Machine tool spindle fault diagnosis model establishment method and machine tool spindle fault diagnosis method

Abstract

The invention discloses a method for establishing a machine tool spindle fault diagnosis model and a machine tool spindle fault diagnosis method, which belongs to the field of machine tool spindle fault diagnosis, including: collecting signals related to the health status of the machine tool spindle to construct a training data set, and dividing it into multiple training subsets; establishing a convolutional neural network, and serving as an Actor module and a Critic module of an Actor-Critic algorithm at the same time, for outputting a strategy distribution and an evaluation value; one training subset corresponds to one training round, and the training process includes: inputting a training sample into the Actor module, outputting the strategy distribution at the current moment, and calculating the state at the next moment; inputting the state at the current and next moments into the Critic module, and outputting evaluation values ​​for the two states; all rounds are one iteration, and after each iteration, the network parameters are updated according to the losses of the Actor module and the Critic module; after the iteration is terminated, the Actor module is used as a machine tool spindle fault diagnosis model. The invention can improve the accuracy and robustness of the fault diagnosis result.

Description

Technical Field

[0001] The present invention belongs to the field of machine tool spindle fault diagnosis, and more specifically, relates to a machine tool spindle fault diagnosis model establishment method and a machine tool spindle fault diagnosis method. Background Art

[0002] Machine tools are very important working machines in the industrial field. Among them, the machine tool spindle is an important part of the machine tool. During the processing, a sudden failure of the machine tool spindle will affect the quality of the processed parts, reduce the working efficiency of the machine tool and the qualified rate of the parts, and increase the manufacturing cost. In severe cases, it will cause the machine tool to break, hit, and other phenomena, damage other high-value parts of the machine tool, and may even endanger the personal safety of novice operators. In factories where it is impossible to shut down for a long time, when operators without rich experience face a sudden failure of the machine tool spindle, on the one hand, they cannot quickly find the cause of the failure and reduce work efficiency, and on the other hand, it is easy to cause more serious accidents and increase costs or dangers. Therefore, in recent years, with the breakthrough of artificial intelligence technology, the intelligent diagnosis system of machine tool spindles that does not require the participation of machine tool operators has been continuously developing.

[0003] In the field of machine tool spindle fault diagnosis, the existing intelligent diagnosis methods mainly collect fault signals through sensors, manually extract fault characteristic parameters, and then use a neural network model based on deep learning to train the fault characteristic parameters. For example, a Chinese patent document with publication number CN 105081879A and publication date November 25, 2015 discloses a method for fault diagnosis and prediction of CNC machine tool spindles, which uses advanced artificial intelligence technologies such as neural networks, expert systems and support vector machines for fault diagnosis and prediction, and quickly analyzes the measurement data transmitted from high-precision laser displacement sensors, infrared temperature sensors and acceleration sensors through a hybrid reasoning model.

[0004] The CNC machine tool spindle state detection method disclosed in the patent document needs to extract fault feature parameters first, then use the fault parameters as training data of the neural network to train the model, and finally the network analyzes the data of the machine tool operation to give the fault result. Obviously, there are four shortcomings here: First, the method based on manual feature extraction is prone to lose a lot of data information, and has high requirements on the operator's experience; second, in the data processing process of this method, the multiple signals collected do not have a data fusion process, and the data information is not fully utilized; third, there is no optimization of the neural network. The neural network based on simple artificial intelligence technology is very easy to be limited to the local optimum, resulting in the diagnosis model only having high-precision diagnosis results for the training data and lacking generalization ability; fourth, a health status indicator of the current state of the spindle is not calculated, so the operator does not have a clear understanding of the health status of the current machine tool spindle. Summary of the invention

[0005] In view of the defects of the prior art and the need for improvement, the present invention provides a method for establishing a machine tool spindle fault diagnosis model and a machine tool spindle fault diagnosis method, the purpose of which is to optimize the network's search ability for the global optimum, reduce the degree of dependence on expert experience, and improve the accuracy and robustness of the fault diagnosis results by constructing a deep network that can autonomously identify fault characteristics and combining the reinforcement learning algorithm Actor-Critic when training the neural network to add decision-making capabilities.

