Fault identification method for machine tool spindle, fault identification model training method and device
By combining two fault identification models, using pre-trained models and feature extraction technology, the problem of insufficient fault data in the machine tool spindle system is solved, the accuracy and reliability of fault identification are improved, and efficient identification of machine tool spindle faults is achieved.
Patent Information
- Application Number
- CN202110895587.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-05
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2041-08-05
AI Technical Summary
It is difficult for the prior art to obtain sufficient fault data samples of the machine tool spindle, resulting in failure prediction based on artificial intelligence technology in machine tool spindle systems and failure to accurately identify faults.
Using the combination method of two fault recognition models, the first fault recognition model and the second fault recognition model are trained through neural network and used the weight value of the pre-trained model, combined with order analysis and wavelet packet decomposition, extract fault features and reduce individual The amount of data required for model training improves recognition accuracy.
It improves the accuracy and reliability of machine tool spindle fault identification, reduces the demand for the number of fault data samples, and enhances the generalization ability and stability of the model.
Smart Images

Figure CN113762344B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of fault identification, and in particular to a fault identification method for a machine tool spindle. Background Art
[0002] As one of the most important and widely used processing equipment in modern mechanical manufacturing, the operating status of machine tools will affect the working efficiency of the entire manufacturing system. Due to frequent use, complex mechanical structure, and harsh working environment, machine tool equipment, especially the spindle system as the core component of machine tools and which maintains high-speed rotation for a long time, is usually difficult to accurately locate and predict various faults (such as spindle imbalance, spindle friction, etc.) through traditional physical models. Once a fault occurs, it is very likely to cause significant property losses and serious consequences.
[0003] In recent years, with the development and popularization of artificial intelligence and machine learning technologies, many scholars and researchers have tried to apply data-driven algorithm models to the field of fault prediction and diagnosis, and have achieved certain results. The core idea of these methods is to fully train machine learning models by collecting a large amount of equipment operation data under normal and fault conditions, so that the trained models can identify and locate various types of faults based on the input equipment operation data.
[0004] However, for the spindle system of machine tools, in the actual production process, due to the special characteristics of the machine tool spindle system such as high-speed rotation, it is extremely difficult to obtain enough fault data samples for model training for all types of faults, making it difficult for fault prediction technology based on artificial intelligence technology to achieve the expected effect in this field, which has become an industry pain point in this field. Summary of the invention
[0005] The technical problem to be solved by the present invention is to overcome the defect in the prior art that it is difficult to obtain sufficient data samples of a machine tool spindle for model training, and thus it is impossible to obtain an accurate model for identifying machine tool spindle faults, and to provide a machine tool spindle fault identification method, a fault identification model training method and a device.
[0006] The present invention solves the above technical problems through the following technical solutions:
[0007] In a first aspect, a method for identifying a fault of a machine tool spindle is provided, comprising:
[0008] Acquiring operation data of the machine tool spindle;
[0009] The operation data is input into a first fault identification model and a second fault identification model; wherein the first fault identification model is obtained by training a neural network according to first fault sample data corresponding to a first fault type of the machine tool spindle, and the second fault identification model is obtained by training a pre-trained model according to second fault sample data corresponding to a second type of fault of the machine tool spindle, and the pre-trained model is constructed according to model parameters of the first fault identification model;
[0010] Fault identification is performed on the machine tool spindle using the first fault identification model and the second fault identification model.
[0011] Optionally, inputting the operating data into the first fault identification model and the second fault identification model comprises:
[0012] respectively determining a fault feature corresponding to the first fault type and a fault feature corresponding to the second fault type;
[0013] inputting first operating data in the operating data that matches the fault feature corresponding to the first fault type into the first fault identification model;
[0014] Second operating data in the operating data that matches the fault feature corresponding to the second fault type is input into the first fault identification model.
[0015] Optionally, the model parameters include weight values;
[0016] The first fault identification model has the same weight value as the pre-trained model.
[0017] In a second aspect, a training method for a fault identification model is provided, wherein the fault identification model is used to identify a fault of a machine tool spindle, and the training method comprises:
[0018] Using first fault sample data corresponding to a first fault type of the machine tool spindle, training a first neural network to obtain a first fault identification model for identifying the first fault type;
[0019] Constructing a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model;
[0020] The pre-trained model is trained using second fault sample data corresponding to a second fault type of the machine tool spindle to obtain a second fault identification model for identifying the second fault type.
