CN114676716A

CN114676716A - Method, device and medium for predicting residual life of tool

Info

Publication number: CN114676716A
Application number: CN202011430089.7A
Authority: CN
Inventors: 陈鑫; 戴永恒; 王磊; 马丽丽; 洪煦; 武少波; 王鹏达
Original assignee: Diankeyun Beijing Technology Co ltd
Current assignee: Diankeyun Beijing Technology Co ltd
Priority date: 2020-12-09
Filing date: 2020-12-09
Publication date: 2022-06-28

Abstract

The invention provides a method, a device and a medium for predicting the residual life of a cutter, wherein the method comprises the following steps: collecting a cutter state signal; obtaining a training set containing training samples and a testing set containing testing samples based on the collected tool state signals; extracting training samples from a training set, inputting the training samples into an initialized mixed model comprising a convolution depth confidence network CDBN model and a bidirectional long-short term memory (BilTM) model, and training the mixed model; the CDBN model comprises a multi-layer convolution limited Boltzmann machine (CRBM), training samples in the training set are firstly input into the CDBN model, and the output of the CDBN model is input into the BiLSTM model; and inputting the test sample into the trained mixed model, and outputting a prediction result of the residual service life of the cutter. The CDBN-BilSTM-based tool residual life prediction method, device and medium have self-learning capability, can greatly improve the intelligence of prediction, and improve the prediction precision of the tool residual life.

Description

Method, device and medium for predicting residual life of tool

Technical Field

The invention relates to the technical field of fault diagnosis, in particular to a tool residual life prediction technology, and more particularly to a tool residual life prediction method, a device and a medium based on a CDBN-BilSTM model.

Background

In the machining process, the cutting machining is the most main machining mode for part forming, and the abrasion state of a cutter directly influences the machining precision, the surface quality and the production efficiency of parts. Therefore, the prediction of the wear state and the residual service life of the cutter has very important significance for ensuring the machining quality and realizing continuous automatic machining.

The tool wear monitoring method is divided into a direct measuring method and an indirect measuring method, the direct measuring method adopts an optical measuring method, a resistance measuring method, a ray measuring method and the like to directly obtain the wear value of the tool, but the tool wear state in the machining process cannot be detected in real time due to the interference of cooling liquid and the like in the production process, and the tool wear monitoring method is less in application in actual industrial production; the indirect measurement method is that a sensor acquires signals in real time in the cutting process of the cutter, and machine learning models such as back propagation, a support vector machine and the like are adopted to monitor the abrasion loss of the cutter after data processing and characteristic extraction. But traditional machine learning models suffer from instability in the quality of the extracted features. Meanwhile, a shallow layer structure algorithm is adopted for training a sample, the convergence rate of the model is not controllable, and the model is easy to fall into a local optimal solution, so that an effective diagnosis model is difficult to establish by using the traditional mode identification method, and development and research of a cutter wear diagnosis algorithm and a model with self-learning capability are urgently needed.

Disclosure of Invention

In view of the problems in the prior art, the invention provides a method, a device and a medium for predicting the residual life of a tool based on CDBN-BilSTM, which have self-learning capability and can greatly improve the intelligence of prediction.

According to an aspect of the present invention, there is provided a tool remaining life prediction method, including the steps of:

collecting a cutter state signal;

obtaining a training set containing training samples and a testing set containing testing samples based on the collected tool state signals;

extracting training samples from a training set, inputting the training samples into an initialized mixed model comprising a convolution depth confidence network CDBN model and a bidirectional long-short term memory (BilTM) model, and training the mixed model; the CDBN model comprises a multi-layer convolution limited Boltzmann machine (CRBM), training samples in the training set are firstly input into the CDBN model, and the output of the CDBN model is input into the BiLSTM model;

and inputting the test sample into the trained mixed model, and outputting a prediction result of the residual service life of the cutter.

In an embodiment of the present invention, the step of acquiring the tool state signal includes: collecting a cutter state signal through a vibration sensor; the tool state signal includes: the time domain information, the amplitude information, the time difference information or the frequency domain information of the vibration signal acquired by the vibration sensor.

In an embodiment of the present invention, the obtaining a training set including training samples and a test set including test samples based on the collected tool state signals includes: carrying out signal decomposition on the acquired cutter state signals by adopting continuous wavelet transform; a training set containing training samples and a test set containing test samples are obtained based on the decomposed signals.