[0006] To achieve the above object, according to one aspect of the present invention, a method for establishing a machine tool spindle fault diagnosis model is provided, comprising:

[0007] Acquire signals related to the health status of a machine tool spindle to construct a training data set; divide the training data set into a plurality of training subsets;

[0008] Establish a convolutional neural network, which serves as both the Actor module and the Critic module of the Actor-Critic algorithm. The convolutional neural network is used to compress high-dimensional information into low-dimensional information, and output a strategy distribution π and an evaluation value V. The strategy distribution π represents the probability that the input signal belongs to each preset fault type, and the evaluation value V represents the evaluation of the state corresponding to the strategy distribution π.

[0009] The training samples in a training subset are used as training data for one round of the Actor-Critic algorithm. For each training sample, the training process of the Actor-Critic algorithm includes: inputting the training sample as the current state into the Actor module, so that the Actor module outputs the policy distribution at the current moment; calculating the state at the next moment after executing the action according to the policy distribution at the current moment; inputting the current state and the state at the next moment into the Critic module, so that the Critic module outputs the evaluation value for the current state and the next state;

[0010] All rounds of training constitute one iterative training. After each iterative training, the parameters of the convolutional neural network are updated according to the losses of the Actor module and the Critic module.

[0011] Repeat the iterative training until the preset iteration termination condition is met; after the iteration is terminated, the Actor module is used as the machine tool spindle fault diagnosis model.

[0012] The machine tool spindle fault diagnosis model established by the present invention is completely driven by data to complete the machine tool spindle fault diagnosis, which avoids dependence on expert experience, improves the efficiency of fault diagnosis system development, and reduces development costs; and, when establishing the machine tool spindle fault diagnosis model, the present invention uses the same convolutional neural network as both the Actor module and the Critic module of the Actor-Critic algorithm, so that the neural network has the functions of self-evaluation and self-improvement, and combines the influence of the current selection on the future reward in the Actor-Critic algorithm, so that the neural network can explore the global optimum and avoid falling into the local optimum, thereby improving the generalization ability and ensuring the accuracy and robustness of the fault diagnosis result.

[0013] Furthermore, in each iteration of training, the loss function used is:

[0014]

[0015] Among them, Loss(θ) represents the overall loss, L represents the total number of training rounds, and l represents the training round number. and They represent the losses of the Actor module and the Critic module in the lth training round, λ represents the loss weight parameter, and θ represents the model parameter.

[0016] At the end of each iterative training, the present invention updates the parameters of the convolutional neural network, and the loss function used includes the losses of the Actor module and the Critic module at the same time, further ensuring the self-evaluation and self-improvement functions of the convolutional neural network.

[0017] Further,

[0018] Where T represents the maximum number of time steps in a round; π(a t |s t ) represents the strategy distribution at time t, a t Denotes the distribution from the strategy π(a t |s t ) in the action selected; A(s t ,a t ) represents the advantage function, A(s t ,a t )=r t +γV t+1 -V t , r t Indicates execution of action a t The reward obtained later, V t and V t+1They represent the evaluation values ​​of the states at time t and time t+1 respectively, and γ represents the discount factor for each round.

[0019] Further,

[0020] in, Represents the cumulative sum of future rewards.

[0021] Furthermore, in each training subset, the number of training samples is greater than 1.

[0022] When the present invention combines the Actor-Critic algorithm for model training, multiple training samples are used in each round, thereby being able to retain the impact of the current action on subsequent actions during the training process, thereby improving the model training effect.

[0023] Furthermore, in each training subset, the number of training samples is equal to the number of preset fault types.

[0024] When the present invention combines the Actor-Critic algorithm for model training, the number of training samples used in each round is the same as the number of fault types, thereby achieving the best model training effect.

