the bearing retainer is normalized according to the health index curve as the remaining service life label value, which reduces the reliance on manual calibration, saves time and effort, reduces the influence of subjective factors, improves the accuracy of data label calibration, and improves the training effect and prediction accuracy of the model.

[0133] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0134] There are a few points to note:

[0135] (1) The drawings of the embodiments of the present invention only involve structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0136] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the film or region is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a film, a region or a substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0137] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0138] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for detecting the life of a bearing cage for a CNC machine tool, characterized in that: include: S1: Acquire historical sensor data of bearing cages for CNC machine tools; S2: extracting data features of the historical sensor data; S3: According to the data features of the historical sensor data, the sensor data at each moment is screened by a safety zone recognition model based on a clustering algorithm to remove abnormally distributed data; S4: Impute missing data after removal; S5: construct a health index curve based on the interpolated sensor data; S6: normalizing the remaining service life of the bearing retainer according to the health index curve to obtain a remaining service life label value; S7: Record historical sensor data and corresponding remaining service life label values ​​as training samples; S8: training a remaining useful life detection model based on a long short-term memory neural network through the training samples; S9: Acquire real-time sensor data of bearing cages for CNC machine tools; S10: The remaining service life of the bearing retainer for CNC machine tools is detected by using the trained remaining service life detection model based on the long short-term memory neural network.

2. The life detection method of the bearing retainer for a CNC machine tool according to claim 1, characterized in that: The data features include time domain features and frequency domain features.

3. The life detection method of the bearing retainer for a CNC machine tool according to claim 1, characterized in that: The S3 specifically includes: S301: Construct a feature data set according to the data features of the historical sensor data: in, P represents the feature dataset, p ij Indicates i The moment j The feature value of feature dimension, , T Indicates the total duration, , n Indicates the total number of feature dimensions; S302: Clustering the historical sensor data based on a K-Means clustering algorithm; S303: Calculate the safety boundary value of each category in each feature dimension: in, r kj Indicates k The categories in j The safety boundary value under the feature dimension, Indicates j The average value of the feature values ​​of feature dimensions, T k Indicates k The total duration of data contained in each category; S304: Determine the safety area of ​​each category according to the safety boundary value of each feature dimension: in, x j Indicates j The variable value of the dimension, c kj Indicates k The center of the cluster j The feature value of feature dimensions; S305: Determine whether the sensor data at each moment is within the safety area; if so, retain it; otherwise, delete it to remove abnormally distributed data.

4. The life detection method of the bearing retainer for a CNC machine tool according to claim 1, characterized in that: The S4 is specifically: The missing data after removal were interpolated by linear interpolation.

5. The life detection method of the bearing retainer for a CNC machine tool according to claim 3, characterized in that: The S5 specifically includes: S501: Determine the target intersection point of the line between the sensor data and the cluster center at each moment in the safety area: in, I i Indicates i The target intersection point of the line between the sensor data at the time instant and the cluster center on the safety area, I ij Indicates i The intersection of the line between the sensor data at time instant and the cluster center on the safety area is j The value of the feature dimension; S502: Calculate the distance between the sensor data at each moment and the target intersection point, and use it as the health index of the sensor data at each moment: in, H i Indicates i The health index at all times, p ij Indicates i The moment j The feature value of feature dimensions; S503: Construct a health index curve according to the health index of the sensor data at each moment.

6. The life detection method of the bearing retainer for a CNC machine tool according to claim 1, characterized in that: The S6 specifically includes: S601: Determine a turning point in the health index curve according to the health index curve; S602: Dividing the health index curve into a healthy stage and a deteriorating stage according to the turning point; S603: According to the turning point, the remaining service life of the bearing retainer is normalized: in, Y t express t The remaining service life tag value at the moment, t Indicates the time, T h Indicates the turning point. T s Indicates the failure time point.

7. The life detection method of the bearing retainer for a CNC machine tool according to claim 1, characterized in that: The remaining service life detection method of the remaining service life detection model based on the long short-term memory neural network specifically includes: Input sensor data into the LSTM neural network; Through the long short-term memory neural network, the hidden state at each moment is extracted; The hidden states at each moment are aggregated to predict the remaining service life at the next moment.

8. The life detection method of the bearing retainer for a CNC machine tool according to claim 1, characterized in that: The S8 specifically includes: S801: constructing a mean square error loss function according to the remaining useful life prediction value and the remaining useful life label value; S802: Based on the mean square error loss function, using the Adam optimizer, the model parameters of the remaining useful life detection model based on the long short-term memory neural network are optimized.

9. The life detection method of the bearing retainer for a CNC machine tool according to claim 1, characterized in that: After S1, it also includes: Based on the historical sensor data, similar sensor data is generated by generating an adversarial network, and the similar sensor data is used to train a remaining service life detection model based on a long short-term memory neural network.

10. A life detection system for a bearing cage for a CNC machine tool, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the life detection method for a bearing retainer for a CNC machine tool as claimed in any one of claims 1 to 9 is implemented.

Images