In an embodiment of the present invention, the extracting training samples from the training set is input into an initialized hybrid model including a convolutional deep belief network CDBN model and a bidirectional long-short term memory (BiLSTM) model, and the training on the hybrid model includes: randomly extracting training samples from a training set; and inputting the extracted training samples into the mixed model, obtaining errors through forward propagation, and realizing fine adjustment of model parameters through backward propagation until the training is finished.

In an embodiment of the invention, each layer of CRBM includes a convolutional neural network CNN and a restricted boltzmann machine RBM, which includes a visible layer and a hidden layer.

In an embodiment of the present invention, the BiLSTM model includes: one or more BilTM layers, each BilTM layer comprising a forward LSTM and a backward LSTM, each of the forward LSTM and the backward LSTM comprising a forgetting control gate, an input control gate, and an output control gate.

In one embodiment of the invention, a Dropout layer is added into the BilSTM model; introducing a lucky RELU activation function in a full-link layer of the BilSTM model.

In one embodiment of the present invention, the first and second electrodes are,

the expression of the wavelet transform is as follows:

where a is the scale, τ is the translation, t is the time, WT_f(a, τ) is the wavelet transform coefficient, ψ is the wavelet basis, and the wavelet transform has two variables: the scale a and the amount of translation τ. The scale a controls the expansion and contraction of the wavelet function, and the translation amount tau controls the translation of the wavelet function. The scale corresponds to frequency (inverse ratio) and the amount of translation τ corresponds to time.

In another aspect of the present invention, there is also provided a device for predicting remaining life of a tool, the device including a processor and a memory, the memory having stored therein computer instructions, the processor being configured to execute the computer instructions stored in the memory, and the device implementing the steps of the method when the computer instructions are executed by the processor.

In another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the method as set forth above.

The method and the device for predicting the residual life of the cutter based on the CDBN-BilSTM model are based on a mixed model of a Convolutional Deep Belief Network (CDBN) and a bidirectional Long Short Term Memory network (Bi-directional Long Short-Term Memory, BilSTM), have self-learning capability, greatly improve the 'intelligence' of prediction, reduce the time complexity and the space complexity of a model algorithm, and improve the stability of a predicted value and the prediction precision of the residual life.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

It will be appreciated by those skilled in the art that the objects and advantages that can be achieved with the present invention are not limited to the specific details set forth above, and that these and other objects that can be achieved with the present invention will be more clearly understood from the detailed description that follows.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention.

FIG. 1 is a schematic flow chart of a tool remaining life prediction method based on a CDBN-BilSTM model according to an embodiment of the invention.

FIG. 2 is a schematic flow chart of a tool remaining life prediction method based on a CDBN-BilSTM model according to another embodiment of the invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the following embodiments and accompanying drawings. The exemplary embodiments and descriptions of the present invention are provided to explain the present invention, but not to limit the present invention.

It should be noted that, in order to avoid obscuring the present invention with unnecessary details, only the structures and/or processing steps closely related to the scheme according to the present invention are shown in the drawings, and other details not so relevant to the present invention are omitted.

It should be emphasized that the term "comprises/comprising" when used herein, is taken to specify the presence of stated features, elements, steps or components, but does not preclude the presence or addition of one or more other features, elements, steps or components.

The method aims to solve the problem that an effective diagnosis model is difficult to establish by a traditional tool life prediction method. The invention provides a novel cutter life prediction method, which creatively provides a Convolution Depth Belief Network (CDBN) -bidirectional Long Short Term Memory (Bi-directional Long Short Term Memory, BilSTM) mixed model, wherein the CDBN-BilSTM model is a mixed model based on a CDBN model and a BilSTM network model, the mixed model realizes automatic filtering of a signal end through a Convolutional Neural Network (CNN) and remarkably improves the stability of a predicted value. And then embedding the BilSTM model into the CDBN, establishing a time sequence relation between signal characteristics in two directions, and operating in two directions to strengthen the memory function of the sequence and further improve the prediction precision of the residual service life of the cutter. The model has self-learning capability, the 'intelligence' of prediction is greatly improved, the time complexity and the space complexity of a model algorithm are reduced, and the stability of a predicted value and the prediction precision of the residual life are improved.

