High-speed gear milling machine spindle box fault diagnosis method and system based on fault mechanism simulation and data fusion

Abstract

The invention provides a fault diagnosis method and system for a spindle box of a high-speed gear milling machine based on fault mechanism simulation and data fusion. The method mainly comprises the following steps: S1, calculating a tooth profile wear curve of a gear based on an Archard wear principle; s2, three-dimensional modeling software is used for building system models in different wear states, and simulation data sets under different wear degrees are obtained with the help of dynamics simulation software; s3, establishing a data acquisition system to obtain an experimental vibration data set under the actual working condition of the spindle box; s4, constructing a spindle box fault diagnosis model by using the analogue simulation data set obtained in the step S2; and S5, performing fault diagnosis on the experimental vibration data set obtained in the step S3 by using the spindle box fault diagnosis model constructed in the step S4, and identifying the health condition of the spindle box system. Important data support is provided for a main axle box fault diagnosis and health management system, and the diagnosis accuracy and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis of a gear milling machine spindle box system, and in particular to a gear milling machine spindle box fault diagnosis method and system based on fault mechanism simulation and data fusion. Background Art

[0002] High-speed gear milling machines are highly efficient profiling gear processing equipment used in large gears for wind power, shield tunneling, and marine applications. Their spindle box is the core component of the entire transmission system. Its primary function is to transmit the motor's output power through a four-stage gear reduction to the cutting tool for high-speed cutting, removing the blank's machining allowance. Its dynamic characteristics directly affect the machining accuracy and stability of the cutterhead. However, the spindle box's long-term operation at high speeds and heavy loads increases the gear failure rate, leading to a significant increase in gear pair transmission errors. This anomaly is transmitted to the cutterhead through the drive chain, manifesting as an abnormally high amplitude of cutterhead vibration accompanied by an accumulation of energy in the characteristic frequency components. This type of vibration not only accelerates tool wear but also forms periodic chatter marks on the workpiece surface, severely reducing machining quality.

[0003] Existing fault diagnosis methods include data-driven and physical models. The former relies on real-time data collected by sensors or monitoring equipment. It uses a large amount of historical data and fault samples to train algorithms, extract fault characteristics, and predict fault types. However, in-service machine tools cannot conduct sufficient experiments to collect the required historical data and fault samples. The latter uses mathematical models based on the structure and operating principle of the spindle box to obtain simulation data for fault diagnosis. However, mathematical models cannot simulate the actual operating status and fault conditions of the spindle box system.

[0004] Therefore, establishing a mapping relationship between gear fault status and cutterhead vibration characteristics becomes a key challenge. It is necessary to develop a cross-scale diagnostic model that can synchronously capture the micromorphological changes and macroscopic vibration responses of the gear to solve the problem of insufficient sensitivity of traditional methods in early fault detection and provide reliable judgment criteria for the health management system of the gear milling machine. Summary of the Invention

[0005] In response to the above-mentioned problems, the present invention proposes a method and system for diagnosing spindle box faults of high-speed gear milling machines based on fault mechanism simulation and data fusion. First, by analyzing the gear wear failure mechanism, a simulation physical model is constructed, and simulation data sets under different wear degrees are obtained through simulation. Secondly, a spindle box data acquisition system is built to obtain an experimental vibration data set of the spindle box under actual working conditions. Then, the simulation data set is used to construct a spindle box system fault diagnosis model. Finally, the trained spindle box system fault diagnosis model is used to perform fault judgment on the experimental vibration data set to identify the health status of the spindle box system.

[0006] The present invention is achieved through the following technologies:

[0007] A method for diagnosing a spindle box fault of a high-speed gear milling machine based on fault mechanism simulation and data fusion comprises the following steps:

[0008] Step S1: Calculate the gear wear curve based on the Archard wear calculation formula. Use a nonlinear function to further divide the gear wear stages, namely the running-in stage, stable wear stage, and severe wear stage, into twelve intervals, A to L, and calculate the gear tooth profile wear curve corresponding to each interval.

[0009] Step S2: Use 3D modeling software to build a spindle box system model under different wear states, and use dynamic simulation software to simulate the influence of gears under different wear conditions on the cutter head vibration, and obtain simulation data sets under different gear wear degrees;

[0010] Step S3: Build a data acquisition system for the milling machine spindle box, install vibration acceleration sensors on the bearing seats of the first, third, and fifth axes to capture the vibration signals of the spindle box under actual working conditions and obtain the corresponding experimental vibration data set.