[0021] Optionally, it also includes:
[0022] determining a fault signature corresponding to a first fault type;
[0023] Performing order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain first fault sample data corresponding to the fault characteristics of the first fault type;
[0024] and / or, determining a fault characteristic corresponding to a second fault type;
[0025] Performing order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain second fault sample data corresponding to the fault feature of the second fault type.
[0026] Optionally, the model parameters include weight values;
[0027] Constructing a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model, including:
[0028] The weight value of the pre-trained model is configured according to the weight value of the first fault identification model.
[0029] Optionally, it also includes:
[0030] Using the second fault sample data, training a second neural network to obtain a verification model of the second fault identification model, wherein the second neural network has the same network structure as the pre-trained model;
[0031] Performing a five-fold cross validation on the second fault identification model according to the validation model;
[0032] The verified second fault identification model is determined as the final second fault identification model.
[0033] In a third aspect, a fault identification device for a machine tool spindle is provided, comprising:
[0034] An acquisition module, used for acquiring the operation data of the machine tool spindle;
[0035] An input module, used to input the operation data into a first fault recognition model and a second fault recognition model; wherein the first fault recognition model is obtained by training a neural network according to first fault sample data corresponding to a first fault type of the machine tool spindle, and the second fault recognition model is obtained by training a pre-trained model according to second fault sample data corresponding to a second type of fault of the machine tool spindle, and the pre-trained model is constructed according to model parameters of the first fault recognition model;
[0036] An identification module is used to perform fault identification on the machine tool spindle using the first fault identification model and the second fault identification model.
[0037] Optionally, the input module includes:
[0038] a determination unit, configured to respectively determine a fault feature corresponding to the first fault type and a fault feature corresponding to the second fault type;
[0039] a first input unit, configured to input first operating data in the operating data that matches a fault feature corresponding to the first fault type into the first fault identification model;
[0040] The second input unit is used to input second operating data in the operating data that matches the fault feature corresponding to the second fault type into the first fault identification model.
[0041] Optionally, the model parameters include weight values;
[0042] The first fault identification model has the same weight value as the pre-trained model.
[0043] In a fourth aspect, a training device for a fault identification model is provided, wherein the fault identification model is used to identify a fault of a machine tool spindle, and the training device comprises:
[0044] A first training module, used to train a first neural network using first fault sample data corresponding to a first fault type of the machine tool spindle, to obtain a first fault recognition model for identifying the first fault type;
[0045] A construction module, used to construct a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model;
[0046] The second training module is used to train the pre-trained model using second fault sample data corresponding to the second fault type of the machine tool spindle to obtain a second fault identification model for identifying the second fault type.
[0047] Optionally, it also includes:
[0048] A first determination module is used to determine a fault feature corresponding to a first fault type, and perform order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain first fault sample data corresponding to the fault feature of the first fault type;
[0049] And / or, a second determination module is used to determine the fault characteristics corresponding to the second fault type, and perform order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain second fault sample data corresponding to the fault characteristics of the second fault type.
[0050] Optionally, the model parameters include weight values;
[0051] The building blocks are specifically used for:
[0052] The weight value of the pre-trained model is configured according to the weight value of the first fault identification model.
[0053] Optionally, it also includes:
[0054] A verification module is used to use the second fault sample data to train a second neural network to obtain a verification model of the second fault identification model, wherein the second neural network has the same network structure as the pre-trained model; and according to the verification model, a five-fold cross-validation is performed on the second fault identification model, and the second fault identification model that passes the verification is determined as the final second fault identification model.
[0055] The positive and progressive effect of the present invention is that faults of a machine tool spindle are identified by using a first fault identification model and a second fault identification model. Since the second fault identification model is obtained by fine-tuning the first fault identification model based on the first fault identification model using fault sample data corresponding to the second fault type, it stores and accumulates the knowledge learned in the first fault type, which can improve the accuracy and reliability of the second fault identification model, thereby improving the accuracy and reliability of fault identification of the machine tool spindle. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 A flowchart of a method for training a fault identification model provided by an exemplary embodiment of the present invention;
[0057] Figure 2a A flow chart of a method for identifying a fault of a machine tool spindle provided by an exemplary embodiment of the present invention;
[0058] Figure 2b A schematic diagram of an algorithm architecture of a fault identification method for a machine tool spindle provided by an exemplary embodiment of the present invention;
[0059] Figure 3 A module schematic diagram of a training device for a fault identification model provided by an exemplary embodiment of the present invention;
[0060] Figure 4 A schematic diagram of a module of a machine tool spindle fault identification device provided by an exemplary embodiment of the present invention;
[0061] Figure 5 The present invention is a schematic structural diagram of an electronic device according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION
[0062] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.