[0025] Furthermore, signals related to the health status of the machine tool spindle are collected to construct a training data set, including:

[0026] Collecting multiple signals related to the health status of the machine tool spindle at the same sampling rate and normalizing the collected signals;

[0027] The normalized signal is synchronously segmented, and multiple signals in the same time period are used to construct a training sample; each training sample contains multiple data points, and each data point is formed by stacking multiple signals at the same sampling point in sequence; all training samples constitute a training data set.

[0028] The present invention simultaneously collects a variety of signals related to the health status of the machine tool spindle, and constructs training samples by synchronous segmentation and signal stacking, thereby realizing data fusion, thereby making full use of the effective information of the data and effectively improving the accuracy of the model in fault diagnosis.

[0029] According to another aspect of the present invention, a method for diagnosing a machine tool spindle fault is provided, comprising:

[0030] Signals related to the health status of the machine tool spindle are collected and processed into samples to be tested; the samples to be tested are input into the machine tool spindle fault diagnosis model obtained by the machine tool spindle fault diagnosis model establishment method provided by the present invention, so that the machine tool spindle fault diagnosis model outputs the corresponding strategy distribution, and the fault type with the highest probability in the strategy distribution is determined as the fault type of the machine tool spindle.

[0031] The machine tool spindle fault diagnosis model established by the machine tool spindle fault diagnosis model establishment method provided by the present invention does not rely on expert experience and has strong generalization ability. Based on the machine tool spindle fault diagnosis model, the machine tool spindle fault diagnosis method provided by the present invention has high reliability and robustness.

[0032] Furthermore, the machine tool spindle fault diagnosis method provided by the present invention also includes:

[0033] according to Calculate the health status index of the machine tool spindle;

[0034] Where HI represents the health status index of the machine tool spindle; C represents the number of preset fault types, i represents the fault type number; a i is the i-th element of the strategy distribution output by the machine tool spindle fault diagnosis model, which corresponds to the probability of the i-th type of fault occurring in the machine tool spindle state; h i is the unit vector of the health state, where the number of elements is C, the i-th element is 1, and the rest are 0.

[0035] The health status index of the machine tool spindle calculated by the present invention based on the above formula quantifies the distance between the fault diagnosis result output by the model and the actual standard fault category, and accurately quantifies the current health status of the machine tool spindle, which is helpful to help operators and maintainers understand the working status of the spindle more clearly in real time, improve work efficiency, and avoid sudden dangerous situations.

[0036] According to another aspect of the present invention, a computer-readable storage medium is provided, comprising a stored computer program; when the computer program is executed by a processor, the device where the computer-readable storage medium is located is controlled to execute the machine tool spindle fault diagnosis model establishment method provided by the present invention, and / or the machine tool spindle fault diagnosis method provided by the present invention.

[0037] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0038] (1) The machine tool spindle fault diagnosis model established by the present invention is completely driven by data to complete the machine tool spindle fault diagnosis, avoiding dependence on expert experience, improving the efficiency of fault diagnosis system development, and reducing development costs; and, when establishing the machine tool spindle fault diagnosis model, the present invention uses the same convolutional neural network as both the Actor module and the Critic module of the Actor-Critic algorithm, so that the neural network has the functions of self-evaluation and self-improvement, and combines the influence of the current selection on the future reward in the Actor-Critic algorithm, so that the neural network can explore the global optimum and avoid falling into the local optimum, thereby improving the generalization ability and ensuring the accuracy and robustness of the fault diagnosis results.

[0039] (2) The method for establishing a machine tool spindle fault diagnosis model provided by the present invention, when combined with the Actor-Critic algorithm for model training, uses the same convolutional neural network to represent both the Actor module and the Critic module, which can effectively reduce network parameters and improve operating efficiency.

[0040] (3) The method for establishing a machine tool spindle fault diagnosis model provided by the present invention, when combined with the Actor-Critic algorithm for model training, uses multiple training samples in each round, thereby being able to retain the impact of the current action on the subsequent actions during the training process, thereby improving the model training effect; in its preferred embodiment, the number of training samples used in each round is the same as the number of fault types, which can obtain the best model training effect.