[0011] Step S4: constructing a headstock system fault diagnosis model using the simulation data set obtained in step S2;

[0012] Step S5: Use the spindle box system fault diagnosis model constructed in step S4 to perform fault diagnosis on the experimental vibration data set obtained in step S3 to identify the health status of the spindle box system.

[0013] In step S1, the specific steps are as follows:

[0014] Step S11: Considering that gears are affected by factors such as load changes, lubrication conditions, and operating environment during operation, the gear wear process has complex nonlinear characteristics. Therefore, in order to accurately model the gear wear process and accurately predict gear wear, the gear wear curve must first be calculated using the Archard wear model;

[0015] Wear calculation formula:

[0016] Where V is the wear volume, S is the relative sliding distance, K is the wear coefficient, W is the normal load at the contact point, and H is the surface hardness of the wear surface;

[0017] Step S12: According to the relevant provisions of JB / T 5664-2007 Failure Criteria for Heavy-Duty Gears, for equipment with high safety requirements, when the sum of the wear on both sides of the tooth root of a certain gear ΔS (mm) and the gear normal module m n(mm) percentage value M (M = ΔS / m n ) exceeds 15%, the gear is considered to have failed. Based on this criterion, in order to more accurately evaluate the wear state and evolution process of the gear, gear wear is divided into three stages: when M is between 0-3%, it is the running-in stage; when M is between 3%-8%, it is the stable wear stage; when M is between 8%-15%, it is the severe wear stage; when M exceeds 15%, the gear is considered to have failed;

[0018] Step S13: Next, a nonlinear function is used to further divide the gear wear phases of the running-in, stable, and severe wear stages into twelve intervals, A to L, and the gear tooth profile wear curve corresponding to each interval is calculated. The nonlinear time division method is used to ensure that the time intervals during the running-in and severe wear stages are short, while the time interval during the stable wear stage is long. This improves the accuracy of wear changes in a short period of time and adapts to the changing characteristics of the wear rate, thereby improving overall calculation efficiency and prediction accuracy.

[0019] Nonlinear function expression:

[0020] Where: t i is the end time of the ith interval, T is the total time, N is the number of intervals, b is the exponential base, which is used to control the growth rate; c is the intermediate node coefficient.

[0021] In step S2, the specific steps are as follows:

[0022] Step S21: Use 3D modeling software to construct a 3D geometric model of the spindle box system, including the gear set, bearings, and related supporting structures inside the spindle box, and construct a corresponding fault gear model based on the tooth profile wear curve;

[0023] Step S22: importing the spindle box system model into multi-body dynamics simulation software, applying corresponding constraints to the simulation model according to actual conditions, and obtaining simulation data sets under different gear wear conditions;

[0024] In step S3, the specific steps are as follows:

[0025] Build a data acquisition system for the spindle box of the gear milling machine, select appropriate vibration acceleration sensors and installation methods according to measurement requirements, and install the vibration acceleration sensors on the bearing seats of the first, third, and fifth axes of the spindle box using magnetic attraction.

[0026] In step S4, the specific steps are as follows:

[0027] Step S41: Based on the temporal characteristics of gear wear faults, a fault diagnosis model is established that integrates adaptive wavelet transform, self-attention long short-term memory network (AWT-SA-LSTM) and dynamic gated convolutional neural network (DG-CNN); the model includes two convolutional layers, two pooling layers, two LSTM layers, one fully connected layer and a softmax layer;

[0028] Step S42: performing data normalization on the simulation data set and the spindle box experimental vibration data set, dividing the normalized data set into a training set and a validation set in a ratio of 7:3, and using the spindle box experimental vibration data set as the test set;

[0029] Step S43: Perform adaptive wavelet transform on the pre-processed data set to obtain time-frequency features at different scales. The adaptive wavelet transform formula is:

[0030] Where W x (a * ,b) is the wavelet transform coefficient, is the adaptive wavelet basis function, a * is the optimal scaling factor;

[0031] Step S44: Use the time-frequency features after adaptive wavelet transformation as input and use the dynamic gate control convolutional neural network (DG-CNN) to extract features. The calculation formula of the dynamic gate control mechanism is:

[0032] Where, is the output feature of the i-th neuron in layer l, g(f(·)) is the dynamic gating function; is the i-th convolution kernel of the l-th layer, is the i-th bias term of the l-th layer; is the gate weight, σ(·) is the activation function Sigmoid, which limits the output to (0,1), and W g is the gating weight matrix, b g is the gate bias term;

[0033] Step S45: Input the features extracted by DG-CNN into the self-attention long short-term memory network (SA-LSTM) for temporal modeling. The attention calculation formula is:

[0034] Where Q, K, and V are query, key, and value matrices, respectively;

[0035] Step S46: Use the fully connected layer and Softmax classifier to identify the gear wear state, complete the model training phase, and accurately classify the twelve types of gear wear conditions. The fully connected layer dimensionality reduction calculation formula is: y = Softmax (W fc ·h T +b fc );

[0036] Where W fc is the weight matrix of the fully connected layer, b fc is the bias term of the fully connected layer;

[0037] Step S47: Use the validation set to evaluate and validate the model, calculate performance indicators: accuracy, precision and build a confusion matrix, and further optimize the fault diagnosis model.

[0038] In step S5, the specific steps are as follows:

[0039] The experimental vibration data set collected in step S3 is input into the constructed headstock system fault diagnosis model to perform fault diagnosis and identify the health status of the headstock system.

[0040] A high-speed gear milling machine spindle box fault diagnosis system based on fault mechanism simulation and data fusion, including: Simulation system: used to simulate the simulation data when the spindle box system fails; Data acquisition system: used to collect spindle box experimental vibration data; Fault diagnosis system: used to perform normalization processing, feature extraction, and state identification on simulated vibration data to build a fault diagnosis model; Condition identification system: used to identify the actual health condition of the spindle box.

[0041] The present invention has the following benefits:

[0042] The present invention proposes a method and system for diagnosing the spindle box fault of a high-speed gear milling machine based on fault mechanism simulation and data fusion, which comprehensively considers the two fault diagnosis methods of physical model and data drive, makes up for the limitations of a single method, and improves the accuracy and adaptability of fault identification. Specifically, the wear amount of the gear is calculated by the Archard wear model, and a simulation data set under different wear states is constructed. The adaptive wavelet transform (AWT) is combined to improve the time-frequency resolution of the vibration signal, and the dynamic gated convolutional neural network (DG-CNN) is used to extract deep local features. Then, the self-attention long short-term memory network (SA-LSTM) is combined to learn the temporal correlation information, thereby constructing a gear wear fault diagnosis model. Finally, the actual operation of the spindle box is judged to identify its health status. In general, the present invention makes up for the shortcoming of the small number of spindle box fault samples by combining the two fault diagnosis methods of physical model and data drive, and provides data support for the construction of a spindle box health management system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of the fault diagnosis method of the present invention.

[0044] Figure 2 Schematic diagram of wear variation over time according to the present invention.

[0045] Figure 3 Schematic diagram of gear tooth profile wear of the present invention.

[0046] Figure 4 It is a schematic diagram of the spindle box structure of the present invention.

[0047] Among them: 1. Spindle box motor; 2. Pulley drive; 3. First-stage right driving wheel; 4. Second-stage right driving wheel; 5. Third-stage right driving wheel; 6. Fourth-stage right driving wheel; 7. First-stage left driving wheel; 8. Second-stage left driving wheel; 9. Third-stage left driving wheel; 10. Fourth-stage left driving wheel; 11. Cutter disc.

[0048] Figure 5 This is a comparison diagram of the spindle box simulation signal and the actual vibration signal of the present invention.

[0049] Figure 6 This is a schematic diagram of the data acquisition module of the present invention.

[0050] Figure 7 It is the algorithm flow chart of the present invention. DETAILED DESCRIPTION

[0051] In order to make the present invention clearer, the present invention will be described more completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0052] Reference Attachment Figure 1The present invention provides a method for diagnosing a high-speed gear milling machine spindle box fault based on fault mechanism simulation and data fusion, comprising the following steps:

[0053] Step S1: Calculate the gear wear curve based on the Archard wear calculation formula. Use a nonlinear function to further divide the gear wear stages of the running-in phase, stable wear phase, and severe wear phase into twelve intervals, A to L, and calculate the gear tooth profile wear curve corresponding to each interval.