[0063] Figure 1A flowchart of a method for training a fault identification model provided by an exemplary embodiment of the present invention, wherein the trained fault model is used to identify the fault of a machine tool spindle, and the method comprises the following steps:
[0064] Step 101: Use first fault sample data corresponding to a first fault type of a machine tool spindle to train a neural network to obtain a first fault identification model for identifying the first fault type.
[0065] For different fault types, corresponding to different fault features for fault identification, it is necessary to determine the fault features for identifying various fault types based on the operation data of the machine tool spindle. The operation data of the machine tool spindle can be obtained by, but not limited to, sensors installed on the machine tool spindle, such as various sensors installed near the bearings. According to the characteristic parameters extracted from the signals such as vibration, temperature, and speed monitored by various sensors, the fault features for identifying various fault types of the machine tool spindle are determined through empirical analysis and calculation, and then the fault sample data required for training the fault identification model for identifying various fault types is determined. If the number of fault types of the machine tool spindle is n, the corresponding fault sample data set can be expressed as D1, D2, ..., Dn.
[0066] The operation data contains various types of signals, such as vibration signals, temperature signals, speed signals, etc. Since the vibration signal contains extremely rich information, a large amount of feature information can be extracted. When determining the fault sample data of the corresponding fault type based on the operation data, order analysis and wavelet packet decomposition are used to extract the frequency domain and time-frequency domain feature information in the vibration signal as fault sample data. The extracted frequency domain feature information mainly includes the amplitude of the first and second frequencies of the vibration signal in three directions; the time-frequency domain feature information mainly includes the standard deviation of the low-pass coefficients and the energy distribution coefficient of each order of sub-wavelets after wavelet packet decomposition; time domain signals such as speed signals and temperature signals can be directly used as feature information.
[0067] In one embodiment, after the operation data is acquired, the operation data is preprocessed, for example, outliers with empty signals or outliers with serious signal interference are removed, and the preprocessed operation data is used to determine fault sample data.
[0068] After determining the fault characteristics of each fault type, the operation data corresponding to each fault characteristic of the machine tool spindle in the fault state and the normal operation state are obtained respectively, and the fault sample data of the corresponding fault type is determined according to the operation data. The first fault sample data is the fault sample data corresponding to the first fault type of the machine tool spindle, and the second fault sample data described below is the fault sample data corresponding to the second fault type of the machine tool spindle.
[0069] In one embodiment, after determining the fault characteristics of each fault type, the parameters corresponding to each fault characteristic in the operation data are labeled to identify the fault type corresponding to the parameter. The corresponding relationship between the parameter and the label may be a one-to-one relationship or a one-to-many relationship.
[0070] For example, assuming that the operating data includes parameters a, b, c, and d, and the parameter corresponding to fault feature A is parameter a, the label of parameter a is configured as A_Label (indicating the degree of fault and abnormality), and the data corresponding to parameter a is represented as fault sample data A_Data corresponding to fault feature A; assuming that the parameters corresponding to fault feature B are parameters a, c, and d, the labels of parameters a, c, and d are configured as B_Label (indicating the degree of fault and abnormality), and the data corresponding to parameters a, c, and d are represented as fault sample data A_Data corresponding to fault feature B. The operating data collected under the normal operating state of the machine tool spindle is identified as N_Data, and the label of the corresponding parameter is identified as N_Label. By analogy, the data corresponding to all fault types of the machine tool spindle can be labeled.
[0071] When training the first fault identification model, the first fault sample data is input into the neural network, and the neural network is trained to obtain the first fault identification model for identifying the first fault type. The first fault sample data is generally the fault sample data with a large amount of data or the largest amount of data among the fault sample data corresponding to each type of fault. Eighty percent of the fault sample data are randomly selected to directly train the model, and the remaining twenty percent of the samples are used as a verification set. The model is iterated until the model is verified using the verification set, and the iteration is stopped when the accuracy of the model is no longer improved, and the first fault identification model that has been trained is obtained. The first fault identification model is used to identify whether the machine tool spindle has a fault of the first fault type based on the operation data of the machine tool spindle.
[0072] In one embodiment, during the model training process, Adam (a first-order optimization algorithm) and a classification cross entropy function are used as the training optimizer and loss function of the model, respectively.