The convolutional Deep Belief Network CDBN is a model formed by integrating a Convolutional Neural Network (CNN) into a Deep Belief Network (DBN). The DBN is a model adopting an unsupervised greedy layer-by-layer training method, can avoid manual operation of feature extraction and selection, has the capability of processing high-dimensional and nonlinear data, and can effectively prevent the problems of dimension disaster and the like, so that the method is applied to the fault diagnosis problem of the cutter in the new period of industrial big data era. The DBN can directly process original data, a machine can automatically learn signal characteristics, and the processes of characteristic extraction and selection are omitted, so that the method has great advantages in the aspects of processing high-dimensional and nonlinear data and the like, and the intelligence of residual life prediction is improved. However, the output of a conventional DBN is less resistant to "noise". Therefore, the convolutional neural network CNN is fused into the DBN, and the automatic filtering of the signal end is realized through the CNN, so that the stability of the predicted value is obviously improved.

Furthermore, the invention embeds a bidirectional long-short term memory (BilSTM) model into the CDBN, and more finely reflects the processing conditions of the cutter in different time periods by bidirectionally establishing a time sequence relation among signal characteristics, thereby improving the prediction precision of the residual life of the cutter.

The tool remaining life prediction method will be described in detail below with reference to the drawings.

Fig. 1 is a schematic flow chart of a tool remaining life prediction method based on a CDBN-BiLSTM model in an embodiment of the present invention, as shown in fig. 1, the method includes the following steps:

step S110: and collecting a cutter state signal.

The tool in the embodiment of the present invention may include turning tools, milling tools, boring tools, drill bits, etc., but is not limited thereto. The step can acquire the state signal of the cutter through the vibration sensor.

In the aspect of mounting the vibration sensor: the vibration sensor is directly installed at the rear end of the main shaft through a screw, is tightly attached to the main shaft and ensures no axial and radial play.

In the aspect of signal acquisition: aiming at the residual life prediction of the cutter, because the vibration sensor has the advantages of high measurement precision, wide frequency and amplitude response, easiness in installation and monitoring and the like, the cutter state signal is acquired through the vibration sensor, and generally comprises time domain information (such as an average value, a mean square value, a root mean square value, a variance and the like), amplitude information (such as probability distribution, probability density and the like), time difference information (such as an autocorrelation function, a cross-correlation function and the like) or frequency domain information (such as self-power spectral density, cross-power spectral density, a frequency response function and the like) of the vibration signal. The invention focuses on the research based on frequency domain signals collected in the process of machining the cutter.

Step S120: a training set containing training samples and a test set containing test samples are obtained based on the acquired tool state signals.

As an example, this step may include: carrying out signal decomposition on the acquired cutter state signals by adopting continuous wavelet transform; a training set containing training samples and a test set containing test samples are obtained based on the decomposed signals.

In some embodiments, the acquired tool state signals may be preprocessed before signal decomposition of the tool state signals by using continuous wavelet transform, and the preprocessing may include filtering and denoising, because the quality of the denoising effect of the original signals directly affects the tool wear state prediction accuracy. Because the collected signals directly come from the cutting working area, the actually collected data have the defects of unequal length, large noise interference, data redundancy and the like, and the original signals generally need to be denoised. The noise reduction of the original signal can improve the prediction precision of the tool wear state. Of course, the pre-treatment step is an optional step and is not necessary.

After signal preprocessing, the signal decomposition is selected to be carried out on all preprocessed signals through continuous wavelet transform. The continuous wavelet transform is proposed by Y.Meyer and S.Mallat, can automatically adjust the size of a window according to the signal acquisition frequency, is a self-adaptive time-frequency analysis method, has the advantages of well finding the local time-frequency characteristics of a non-stationary signal, namely a vibration signal, and is integral transform according to the definition of the continuous wavelet transform. The continuous wavelet transform expression is as follows:

where a is the scale, τ is the translation, t is the time, WT_f(a, τ) is the wavelet transform coefficient, ψ is the wavelet basis, and the wavelet transform has two variables: scale a and amount of translation τ. The scale a controls the expansion and contraction of the wavelet function, and the translation amount tau controls the translation of the wavelet function. The scale corresponds to frequency (inverse ratio) and the amount of translation τ corresponds to time. Since the wavelet basis psi has two parameters of scale and displacement, it will expand at the wavelet basis psiMeaning that a function of time is projected onto a two-dimensional time-scale phase plane. Based on the characteristics of the wavelet base, the function is beneficial to extracting the characteristics after being projected to a wavelet transform domain. Meanwhile, the original time sequence is decomposed into a series of superposition of fundamental wavelet, so that a time-frequency signal with good time domain and frequency domain localization is constructed, and the abrasion conditions of different cutting edges of the cutter can be judged by analyzing the signal, thereby laying a good foundation for the residual life prediction of the cutter. As will be described in the following steps, a DBN and CNN-based hybrid model (i.e. CDBN) is then applied to realize feature extraction and information filtering of the tool state signals, and then a BilSTM model is used to bidirectionally establish a timing relationship between signal features, so that the operating conditions and internal associations of the tool in different time periods are more finely reflected, and the prediction accuracy of the residual life is improved.