[0041] (4) The method for establishing a machine tool spindle fault diagnosis model provided by the present invention simultaneously collects a variety of signals related to the health status of the machine tool spindle, and constructs training samples by synchronous segmentation and signal stacking, thereby realizing data fusion, thereby making full use of the effective information of the data and effectively improving the accuracy of the model in fault diagnosis.

[0042] (5) The machine tool spindle fault diagnosis method provided by the present invention provides a quantitative calculation method for the health status index while performing fault diagnosis, which is helpful to help operators and maintainers understand the working status of the spindle more clearly in real time, improve work efficiency, and avoid sudden dangerous situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram of a method for establishing a machine tool spindle fault diagnosis model provided by an embodiment of the present invention;

[0044] Figure 2 A schematic diagram of a convolutional neural network provided by an embodiment of the present invention;

[0045] Figure 3It is a schematic diagram of the existing Actor-Critic algorithm;

[0046] Figure 4 A schematic diagram of data fusion provided by an embodiment of the present invention;

[0047] Figure 5 A pseudo code schematic diagram of a method for establishing a machine tool spindle fault diagnosis model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0049] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the accompanying drawings are used to distinguish similar objects but not necessarily to describe a specific order or sequence.

[0050] In order to solve the problem that the existing machine tool spindle fault diagnosis method based on artificial intelligence relies on expert experience, has insufficient generalization ability, and has poor reliability and stability, the present invention provides a method for establishing a machine tool spindle fault diagnosis model and a machine tool spindle fault diagnosis method, the overall idea of ​​which is to build a deep network that can autonomously identify fault features, combine the reinforcement learning algorithm Actor-Critic to add decision-making capabilities when training the neural network, so as to optimize the network's search ability for the global optimum, reduce the degree of dependence on expert experience, and improve the accuracy and robustness of the fault diagnosis results; further, through the method of multi-data fusion, make full use of the effective information in the data to improve the accuracy and reliability of fault diagnosis; on this basis, further propose a quantitative calculation method for the health status of the machine tool spindle to clearly and real-time reflect the health status of the machine tool spindle. The following is an embodiment.

[0051] Embodiment 1:

[0052] A method for establishing a machine tool spindle fault diagnosis model, such as Figure 1 and Figure 5 As shown, including:

[0053] Acquire signals related to the health status of a machine tool spindle to construct a training data set; divide the training data set into a plurality of training subsets;

[0054] A convolutional neural network is established, which serves as both the Actor module and the Critic module of the Actor-Critic algorithm. The convolutional neural network is used to compress high-dimensional information into low-dimensional information, and output a strategy distribution π and an evaluation value V. The strategy distribution π represents the probability that the input signal belongs to each preset fault type. The strategy distribution π is specifically a one-dimensional vector, in which the number of elements is the same as the number of fault types, and the sum of all elements is 1, and each element represents the probability that the network determines that the input signal belongs to the corresponding fault type. The evaluation value V represents the evaluation of the state corresponding to the strategy distribution π, and V is specifically a rational number. In this embodiment, the structure of the established convolutional neural network is as follows: Figure 2 As shown;

[0055] The training samples in a training subset are used as the training data for one round of the Actor-Critic algorithm. For each training sample, the training process of the Actor-Critic algorithm is as follows: Figure 3 As shown, it includes: inputting the training sample as the current state into the Actor module, so that the Actor module outputs the policy distribution at the current moment; calculating the state at the next moment after executing the action according to the policy distribution at the current moment; inputting the current state and the state at the next moment into the Critic module, so that the Critic module outputs the evaluation value for the current state and the next state;

[0056] All rounds of training constitute one iterative training. After each iterative training, the parameters of the convolutional neural network are updated according to the losses of the Actor module and the Critic module.

[0057] Repeat the iterative training until the preset iteration termination condition is met; after the iteration is terminated, the Actor module is used as the machine tool spindle fault diagnosis model.