[0073] The neural network may be, but is not limited to, a deep neural network, the number of input layer neuron nodes of the deep neural network matches the number of fault features corresponding to the first fault type. For example, if the number of fault features corresponding to the first fault type is m, the number of input layer neuron nodes should also be m. The number of hidden layers k of the deep neural network and the number of neurons in each hidden layer are expressed as N1, N2, ..., Nk. In order to make the gradient of the model drop smoothly during the training process, the number of neurons in each layer is reduced layer by layer, that is, N1≥N2≥...≥Nk is satisfied. The network parameters of the neural network, including the network structure (including the number of hidden layers, the number of input layer neuron nodes, etc.) and the model parameters (initial values of weight values), can be selected through a large number of experiments to select the structure and parameters with better fault recognition accuracy and stability.
[0074] In one embodiment, the number of normal sample data (sample data of the machine tool spindle in normal operating state) and fault sample data (sample data of the machine tool spindle in fault state) in the fault sample data corresponding to each type of fault is consistent, so as to avoid the decrease in model accuracy due to unbalanced sample data.
[0075] Step 102: construct a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model.
[0076] Among them, the model parameters include weight values.
[0077] The weight value of the pre-trained model is the same as the weight value of the first fault recognition model, that is, the pre-trained model and the first fault recognition model share the weight value, and the weight value of the first fault recognition model is used as the initial value of the weight value of the second fault recognition model. The number of neuron nodes in the input layer of the pre-trained model matches the number of fault features corresponding to the second fault type.
[0078] Step 103: Use the second fault sample data corresponding to the second fault type of the machine tool spindle to train the pre-trained model to obtain a second fault identification model for identifying the second fault type.
[0079] When training the second fault identification model, the second fault sample data is input into the pre-trained model, and the pre-trained model is further trained to obtain the second fault identification model for identifying the second fault type. The second fault sample data is generally the fault sample data corresponding to each fault type, and the fault sample data has a smaller amount of data. Since the second fault type and the first fault type are both machine tool spindle faults and have similar characteristics, the operation data (first fault sample data) collected from the machine tool spindle running in the first fault type state can also partially characterize the abnormal law when the second fault type occurs. Therefore, on the basis of the first fault identification model, the fault sample data corresponding to the second fault type is used to fine-tune the first fault identification model to obtain the second fault identification model for identifying the second fault type. Compared with using the second fault sample data alone to train the neural network, the amount of data required for the second fault sample data will be greatly reduced. In addition, when training the second fault identification model, the first fault sample data is also combined to achieve a more comprehensive analysis of the operating data from a high dimension, thereby identifying potential faults. The knowledge learned from other faults is reused in the prediction model of similar faults, thereby greatly reducing the number of samples required for training a single model and solving the industry pain point of difficulty in obtaining a large amount of fault data on machine tool spindle systems.
[0080] During training, 80% of the fault sample data is randomly selected to directly train the model, and the remaining 20% of the samples are used as the validation set. The model is iterated until the validation set is used to verify the model. The iteration is stopped when the accuracy of the model is no longer improved, and the trained second fault identification model is obtained. The second fault identification model is used to identify whether the machine tool spindle has a fault of the second fault type based on the operating data of the machine tool spindle. The model has a strong generalization ability and can improve the stability and accuracy of fault identification.
[0081] In one embodiment, the second fault sample data is also input into the neural network to train the neural network. The network structure of the neural network is the same as that of the pre-trained model. The initial weight value of each neuron node is a random value. A verification model for verifying the accuracy of the second fault identification model is obtained. The verification model is used to verify the accuracy of the second fault identification model. Specifically, a five-fold cross validation can be used, and the training and testing are repeated five times to calculate the comprehensive prediction accuracy of each fault identification model. In step 103, the second fault identification model that has passed the verification is determined as the final second fault identification model.
[0082] In the embodiment of the present invention, on the basis of the first fault identification model, the fault sample data corresponding to the second fault type is used to fine-tune the first fault identification model, and a second fault identification model for identifying the second fault type is obtained. Compared with the use of the second fault sample data alone to train the neural network, the amount of the second fault sample data required will be greatly reduced. In addition, the training of the second fault identification model is obtained based on the training of the first fault identification model, and the knowledge learned from similar fault types is stored and accumulated, so as to achieve a more comprehensive analysis of the operating data from a high latitude, identify potential faults, and improve the accuracy and reliability of the second fault identification model, thereby improving the accuracy and reliability of the state assessment and life prediction of the machine tool spindle.