[0054] Step S2: Use 3D modeling software to build a spindle box system model under different wear states, and use dynamic simulation software to simulate the impact of different gear wear degrees on cutter head vibration, and obtain simulation data sets under different gear wear degrees.

[0055] Step S3: Build a data acquisition system for the milling machine spindle box, install vibration acceleration sensors on the bearing seats of the first, third, and fifth axes to capture the vibration signals of the spindle box under actual working conditions and obtain the corresponding experimental vibration data set.

[0056] Step S4: construct a spindle box system fault diagnosis model using the simulation data set obtained in step S2.

[0057] Step S5: Use the spindle box system fault diagnosis model constructed in step S4 to perform fault diagnosis on the experimental vibration data set obtained in step S3 to identify the health status of the spindle box system.

[0058] Reference Attachment Figure 2 and attached Figure 3 , step S1 specifically includes:

[0059] (1) Considering that the gear is affected by factors such as load changes, lubrication conditions, and working environment during operation, the gear wear process has complex nonlinear characteristics. Therefore, in order to accurately model the gear wear process and accurately predict gear wear, the gear wear curve must first be calculated using the Archard wear model.

[0060] Wear calculation formula:

[0061] Where V is the wear volume, S is the relative sliding distance, K is the wear coefficient, W is the normal load at the contact point, and H is the surface hardness of the worn surface.

[0062] (2) According to the relevant provisions of "JB / T 5664-2007 Failure Criteria for Heavy-Duty Gears", for equipment with high safety requirements, when the sum of the wear on both sides of the tooth root of a certain gear ΔS (mm) is equal to the gear normal module m n (mm) percentage value M (M = ΔS / m n) exceeds 15%, the gear is considered to have failed. Based on this criterion, to more accurately assess the gear wear state and its evolution process, gear wear is divided into three stages: when the ratio M is within the range of 0-3%, the gear is in the running-in stage; when the ratio M is within the range of 3%-8%, the gear is in the stable wear stage; when the ratio M is within the range of 8%-15%, the gear enters the severe wear stage; when the ratio M exceeds 15%, the gear wear has reached the failure standard and should be determined as gear failure.

[0063] (3) Secondly, a nonlinear function is used to further divide the gear wear phase of the running-in stage, the stable wear stage, and the severe wear stage into twelve intervals from A to L, and the gear tooth profile wear curve corresponding to each interval is calculated. The nonlinear time division method is used to make the time intervals in the running-in stage and the severe wear stage smaller, while the time interval in the stable wear stage larger, so as to improve the accuracy of wear changes in a short period of time and adapt to the changing characteristics of wear rate, thereby improving the overall calculation efficiency and prediction accuracy.

[0064] Nonlinear function expression:

[0065] Where: t i is the end time of the ith interval, T is the total time, N is the number of intervals, b is the exponential base, which is used to control the growth rate; c is the intermediate node coefficient.

[0066] Reference Attachment Figure 4 and attached Figure 5 , step S2 specifically includes:

[0067] (1) The schematic diagram of the spindle box structure involved in the present invention is shown in the attached figure. Figure 4 As shown, the working principle of the spindle box is that the output speed and torque of the spindle box motor 1 are input into the spindle box through the pulley transmission 2, and finally output the speed and torque to the cutter head through the four-stage gear transmission to process the workpiece.

[0068] (2) Use 3D modeling software to build a 3D geometric model of the spindle box system, including the gear set, bearings and related support structures inside the spindle box, and Figure 3 The corresponding gear wear model is constructed based on the tooth profile wear curve.

[0069] (3) The gear wear models under N conditions are named WearA to WearL, representing N intervals of wear from zero to severe wear.

[0070] (4) Import the spindle box system model into the multi-body dynamics simulation software, impose corresponding constraints on the simulation model according to the actual situation, set the simulation time to 10s, and the number of simulation steps to 20k steps.

[0071] (5) Export the simulated vibration data and name the data sets from WearA to WearL.

[0072] Reference Attachment Figure 6 , step S3 specifically includes:

[0073] (1) Build a data acquisition system for the milling machine spindle box, select appropriate vibration acceleration sensors and installation methods according to measurement requirements, and install the vibration acceleration sensors on the bearing seats of the first, third, and fifth axes of the spindle box using magnetic attraction.