[0058] In this embodiment, there are 9 specific fault types that need to be identified, as shown in Table 1. In order to make full use of the effective information in the data and improve the reliability and accuracy of the machine tool spindle fault diagnosis, the signals collected in this embodiment are related to the health status of the machine tool spindle, specifically the vibration signals in three vertical directions when the machine tool spindle is working, which can be collected by a three-phase acceleration sensor. After the corresponding signals are collected, a training data set will be constructed by data fusion. It should be noted that this is only an exemplary description of the present invention and should not be understood as the only limitation of the present invention. According to actual application requirements, the present invention can diagnose any type and any number of faults, and can also collect any type and any number of signals to complete fault diagnosis, as long as the corresponding fault type is a fault that may actually occur during the operation of the machine tool spindle, and the collected signals are collected from the site and are related to the health status of the machine tool spindle. More examples will not be listed here one by one.

[0059] Table 1

[0060] Fault Status Serial number Bearing inner ring failure and normal tool wear (IN) 1 Bearing inner ring failure and moderate tool wear (IM) 2 Bearing inner ring failure and severe tool wear (IS) 3 Bearing outer ring failure and normal tool wear (ON) 4 Bearing outer ring failure and moderate tool wear (OM) 5 Bearing outer ring failure and severe tool wear (OS) 6 Bearings are normal and tool wear is normal (NN) 7 Bearings are normal and tool wear is moderate (NM) 8 Bearing is normal and tool is severely worn (NS) 9

[0061] like Figure 1 and Figure 4 As shown, in this embodiment, signals related to the health status of the machine tool spindle are collected to construct a training data set, including:

[0062] The vibration signals of the machine tool spindle in three vertical directions are collected at the same sampling rate. The collected signals are recorded as signals of channel x, channel y, and channel z in sequence, and the collected signals are normalized.

[0063] The normalized signal is segmented synchronously, and multiple signals in the same time period are used to construct a training sample; Figure 4 As shown in the figure, after synchronous segmentation, the signals of the three channels are segmented into multiple data segments of length N; the three channel signals in the same time period are represented as x1~x N ,y1~y N and z1~z N , used to construct a training sample;

[0064] Each training sample contains multiple data points, and each data point is formed by stacking multiple signals at the same sampling point in sequence; Figure 4 As shown in the figure, x1, y1, and z1 are stacked in sequence as the first data point of the training sample. i ,y i 、z i After stacking in sequence, they are used as the i-th data point of the training sample, and so on;

[0065] The training data set is composed of all training samples;

[0066] This embodiment constructs training samples in the above manner and realizes data fusion, thereby making full use of effective information in the data and effectively improving the accuracy of fault diagnosis performed by the model.

[0067] like Figure 1 and Figure 5 As shown, in each iteration of training, the loss function used is:

[0068]

[0069]

[0070]

[0071] Among them, Loss(θ) represents the overall loss, L represents the total number of training rounds, and l represents the training round number. and They represent the losses of the Actor module and the Critic module in the lth training round, λ represents the loss weight parameter, θ represents the model parameter; T represents the maximum time step of a round; π(a t |s t ) represents the strategy distribution at time t, a t Denotes the distribution from the strategy π(a t |s t ) in the action selected; A(s t ,a t ) represents the advantage function, A(s t ,a t )=r t +γV t+1 -V t , r t Indicates execution of action a t The reward obtained later, V t and V t+1 They represent the evaluation values ​​of the states at time t and time t+1 respectively, γ represents the discount factor for each round, represents the cumulative sum of future rewards;

[0072] At the end of each iterative training, this embodiment updates the parameters of the convolutional neural network based on the above loss function. The loss function includes the losses of the Actor module and the Critic module, further ensuring the self-evaluation and self-improvement functions of the convolutional neural network.