[0083] In this embodiment, the fault type of the machine tool spindle may include but is not limited to mechanical faults such as bearing wear, spindle imbalance, loose bearing, hydraulic system fault, pneumatic system fault, and lubrication system fault, etc. The first fault type may be any of the above fault types, and the second fault type may also be any of the above fault types, and the first fault type and the second fault type are different. That is, the fault identification model for identifying bearing wear faults can be used as a pre-training model, and the pre-training model can be fine-tuned using fault sample data corresponding to any fault type among spindle imbalance faults, bearing loose faults, hydraulic system faults, pneumatic system faults, and lubrication system faults to obtain a fault identification model for identifying the corresponding fault type; or, the fault identification model for identifying spindle imbalance faults can be used as a pre-training model, and the pre-training model can be fine-tuned using fault sample data corresponding to any fault type among bearing wear faults, bearing loose faults, hydraulic system faults, pneumatic system faults, and lubrication system faults to obtain a fault identification model for identifying the corresponding fault type; or, the fault identification model for identifying pneumatic system faults can be used as a pre-training model, and the pre-training model can be fine-tuned using fault sample data corresponding to any fault type among bearing wear faults, spindle imbalance, bearing loose faults, hydraulic system faults, and lubrication system faults to obtain a fault identification model for identifying the corresponding fault type.
[0084] Figure 2a A flowchart of a fault identification method for a machine tool spindle is provided as an exemplary embodiment of the present invention. The method comprises the following steps:
[0085] Step 201: Obtain the operation data of the machine tool spindle.
[0086] The operation data may be the operation data of the machine tool spindle within a preset time period from the current moment.
[0087] Step 202: Input the operating data into the first fault identification model and the second fault identification model.
[0088] In one embodiment, when operating data is input into a first fault identification model and a second fault identification model, the fault characteristics corresponding to the first fault type and the fault characteristics corresponding to the second fault type are respectively determined, and the first operating data in the operating data that matches the fault characteristics corresponding to the first fault type is input into the first fault identification model, and the second operating data in the operating data that matches the fault characteristics corresponding to the second fault type is input into the first fault identification model.
[0089] The second fault identification model is obtained by training using the fault identification model training method provided in any of the above embodiments.
[0090] Step 203: Use the first fault identification model and the second fault identification model to identify faults of the machine tool spindle.
[0091] It should be noted that the first fault identification model and the second fault identification model are only exemplary. When actually identifying and diagnosing the fault of the machine tool spindle, multiple other fault identification models can be combined to jointly identify the fault of the machine tool spindle. The training process of other fault identification models is similar to the training process of the second fault identification model. The difference is that the pre-trained models of each other fault identification model can be constructed according to the model parameters of the first fault identification model, and can also be constructed according to the model parameters of the second fault identification model.
[0092] Combine multiple fault identification models to diagnose and identify machine tool spindle faults, so as to timely discover machine tool spindle operation faults and take measures to avoid the hazards and losses caused by the faults.
[0093] Figure 2b A schematic diagram of the algorithm architecture of a fault identification method for a machine tool spindle provided as an exemplary embodiment of the present invention, in which a similar fault data set represents a data set of first fault sample data, a similar fault feature set represents a set of fault features extracted from the first fault sample data, a target fault data set represents a data set of second fault sample data, a target fault feature set represents a set of fault features extracted from the second fault sample data, real-time monitoring data represents the operating data of the machine tool spindle, and real-time monitoring data features represent features extracted from the operating data. It can be seen from the figure that the internal weights of the deep neural network are shared, thereby realizing knowledge accumulation and reuse across fault categories, and by storing and reusing knowledge across fault categories, the amount of data of training samples required for a specific type of fault identification model is reduced, thereby solving the industry pain point of difficulty in obtaining specific types of spindle fault data (training samples).
[0094] The historical monitoring data and fault history records (actual fault conditions) of a CNC machine tool in a workshop are used as training samples, and the fault features extracted from the historical monitoring data are used as Figure 2bThe input of the prediction model shown in the figure is to input the actual fault situation as the label result of the model, use random data partitioning to form training sets and test sets for various types of spindle faults, use range normalization as a means of dimensionless, fuse the features of data of different dimensions and scales, build a prediction model, and fine-tune the target fault through transfer learning, so that fault prediction of complex spindle systems can be achieved with only a small number of samples. The specific steps are as follows:
[0095] (1) Obtain historical monitoring data and corresponding fault history records of CNC machine tools in a workshop, where the monitoring data mainly includes vibration, temperature, and speed signals. After removing abnormal samples with empty signal values, the acquired data includes 1608 samples under normal working conditions, 331 samples under spindle imbalance fault conditions (Type A faults), and 666 samples under bearing wear fault conditions (Type B faults), totaling 2605 samples.