[0074] (2) Set the sampling frequency of the vibration signal to 2kHz and the data storage time to 5 minutes.

[0075] Reference Attachment Figure 7 , step S4 specifically includes:

[0076] (1) Based on the temporal characteristics of gear wear faults, a fault diagnosis model is established that integrates adaptive wavelet transform, self-attention long short-term memory network (AWT-SA-LSTM) and dynamic gated convolutional neural network (DG-CNN). The model consists of two convolutional layers, two pooling layers, two LSTM layers, one fully connected layer and a softmax layer.

[0077] (2) Data normalization was performed on the simulation dataset and the actual spindle box dataset. The normalized dataset was divided into a training set and a validation set in a ratio of 7:3. The spindle box experimental vibration dataset was used as the test set. The normalized data input dimension is a single-channel time series, expressed as N × 20,000 × 1, where N is the total number of samples and 20,000 is the single-sample time step.

[0078] (3) Using Morlet wavelet basis function As the initial basis function, the scale factor a is dynamically optimized by the gradient descent algorithm. * , the search range is limited to a∈[1,64], the number of iterations is 100, 64 scale time-frequency feature matrices are generated, and the output dimension is N×20000×64. The adaptive wavelet transform formula is:

[0079] Where W x (a * ,b) is the wavelet transform coefficient, is the optimized adaptive wavelet basis function.

[0080] (4) The time-frequency features after adaptive wavelet transformation are used as input and feature extraction is performed using the dynamic gate control convolutional neural network (DG-CNN). The calculation formula of the dynamic gate control mechanism is:

[0081] Where, is the output feature of the i-th neuron in layer l, g(f(·)) is the dynamic gating function; is the i-th convolution kernel of the l-th layer, is the i-th bias term of the l-th layer; is the gate weight, σ(·) is the activation function Sigmoid, which limits the output to (0,1), and W g is the gating weight matrix, b g is the gate bias term.

[0082] Specific parameters include: the neural network consists of two convolutional layers and two pooling layers. The convolution kernel size of the first convolutional layer is 1×3, the number of convolution kernels is 128, the step size is 1, the receiving time-frequency feature input is 20000×64, and the output is a 128-channel feature map with a dimension of 20000×128; the pooling window size of the first pooling layer is 1×2, the step size is 2, the receiving dimension is 20000×128, and the output dimension is 10000×128; the convolution kernel size of the second convolutional layer is 1×3, the number of convolution kernels is 256, the step size is 1, the receiving dimension is 10000×256, and the output is a 256-channel feature map with a dimension of 10000×256; the second pooling layer outputs 5000×256 after the same pooling operation.

[0083] (5) Input the features extracted by DG-CNN into the self-attention long short-term memory network (SA-LSTM) for temporal modeling. The attention calculation formula is:

[0084] Where Q, K, and V are query, key, and value matrices, respectively.

[0085] Specific parameters include: the self-attention long short-term memory network consists of two LSTM layers. The first LSTM layer receives a dimension of 5000×256, has 128 hidden units, outputs the hidden state of all time steps, and has a dimension of 5000×128. Dropout=0.2 is introduced to prevent overfitting; the second LSTM layer receives a dimension of 5000×128, has 64 hidden units, and has a final output dimension of 5000×64.

[0086] (6) Use the fully connected layer and Softmax classifier to identify the gear wear state, complete the model training phase, and accurately classify the N types of gear wear conditions. The fully connected layer dimensionality reduction calculation formula is: y=Softmax(W fc ·h T +b fc )

[0087] Where Wfc is the weight matrix of the fully connected layer, b fc is the bias term of the fully connected layer.

[0088] Specific parameters include: the receiving dimension of the fully connected layer is 5000×64, and the output dimension is N (consistent with the number of fault categories, that is, the number of intervals N).

[0089] (7) Use the validation set to evaluate and verify the model, calculate performance indicators (such as accuracy, precision, etc.) and construct the confusion matrix, and further optimize the fault diagnosis model.

[0090] Reference Attachment Figure 5 , step S5 specifically includes:

[0091] The spindle box experimental vibration data set is input into the constructed spindle box system fault diagnosis model to perform fault diagnosis and identify the health status of the spindle box system.