After wavelet transform processing is performed on the tool state signal, a sample set may be obtained based on the transformed signal, and the sample set may include a training set and a test set, the training set including training samples, and the test set including test samples. Wherein the training samples are derived based on the collected state signals of the tool and known remaining life data of the tool. In an embodiment of the present invention, the training set and the test set may be divided according to a predetermined ratio to form a sample set.

Step S130: training samples are extracted from the training set and input into an initialized hybrid model comprising a CDBN model and a BilSTM model, and the hybrid model is trained.

The mixed model can be called a CDBN-BilSTM model or a CDBN-BilSTM mixed model, wherein the CDBN model comprises a multilayer convolution limited Boltzman machine CRBM (in the embodiment of the invention, the CDBN comprises 4 layers of convolution limited Boltzman machines CRBM), training samples in a training set are firstly input into the CDBN model, and the output of the CDBN model is then input into the BilSTM model.

In one embodiment of the present invention, training samples may be randomly drawn from a training set. After the initialization of the CDBN-BilSTM model is completed, the data set of training samples may be imported into the CDBN-BilSTM model.

The CDBN is a layered probability generation model which is formed by stacking and combining a plurality of convolution limited Boltzmann machines (CRBMs), and the CRBMs are derived from a base model of the limited Boltzmann machines (RBMs). The RBM is a two-layer network model consisting of a visible layer and a hidden layer. Connection exists between each neuron node in adjacent layers, and a connectionless mode is adopted in each layer, which is a basic component of the DBN and is a key preprocessing unit in the field of deep learning. The CRBM is an idea of integrating a convolutional neural network on the basis of an original RBM and improves the idea, but the structure of the CRBM is similar to that of the RBM and adopts a layer-by-layer stacking method. The CRBM has the obvious characteristics that a local field (local field) and a weight are shared, a hidden layer and a visible layer are locally connected, and the weight is shared through convolution, so that the anti-noise capability of the model is enhanced. The CDBN model can automatically extract data features, and the deep architecture of the model can enable extraction of a full range signal by generating a convolution filter, which can reduce a large number of connection weights and can more efficiently learn useful information from neighboring elements. The embodiment of the invention innovatively applies the CDBN model in the prediction of the residual life of the cutter, and shows the good self-learning capability of the model while reducing noise and improving the prediction stability. Since the existing CDBN model can be used in the present invention, the CDBN model will not be described in detail herein.

For the BilSTM model, the model is a special recurrent neural network, and mainly aims to solve the problems of gradient elimination and gradient explosion in the long sequence training process. Briefly, the BilTM is composed of forward LSTM and backward LSTM, one or more BilTM layers can be included in the BilTM model, and in the embodiment of the invention, 2 BilTM layers are adopted, so that compared with a single-layer BilTM layer, the method can extract deeper features. Meanwhile, LSTM introduces the concept of three gates: the forgetting control gate, the input control gate and the output control gate, namely the forward LSTM and the backward LSTM can respectively comprise the forgetting control gate, the input control gate and the output control gate. The forgetting control gate is used for determining which information of the previous hidden layer state is important, the input control gate is used for determining which information of the current state is important, the output control gate is used for determining the next hidden layer state, the hidden layer is used for memorizing and storing the number of nodes of the past state, namely the hidden layer state is represented by selecting the number of the nodes, and the output number generated by each LSTM unit in each time step is represented by the value of the hidden state. LSTM can perform better in longer sequences. The BilSTM is a bidirectional cyclic neural network consisting of a forward LSTM and a backward LSTM, strengthens the time sequence relation between signals from two directions, can better extract vibration signals in a period, and memorizes and saves the signal characteristics of the whole process of tool machining. The BILSTM model is applied to predicting the residual life of the cutter, so that vibration signals in a period can be better extracted, and fault characteristics can be memorized and stored. Since the structure of the BilSTM model is existing, it will not be described in detail here. In the embodiment of the invention, the training sample is firstly input into the CDBN model of the mixed model to realize the feature extraction and information filtration of the cutter state signal, and then the output classification feature data of the CDBN model is directly input into the BilSTM model to establish the time sequence relation among the signal features, thereby reflecting the operation condition and the internal association of the cutter in different time periods more finely.