[0096] (2) Use order analysis and DB4 wavelet packet decomposition to extract frequency domain and time-frequency domain features from the vibration signal. The main input features include: 1. Time domain features - temperature and rotation speed; 2. Frequency domain features - the amplitude of the first and second frequencies of the vibration signal in three directions; 3. Time-frequency domain features - the low-pass energy proportion of the fourth-order sub-wavelet in three directions after wavelet packet decomposition, the fitting coefficients of the first to fourth-order sub-wavelets, and the high-pass energy proportion. Under the premise that vibration, rotation speed, and temperature signals can be obtained, a total of 35 feature columns (2 time domain features, 6 frequency domain features, and 27 time-frequency domain features) that can characterize the operating status of the equipment are obtained as the feature data set of historical records.
[0097] (3) The feature data set is labeled according to the fault history label, and the data set is divided by upsampling so that the number of normal samples and fault samples in the data set corresponding to each fault is consistent, avoiding the decrease in model accuracy due to sample imbalance. The data sample set D1 for Class A faults is obtained, with a total of 662 samples; the data sample set D2 for Class B faults is obtained, with a total of 1332 samples.
[0098] (4) Deep neural networks M1 and M2 for predicting class A and class B faults are constructed respectively, and their hyperparameters are set: the number of input layer neuron nodes is 35, and the number of hidden layer nodes is 32*32*16*16*8*2, that is, the first and second hidden layers contain 32 neurons, the third and fourth layers contain 16 neurons, and the fifth and sixth hidden layers contain 8 and 2 neurons respectively. At the same time, Adam and classification cross entropy functions are used as the training optimizer and loss function of the model respectively (the network structure and related parameters are obtained after a large number of experimental verifications. The deep neural network with this structure has good fault recognition accuracy and stability for the proposed machine tool spindle system fault recognition method).
[0099] (5) Using five-fold cross validation, 80% of the data was extracted from D1 and D2 to train models M1 and M2, and the remaining 20% of the data was used to test the model results. After repeating five times, the prediction accuracy of M1 and M2 before transfer learning was obtained. The results are as follows:
[0100]
[0101] (6) Reconstruct the transfer learning model M1' (the model structure is the same as M1) for predicting Class A faults. At this time, the model M2 for predicting Class B faults has been fully trained. The weight values of the neurons in each hidden layer of M2 are assigned to the hidden layer of M1', so that it inherits the relevant knowledge about Class B faults obtained by M2 through training.
[0102] (7) Use D1 to train the model M1' twice, so that it can perform self-learning based on a small amount of Class A fault data. During the learning process, fold cross validation is used, 80% of the data is extracted from D1 to train the model M1', and the remaining 20% of the data is used to test the model results. After repeating five times, the prediction accuracy of M1' after transfer learning is obtained. The results are as follows:
[0103]
[0104] Taking the spindle imbalance and spindle friction faults of the machine tool spindle system as examples, by collecting small sample operation data of the vibration simulation experimental platform under normal working conditions and two types of faults (A: spindle imbalance, B: bearing wear), deep neural network models M1 and M2 are constructed according to the A-type fault data and B-type fault data respectively for independent identification of the two types of faults. Among them, the recognition accuracy of M1 for A-type faults is 75.5%, and the recognition accuracy of M2 for B-type faults is 89.9%. According to transfer learning, the model M1' for A-type faults is reconstructed, and the weight value of the hidden layer neuron of M1' is shared with the neuron of M2. Based on M2, the data of A-type faults are used for secondary training. The experimental results show that after using the small sample machine tool spindle fault identification method based on transfer learning, the recognition accuracy of A-type faults can be increased from the original 75.5% to 99.4%, which meets the expected purpose and can be put into use.
[0105] Corresponding to the aforementioned training method for a fault identification model and the embodiment of the method for fault identification of a machine tool spindle, the present invention also provides embodiments of a training device for a fault identification model and a device for fault identification of a machine tool spindle.
[0106] Figure 3A module schematic diagram of a training device for a fault identification model provided by an exemplary embodiment of the present invention, wherein the fault identification model is used to identify a fault of a machine tool spindle, and the training device comprises:
[0107] A first training module 31 is used to train a first neural network using first fault sample data corresponding to a first fault type of the machine tool spindle to obtain a first fault recognition model for identifying the first fault type;
[0108] A construction module 32, configured to construct a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model;
[0109] The second training module 33 is used to train the pre-trained model using second fault sample data corresponding to the second fault type of the machine tool spindle to obtain a second fault identification model for identifying the second fault type.
[0110] Optionally, it also includes:
[0111] A first determination module is used to determine a fault feature corresponding to a first fault type, and perform order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain first fault sample data corresponding to the fault feature of the first fault type;
[0112] And / or, a second determination module is used to determine the fault characteristics corresponding to the second fault type, and perform order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain second fault sample data corresponding to the fault characteristics of the second fault type.