[0073] As a preferred implementation, in this embodiment, in each training subset, the number of training samples, that is, the number of training samples used in each round, is equal to the number of preset fault types, which is 9; experiments show that setting the number of training samples in each round to be equal to the number of fault types can obtain the best training effect;

[0074] It should be noted that the setting of the number of training samples used in each round here is only a preferred implementation of the present invention. In some other embodiments of the present invention, the number of training samples in each round can also be set to other values ​​greater than 1, thereby retaining the impact of the current action on the subsequent actions during the training process and improving the model training effect; in some other embodiments, only one training sample can be used directly in each round.

[0075] In general, the machine tool spindle fault diagnosis model established in this embodiment is completely data-driven to complete fault diagnosis, avoiding dependence on expert experience, improving the efficiency of fault diagnosis system development, and reducing development costs; combining the Actor-Critic algorithm to enable the neural network to have decision-making capabilities, imitate the choices of human thinking, and give rewards and punishments to the decisions of the neural network, optimize the neural network algorithm, and help the neural network jump out of the local optimum as much as possible, thereby improving the reliability and robustness of fault diagnosis; using a variety of data fusion methods to construct a training data set, fully utilizing the effective information of the data, and increasing the accuracy of the fault diagnosis results.

[0076] Embodiment 2:

[0077] A method for diagnosing a machine tool spindle fault, comprising:

[0078] Collect signals related to the health status of the machine tool spindle and process them into samples to be tested; input the samples to be tested into the machine tool spindle fault diagnosis model obtained by the machine tool spindle fault diagnosis model establishment method provided in the above embodiment 1, so that the machine tool spindle fault diagnosis model outputs a corresponding strategy distribution, and determines the fault type with the highest probability in the strategy distribution as the fault type of the machine tool spindle;

[0079] In this embodiment, the method of processing the collected signal related to the health status of the machine tool spindle into a sample to be tested can refer to the description in the above embodiment 1, specifically, including:

[0080] The vibration signals of the three vertical directions of the machine tool spindle are collected at the same sampling rate, and the collected signals are normalized;

[0081] The normalized signal is synchronously segmented, and a test sample is constructed using multiple signals in the current time period; the test sample includes multiple data points, and each data point is formed by stacking multiple signals at the same sampling point in sequence;

[0082] The test sample is input into the machine tool spindle fault diagnosis model established in the above embodiment 1, and the corresponding strategy distribution can be obtained; in the strategy distribution, the fault type with the highest probability is expressed as follows:

[0083]

[0084] Among them, a i represents the i-th element in the strategy distribution output by the machine tool spindle fault diagnosis model, C is the total number of fault types, that is, the number of elements in the strategy distribution; a represents the element number with the highest probability in the strategy distribution;

[0085] In order to help operators and maintainers understand the working status of the spindle more clearly in real time, improve work efficiency, and avoid sudden dangerous situations, this embodiment also includes:

[0086] according to Calculate the health status index of the machine tool spindle;

[0087] Where HI represents the health status index of the machine tool spindle; C represents the number of preset fault types, i represents the fault type number; a i is the i-th element of the strategy distribution output by the machine tool spindle fault diagnosis model, which corresponds to the probability of the i-th type of fault occurring in the machine tool spindle state; h i is the unit vector of the health state, where the number of elements is C, the i-th element is 1, and the rest are 0;

[0088] The health status index of the machine tool spindle calculated by the above health status index formula quantifies the distance between the fault diagnosis result output by the model and the actual standard fault category, and accurately quantifies the current health status of the machine tool spindle. HI is a value between 0 and 1. The larger the HI value, the closer the health status of the machine tool spindle is to normal, and the smaller the HI value, the closer the health status of the machine tool spindle is to fault.

[0089] Embodiment 3:

[0090] A computer-readable storage medium includes a stored computer program; when the computer program is executed by a processor, the device where the computer-readable storage medium is located is controlled to execute the machine tool spindle fault diagnosis model establishment method provided in the above-mentioned embodiment 1, and / or the machine tool spindle fault diagnosis method provided in the above-mentioned embodiment 2.