The CDBN-BilSTM mixed model is realized as follows: the CDBN fuses the thought of a convolutional neural network on the basis of a deep confidence network, the radiation range is expanded by adopting the cavity convolution, the aim of quickly extracting the global information of the signal is fulfilled, and then the fine extraction of the local features is realized by adopting the small convolution verification. And (3) performing multi-step convolution operation, down-sampling to an output layer to obtain classified feature data, and inputting the sequence with the extracted features into a BILSTM layer, wherein the number of the BILSTM layers has a large influence on the prediction accuracy of the residual life. The number of the layers of the BILSTM in the deeper layer has an obvious effect on the improvement of the service life prediction precision, but the number of the layers of the BILSTM is continuously increased, so that a serious overfitting problem exists, and due to the consideration, a Dropout layer is added in the full connection layer behind the BILSTM to reduce the overfitting and improve the generalization capability of the model. In addition, the BILSTM model employs a Leaky ReLU activation function at the fully connected layer, which exhibits good performance by assigning a non-zero slope to all negative values. By setting a small constant, the value of the negative axis is preserved so that the information of the negative axis is not lost completely. And finally realizing the prediction of the residual life of the cutter through continuous training and iteration of the model.

In the model training stage, training errors are obtained through forward propagation, fine adjustment of parameters is achieved through backward propagation, iteration termination conditions are judged to be met after the model is converged, model training is finished after the iteration termination conditions are met, and otherwise, the model training is repeatedly circulated.

Step S140: and inputting the test sample into the trained mixed model, and outputting a prediction result of the residual service life of the cutter.

In the embodiment of the invention, the test sample is input into the trained CDBN-BilSTM mixed model, so that the prediction result of the residual life of the cutter is obtained.

The cutter is divided into types such as lathe tool, milling cutter, boring cutter, drill bit, and the frequency spectrum that different types of cutter wearing and tearing state corresponds when processing different material parts often differs great, and the characteristic of signal is drawed work load very greatly, and traditional shallow neural network often can meet the dimension disaster when handling the signal. Meanwhile, the time sequence of the signal is crucial to the residual life prediction analysis, and accurate prediction cannot be performed without strong memory capacity of previous and subsequent information. The invention innovatively fuses the CBDN model and the BilSTM model, firstly uses a mixed model (namely CDBN) based on the DBN and the CNN to realize the feature extraction and information filtration of the cutter state signal, and then establishes the time sequence relation among the signal features through the BilSTM model, thereby reflecting the operating conditions and the internal association of the cutter in different time periods more finely, effectively solving the difficulty in the residual life prediction of the cutter and improving the residual life prediction precision of the cutter.

In another embodiment of the present invention as shown in fig. 2, the step S120 specifically includes:

and step S121, performing continuous wavelet transformation on the cutter state signal obtained by testing, and obtaining a sample set comprising a training set and a testing set based on the transformed signal, wherein the training set comprises training samples, and the testing set comprises testing samples.

The aforementioned step S130 includes:

step S131, randomly extracting training samples in the training set, and inputting the training samples into the initialized CDBN-BilSTM model.

And step S132, the training result is transmitted forward to obtain a training error.

After the training samples in the training set are input into the CDBN-BilSTM model, the training samples are input into each layer of the CDBN model, and the classification result output from the CDBN model is input into the BilSTM model according to the characteristic sequence and is finally output. This is a forward propagation process, and since there is an error between the output result of the model and the actual result, the error between the output estimation result and the actual result is calculated, and the training error is obtained.

Step S133, the error is propagated in the reverse direction, and the gradient of each parameter is calculated. For the CDBN, each layer of CRBM network can only ensure that the weight in its own layer is optimal for the mapping of the feature vector of the layer, but not optimal for the mapping of the feature vector of the entire CDBN, so the back propagation network also propagates the error information from top to bottom to each layer of CRBM network, and fine-tunes the entire CDBN network.

And S134, adjusting the model parameters based on the calculated parameter gradient, thereby updating the CDBN-BilSTM model.

And step S135, after updating the CDBN-BilSTM model, judging whether an iteration termination condition is reached, namely the model converges. After the iteration termination condition is met, the model training is finished (step S136), otherwise, the steps S132-S135 are repeated to repeatedly circulate until the iteration termination condition is reached.