[0092] A high-speed gear milling machine spindle box fault diagnosis system based on fault mechanism simulation and data fusion, including: Simulation system: used to simulate the simulation data when the spindle box system fails; Data acquisition system: used to collect spindle box experimental vibration data; Fault diagnosis system: used to perform normalization processing, feature extraction, and state identification on simulated vibration data to build a fault diagnosis model; Condition identification system: used to identify the actual health condition of the spindle box.

[0093] The damage to the spindle box has multi-source characteristics, and its failure forms include not only gear wear, but also gear system failure modes such as gear breakage, tooth surface pitting, tooth root cracks, and bearing component failures such as bearing pitting and bearing cracks. In addition, assembly process defects may also cause system abnormalities. Although the current simulation model has not yet covered all failure modes, it is also possible to obtain its fault vibration data set by constructing a three-dimensional model of these faults and performing simulations. After normalization and adaptive wavelet transformation, the data of these faults can also be used as input and are applicable to the fault diagnosis model and classification identification described in the present invention. Therefore, the method of the present invention is not only suitable for the diagnosis of gear wear faults, but can also be expanded to apply to the diagnosis of other fault types, and has wide applicability in the health monitoring and predictive maintenance of rotating machinery and gear transmission systems.

[0094] It will be easily understood by those skilled in the art that the above content is merely an embodiment of the present invention and cannot be used to limit the scope of protection of the present invention. Any modifications, equivalent replacements and improvements made on the basis of the technical solution using the technical ideas of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for diagnosing spindle box faults of high-speed gear milling machines based on fault mechanism simulation and data fusion, characterized in that: The following steps are involved: S1: Calculate the gear wear curve based on the Archard wear calculation formula; use a nonlinear function to further divide the gear wear stage, stable wear stage, and severe wear stage into twelve intervals from A to L, and calculate the gear tooth profile wear curve corresponding to each interval; S2: Use 3D modeling software to build a spindle box system model under different wear conditions, and use dynamic simulation software to simulate the impact of different gear wear degrees on cutter head vibration, and obtain simulation data sets under different gear wear degrees; S3: Build a data acquisition system for the milling machine's spindle box. Install vibration accelerometers on the bearing seats of the first, third, and fifth axes to capture the vibration signals of the spindle box under actual working conditions and obtain the corresponding experimental vibration data set. S4: constructing a headstock system fault diagnosis model using the simulation data set obtained in step S2; S5: Using the headstock system fault diagnosis model constructed in step S4, fault diagnosis is performed on the experimental vibration data set obtained in step S3 to identify the health status of the headstock system.

2. A high-speed gear milling machine spindle box fault diagnosis method based on fault mechanism simulation and data fusion according to claim 1, characterized in that: The step S1 of obtaining the gear wear curve of the gear mainly includes the following steps: S11: Considering that gears are affected by factors such as load changes, lubrication conditions, and operating environment during operation, the gear wear process has complex nonlinear characteristics. Therefore, in order to accurately model and predict the gear wear process, the Archard wear model must first be used to calculate the gear wear curve. Wear calculation formula: Where V is the wear volume, S is the relative sliding distance, K is the wear coefficient, W is the normal load at the contact point, and H is the surface hardness of the wear surface; S12: According to the relevant provisions of "JB / T 5664-2007 Heavy Duty Gear Failure Criteria", for equipment with high safety requirements, when the sum of the wear on both sides of the tooth root of a gear ΔS (mm) and the gear normal module m n (mm) percentage value M (M = ΔS / m n ) exceeds 15%, the gear is considered to have failed. Based on this criterion, in order to more accurately evaluate the wear state and evolution process of the gear, gear wear is divided into three stages: when M is between 0-3%, it is the running-in stage; when M is between 3%-8%, it is the stable wear stage; when M is between 8%-15%, it is the severe wear stage; when M exceeds 15%, the gear is considered to have failed. S13: Secondly, a nonlinear function is used to further divide the gear wear running-in stage, stable wear stage, and severe wear stage into twelve intervals from A to L, and the gear tooth profile wear curve corresponding to each interval is calculated; a nonlinear time division method is used to make the time intervals in the running-in stage and severe wear stage shorter, while the time interval in the stable wear stage longer, so as to improve the accuracy of wear changes in a short period of time and adapt to the changing characteristics of wear rate, thereby improving the overall calculation efficiency and prediction accuracy. Nonlinear function expression: Where: t i is the end time of the ith interval, T is the total time, N is the number of intervals, b is the exponential base used to control the growth rate; c is the intermediate node coefficient.