[0091] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for establishing a machine tool spindle fault diagnosis model, characterized in that: include: Collecting signals related to the health status of the machine tool spindle to construct a training data set; Dividing the training data set into a plurality of training subsets; Establish a convolutional neural network as both the Actor module and the Critic module of the Actor-Critic algorithm; the convolutional neural network is used to compress high-dimensional information into low-dimensional information and output the strategy distribution π and the evaluation value V , the strategy distribution π represents the probability that the input signal belongs to each preset fault type, and the evaluation value V represents the evaluation of the state corresponding to the strategy distribution π; A training sample in a training subset is used as training data for one round of the Actor-Critic algorithm. For each training sample, the training process of the Actor-Critic algorithm includes: inputting the training sample as the current state into the Actor module, so that the Actor module outputs the policy distribution at the current moment; calculating the state at the next moment after executing the action according to the policy distribution at the current moment; inputting the current state and the state at the next moment into the Critic module, so that the Critic module outputs the evaluation value for the current state and the next state; All rounds of training constitute one iterative training; after each iterative training, the parameters of the convolutional neural network are updated according to the losses of the Actor module and the Critic module; Repeat the iterative training until the preset iteration termination condition is met; after the iteration is terminated, the Actor module is used as a machine tool spindle fault diagnosis model; The signal related to the health status of the machine tool spindle is collected to construct a training data set, including: collecting a plurality of signals related to the health status of the machine tool spindle at the same sampling rate, and normalizing the collected signals; The normalized signal is synchronously segmented, and multiple signals in the same time period are used to construct a training sample; each training sample contains multiple data points, and each data point is formed by stacking multiple signals at the same sampling point in sequence; all training samples constitute a training data set.

2. The method for establishing a machine tool spindle fault diagnosis model according to claim 1, characterized in that: In each iteration of training, the loss function used is: ； in, represents the overall loss, L represents the total number of training rounds, represents the training round number, and Respectively represent The loss of the Actor module and the Critic module in the training round, represents the loss weight parameter, Represents model parameters.

3. The method for establishing a machine tool spindle fault diagnosis model according to claim 2, characterized in that: ； in, T Indicates the maximum number of time steps in a round; express The strategy distribution at the moment, Indicates distribution from strategy The action selected in represents the advantage function, , Indicates execution of an action The rewards received later, and Respectively for Moment and The evaluation value of the state at time Represents the discount factor for each round.

4. The method for establishing a machine tool spindle fault diagnosis model according to claim 3, characterized in that: ； in, Represents the cumulative sum of future rewards.

5. The method for establishing a machine tool spindle fault diagnosis model according to claim 1, characterized in that: In each training subset, the number of training samples is greater than 1.

6. The method for establishing a machine tool spindle fault diagnosis model according to claim 5, characterized in that: In each training subset, the number of training samples is equal to the number of preset fault types.

7. A method for diagnosing a machine tool spindle fault, characterized in that: include: Collect signals related to the health status of the machine tool spindle and process them into samples to be tested; input the samples to be tested into a machine tool spindle fault diagnosis model obtained by the machine tool spindle fault diagnosis model establishment method described in any one of claims 1 to 6, so that the machine tool spindle fault diagnosis model outputs a corresponding strategy distribution, and determines the fault type with the highest probability in the strategy distribution as the fault type of the machine tool spindle.

8. The machine tool spindle fault diagnosis method according to claim 7, characterized in that: Also includes: according to Calculating a health status index of the machine tool spindle; in, HI Indicates the health status index of the machine tool spindle; C Indicates the number of preset fault types, i Indicates the fault type number; The strategy distribution output by the machine tool spindle fault diagnosis model is i element, which indicates the state of the machine tool spindle. i Probability of class failure; is the unit vector of health status, where the number of elements is C , and among them i elements are 1, and the rest are 0.

9. A computer-readable storage medium, characterized in that: It includes a stored computer program; when the computer program is executed by a processor, it controls the device where the computer-readable storage medium is located to execute the machine tool spindle fault diagnosis model establishment method described in any one of claims 1-6, and / or the machine tool spindle fault diagnosis method described in claim 7 or 8.

Images