The step S140 includes:

step S141, inputting the test sample to the trained hybrid model.

Namely, the test samples in the test set are firstly input into the CDBN model, and the characteristic sequence output by the CDBN model is then input into the BilSTM model.

And step S142, outputting the prediction result of the residual service life of the cutter.

The method utilizes the CDBN-BilSTM model, can more finely reflect the wear conditions and the internal correlation of the cutter in different time periods, and improves the prediction precision of the residual life of the cutter.

In accordance with the foregoing method, the present invention further provides a device for predicting remaining life of a tool, the device comprising a processor and a memory, the memory storing computer instructions, the processor being configured to execute the computer instructions stored in the memory, and the device implementing the steps of the foregoing method when the computer instructions are executed by the processor.

The invention also relates to a storage medium on which a computer program is stored which, when executed by a processor, may implement various embodiments of the method of the invention, the storage medium may be a tangible storage medium such as an optical disc, a Random Access Memory (RAM), a memory, a Read Only Memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of tangible storage medium known in the art.

It is to be understood that the invention is not limited to the specific arrangements and instrumentality described above and shown in the drawings. A detailed description of known methods is omitted herein for the sake of brevity. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present invention are not limited to the specific steps described and illustrated, and those skilled in the art can make various changes, modifications and additions or change the order between the steps after comprehending the spirit of the present invention.

Those of ordinary skill in the art will appreciate that the various illustrative components, systems, and methods described in connection with the embodiments disclosed herein may be implemented as hardware, software, or combinations of both. Whether this is done in hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, plug-in, function card, or the like. When implemented in software, the elements of the invention are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted by a data signal carried in a carrier wave over a transmission medium or a communication link. A "machine-readable medium" may include any medium that can store or transfer information. Examples of a machine-readable medium include electronic circuits, semiconductor memory devices, ROM, flash memory, Erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, Radio Frequency (RF) links, and so forth. The code segments may be downloaded via computer networks such as the internet, intranet, etc.

It should also be noted that the exemplary embodiments mentioned in this patent describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above-described steps, that is, the steps may be performed in the order mentioned in the embodiments, may be performed in an order different from the order in the embodiments, or may be performed simultaneously.

Features that are described and/or illustrated with respect to one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments in the present invention.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made to the embodiment of the present invention by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting remaining life of a tool, the method comprising:

collecting a cutter state signal;

and inputting the test sample into the trained mixed model, and outputting a prediction result of the residual life of the cutter.

2. The method of claim 1,

the step of acquiring a tool state signal comprises: collecting a cutter state signal through a vibration sensor; the tool state signal includes: the time domain information, the amplitude information, the time difference information or the frequency domain information of the vibration signal acquired by the vibration sensor.

3. The method of claim 1, wherein obtaining a training set containing training samples and a test set containing test samples based on the collected tool state signals comprises:

carrying out signal decomposition on the acquired cutter state signals by adopting continuous wavelet transform;

a training set containing training samples and a test set containing test samples are obtained based on the decomposed signals.

4. The method of claim 1, wherein the extracting training samples from the training set is input into an initialized hybrid model comprising a Convolutional Deep Belief Network (CDBN) model and a bidirectional long-short term memory (BilTM) model, and the training of the hybrid model comprises:

randomly extracting training samples from a training set;

and inputting the extracted training samples into the mixed model, obtaining errors through forward propagation, and realizing fine adjustment of model parameters through backward propagation until the training is finished.

5. The method of claim 1, wherein each layer of CRBM comprises a convolutional neural network CNN and a restricted boltzmann machine RBM, the restricted boltzmann machine comprising a visible layer and a hidden layer.

6. The method of claim 1, wherein the BilSTM model comprises: one or more BilTM layers, each BilTM layer comprising a forward LSTM and a backward LSTM, each of the forward LSTM and the backward LSTM comprising a forgetting control gate, an input control gate, and an output control gate.

7. The method of claim 1, further comprising:

adding a Dropout layer into the BilSTM model;

and adding a full connection layer after the BilSTM model and introducing a Leaky RELU activation function.

8. The method of claim 3, wherein the wavelet transform is expressed as:

where a is scale, τ is translation, t is time, WT_f(a, τ) is the wavelet transform coefficient and ψ is the wavelet basis.

9. A tool remaining life prediction device comprising a processor and a memory, wherein the memory has stored therein computer instructions for executing the computer instructions stored in the memory, wherein the device when the computer instructions are executed by the processor implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 8.