3. A method for diagnosing a high-speed gear milling machine spindle box fault based on fault mechanism simulation and data fusion according to claim 1, characterized in that: Acquiring the simulation data set in S2 mainly includes the following steps: S21: Use 3D modeling software to construct a 3D geometric model of the spindle box system, including the gear set, bearings, and related support structures inside the spindle box, and construct the corresponding fault gear model based on the tooth profile wear curve; S22: Import the spindle box system model into the multi-body dynamics simulation software, impose corresponding constraints on the simulation model according to the actual situation, and obtain simulation data sets under different gear wear degrees.

4. A method for diagnosing a high-speed gear milling machine spindle box fault based on fault mechanism simulation and data fusion according to claim 1, characterized in that: The acquisition of the actual data set of the spindle box in the aforementioned S3 mainly includes the following steps: building a spindle box data acquisition system for the gear milling machine, selecting an appropriate vibration acceleration sensor and installation method according to the measurement requirements, and installing the vibration acceleration sensor on the bearing seats of the first, third, and fifth axes of the spindle box using magnetic attraction.

5. The method for diagnosing a high-speed gear milling machine spindle box fault based on fault mechanism simulation and data fusion according to claim 1, characterized in that: The construction of the spindle box system fault diagnosis model in S4 mainly includes the following steps: S41: Based on the temporal characteristics of gear wear faults, a fault diagnosis model is established that integrates adaptive wavelet transform, self-attention long short-term memory (AWT-SA-LSTM) network, and dynamic gated convolutional neural network (DG-CNN). The model consists of two convolutional layers, two pooling layers, two LSTM layers, one fully connected layer, and a softmax layer. S42: performing data normalization on the simulation data set and the spindle box experimental vibration data set, dividing the normalized data set into a training set and a validation set in a ratio of 7:3, and using the spindle box experimental vibration data set as a test set; S43: Perform adaptive wavelet transform on the preprocessed data set to obtain time-frequency features at different scales; the adaptive wavelet transform formula is: Where W x (a * ,b) is the wavelet transform coefficient, is the adaptive wavelet basis function, a * is the optimal scaling factor; S44: The time-frequency features after adaptive wavelet transformation are used as input and feature extraction is performed using the dynamic gate control convolutional neural network (DG-CNN). The calculation formula of the dynamic gate control mechanism is: Where, is the output feature of the i-th neuron in layer l, g(f(·)) is the dynamic gating function; is the i-th convolution kernel of the l-th layer, is the i-th bias term of the l-th layer; is the gate weight, σ(·) is the activation function Sigmoid, which limits the output to (0,1), and W g is the gating weight matrix, b g is the gate bias term; S45: Input the features extracted by DG-CNN into the self-attention long short-term memory network (SA-LSTM) for temporal modeling. The attention calculation formula is: Where Q, K, and V are query, key, and value matrices, respectively; S46: Use the fully connected layer and Softmax classifier to identify the gear wear state, complete the model training phase, and accurately classify the twelve types of gear wear faults. The fully connected layer dimensionality reduction calculation formula is: y=Softmax(W fc ·h T +b fc ): Where W fc is the weight matrix of the fully connected layer, b fc is the bias term of the fully connected layer; S47: Use the validation set to evaluate and validate the model by calculating performance indicators: accuracy, precision and constructing confusion matrix, and further optimize the fault diagnosis model.

6. A method for diagnosing a high-speed gear milling machine spindle box fault based on fault mechanism simulation and data fusion according to claim 1, characterized in that: In the above-mentioned S5, the experimental vibration data set collected in step S3 is input into the constructed headstock system fault diagnosis model to perform fault diagnosis and identify the health status of the headstock system.

7. A high-speed gear milling machine spindle box fault diagnosis system based on fault mechanism simulation and data fusion, characterized in that: include: Simulation system: used to simulate the simulation data when the spindle box system fails; Data acquisition system: used to collect spindle box experimental vibration data; Fault diagnosis system: used to perform normalization processing, feature extraction, and state identification on simulated vibration data to build a fault diagnosis model; Condition identification system: used to identify the actual health condition of the spindle box.

Images