[0113] Optionally, the model parameters include weight values;
[0114] The building blocks are specifically used for:
[0115] The weight value of the pre-trained model is configured according to the weight value of the first fault identification model.
[0116] Optionally, it also includes:
[0117] A verification module is used to use the second fault sample data to train a second neural network to obtain a verification model of the second fault identification model, wherein the second neural network has the same network structure as the pre-trained model; and according to the verification model, a five-fold cross-validation is performed on the second fault identification model, and the second fault identification model that passes the verification is determined as the final second fault identification model.
[0118] Figure 4 A schematic diagram of a module of a machine tool spindle fault identification device provided by an exemplary embodiment of the present invention, the device comprising:
[0119] An acquisition module 41 is used to acquire the operation data of the machine tool spindle;
[0120] An input module 42, used to input the operation data into a first fault recognition model and a second fault recognition model; wherein the first fault recognition model is obtained by training a neural network according to first fault sample data corresponding to a first fault type of the machine tool spindle, and the second fault recognition model is obtained by training a pre-trained model according to second fault sample data corresponding to a second type of fault of the machine tool spindle, and the pre-trained model is constructed according to model parameters of the first fault recognition model;
[0121] The identification module 43 is used to perform fault identification on the machine tool spindle using the first fault identification model and the second fault identification model.
[0122] Optionally, the input module includes:
[0123] a determination unit, configured to respectively determine a fault feature corresponding to the first fault type and a fault feature corresponding to the second fault type;
[0124] a first input unit, configured to input first operating data in the operating data that matches a fault feature corresponding to the first fault type into the first fault identification model;
[0125] The second input unit is used to input second operating data in the operating data that matches the fault feature corresponding to the second fault type into the first fault identification model.
[0126] Optionally, the model parameters include weight values;
[0127] The first fault identification model has the same weight value as the pre-trained model.
[0128] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.
[0129] Figure 5 FIG. 1 is a schematic diagram of the structure of an electronic device according to an exemplary embodiment of the present invention, and shows a block diagram of an exemplary electronic device 50 suitable for implementing the embodiments of the present invention. Figure 5The electronic device 50 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0130] like Figure 5 As shown, the electronic device 50 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 50 may include, but are not limited to: at least one processor 51, at least one memory 52, and a bus 53 connecting different system components (including the memory 52 and the processor 51).
[0131] The bus 53 includes a data bus, an address bus, and a control bus.
[0132] The memory 52 may include a volatile memory, such as a random access memory (RAM) 521 and / or a cache memory 522 , and may further include a read-only memory (ROM) 523 .
[0133] The memory 52 may also include a program tool 525 (or utility) having a set (at least one) of program modules 524, such program modules 524 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.
[0134] The processor 51 executes various functional applications and data processing by running the computer program stored in the memory 52, such as the method provided in any of the above embodiments.
[0135] The electronic device 50 may also communicate with one or more external devices 54 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 55. Furthermore, the model-generated electronic device 50 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 56. As shown, the network adapter 56 communicates with other modules of the model-generated electronic device 50 via a bus 53. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the model-generated electronic device 50, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.
[0136] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.
[0137] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method provided in any of the above embodiments are implemented.
[0138] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.
[0139] In a possible implementation, the present invention can also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps of the thermal runaway warning method for the battery swap station battery described in Example 1.
[0140] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on a user device, partially on a user device, as an independent software package, partially on a user device and partially on a remote device, or entirely on a remote device.
[0141] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.
Claims
1. A method for identifying a fault of a machine tool spindle, characterized in that: include: Acquiring operation data of the machine tool spindle; The operation data is input into a first fault identification model and a second fault identification model; wherein the first fault identification model is obtained by training a neural network according to first fault sample data corresponding to a first fault type of the machine tool spindle, and the second fault identification model is obtained by training a pre-trained model according to second fault sample data corresponding to a second type of fault of the machine tool spindle, and the pre-trained model is constructed according to model parameters of the first fault identification model; wherein the final second fault identification model is determined by the second fault identification model that passes a five-fold cross validation of the second fault identification model according to a verification model; the verification model is obtained by training a second neural network using the second fault sample data; wherein the second neural network has the same network structure as the pre-trained model; Performing fault identification on the machine tool spindle using the first fault identification model and the second fault identification model; The model parameters include weight values; The first fault identification model has the same weight value as the pre-trained model; The number of neuron nodes in the input layer of the neural network matches the number of fault features corresponding to the first fault type, and the number of neurons in each layer decreases layer by layer; The number of neuron nodes in the input layer of the pre-trained model matches the number of fault features corresponding to the second fault type.
2. The method for identifying a fault of a machine tool spindle according to claim 1, characterized in that: The inputting the operation data into the first fault identification model and the second fault identification model comprises: respectively determining a fault feature corresponding to the first fault type and a fault feature corresponding to the second fault type; inputting first operating data in the operating data that matches the fault feature corresponding to the first fault type into the first fault identification model; Second operating data in the operating data that matches the fault feature corresponding to the second fault type is input into the first fault identification model.
3. A method for training a fault identification model, characterized in that: The fault identification model is used to identify the fault of the machine tool spindle, and the training method includes: Using first fault sample data corresponding to a first fault type of the machine tool spindle, training a first neural network to obtain a first fault identification model for identifying the first fault type; Constructing a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model; Using second fault sample data corresponding to a second fault type of the machine tool spindle, training the pre-trained model to obtain a second fault identification model for identifying the second fault type; Using the second fault sample data, training a second neural network to obtain a verification model of the second fault identification model, wherein the second neural network has the same network structure as the pre-trained model; Performing a five-fold cross validation on the second fault identification model according to the validation model; Determine the verified second fault identification model as the final second fault identification model; The model parameters include weight values; The step of constructing a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model comprises: Configuring a weight value of the pre-trained model according to a weight value of the first fault identification model; The number of neuron nodes in the input layer of the neural network matches the number of fault features corresponding to the first fault type, and the number of neurons in each layer decreases layer by layer; The number of neuron nodes in the input layer of the pre-trained model matches the number of fault features corresponding to the second fault type.
4. The method for training a fault identification model according to claim 3, characterized in that: Also includes: determining a fault signature corresponding to a first fault type; Performing order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain first fault sample data corresponding to the fault characteristics of the first fault type; and / or, determining a fault characteristic corresponding to a second fault type; Performing order analysis and wavelet packet decomposition on the operation data of the machine tool spindle to obtain second fault sample data corresponding to the fault feature of the second fault type.
5. A machine tool spindle fault identification device, characterized in that: include: An acquisition module, used for acquiring the operation data of the machine tool spindle; An input module, used for inputting the operation data into a first fault recognition model and a second fault recognition model; wherein the first fault recognition model is obtained by training a neural network according to first fault sample data corresponding to a first fault type of the machine tool spindle, and the second fault recognition model is obtained by training a pre-trained model according to second fault sample data corresponding to a second type of fault of the machine tool spindle, and the pre-trained model is constructed according to the model parameters of the first fault recognition model; wherein the final second fault recognition model is determined by the second fault recognition model that passes a five-fold cross validation of the second fault recognition model according to a verification model; the verification model is obtained by training a second neural network using the second fault sample data; wherein the second neural network has the same network structure as the pre-trained model; an identification module, configured to perform fault identification on the machine tool spindle using the first fault identification model and the second fault identification model; The model parameters include weight values; The first fault identification model has the same weight value as the pre-trained model; The number of neuron nodes in the input layer of the neural network matches the number of fault features corresponding to the first fault type, and the number of neurons in each layer decreases layer by layer; The number of neuron nodes in the input layer of the pre-trained model matches the number of fault features corresponding to the second fault type.
6. A training device for a fault identification model, characterized in that: The fault identification model is used to identify the fault of the machine tool spindle, and the training device includes: A first training module, used to train a first neural network using first fault sample data corresponding to a first fault type of the machine tool spindle, to obtain a first fault recognition model for identifying the first fault type; A construction module, used to construct a pre-trained model of a second fault identification model according to the model parameters of the first fault identification model; A second training module is used to train the pre-trained model using second fault sample data corresponding to a second fault type of the machine tool spindle to obtain a second fault recognition model for identifying the second fault type; a verification module, configured to train a second neural network using the second fault sample data to obtain a verification model of the second fault identification model, wherein the second neural network has the same network structure as the pre-trained model; perform a five-fold cross-validation on the second fault identification model according to the verification model; and determine the second fault identification model that passes the verification as the final second fault identification model; The model parameters include weight values; The construction module is further used to configure the weight value of the pre-trained model according to the weight value of the first fault identification model; The number of neuron nodes in the input layer of the neural network matches the number of fault features corresponding to the first fault type, and the number of neurons in each layer decreases layer by layer; The number of neuron nodes in the input layer of the pre-trained model matches the number of fault features corresponding to the second fault type.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 4 is implemented.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.
