Tool wear state prediction method, device and storage medium

Abstract

The present application relates to a method, device and storage medium for predicting tool wear status. The method comprises preprocessing an original tool wear data set to obtain a data training set and a data test set, using the data training set to perform model training on a pre-constructed original prediction model to obtain an optimized prediction model, and using the optimized prediction model to calculate and process a tool wear characteristic value sequence, a data position characteristic sequence and a data time characteristic sequence to obtain a prediction result, thereby effectively improving tool processing efficiency, and being able to complete the prediction of a complete processing cycle under the tool wear state, thereby improving the accuracy of tool wear data prediction and ensuring product quality.

Description

Technical Field

[0001] The present invention relates to the technical field of tool wear, and in particular to a tool wear state prediction method, device and storage medium. Background Art

[0002] At present, in the process of product processing, the state of tool wear directly affects the surface quality of the processed product, and also has a certain impact on the stability of product processing during the processing. Therefore, it is necessary to understand the wear state of the tool at any time, and then establish an accurate and reliable tool wear prediction model to monitor the tool wear state during the processing of the product, which can not only improve the utilization rate of the tool, but also effectively reduce the production and processing product cost. Therefore, it is very necessary to study an efficient and accurate tool wear prediction model.

[0003] However, existing research on tool wear prediction methods can be roughly divided into two categories: the first method is direct prediction, and the second method is indirect prediction. Direct prediction mainly classifies the tool wear status by using optical images (color camera sensors), but direct prediction takes into account the complexity of the processing environment, which makes it difficult to locate the installation position of the color camera, and taking pictures of the tool requires downtime, which will reduce the tool processing efficiency. Indirect prediction mainly collects cutting force, vibration, acoustic emission and other signal data for analysis and prediction, but this method is mostly based on single-point or short-term prediction levels, and rarely involves the prediction of a complete processing cycle under the tool wear state. Therefore, it is urgent to solve the problems existing in the above two tool wear prediction methods. Summary of the invention

[0004] In view of the problems existing in the background technology, the purpose of the present invention is to provide a tool wear state prediction method, device and storage medium, which can not only effectively improve the tool processing efficiency, but also complete the prediction of a complete processing cycle under the tool wear state, thereby improving the accuracy of tool wear data prediction and ensuring product quality.

[0005] In order to achieve the above object, the present invention provides a method for predicting tool wear state, the prediction method comprising the steps of:

[0006] Obtaining the original tool wear data set;

[0007] Preprocessing the original tool wear data set to obtain a data training set and a data test set;

[0008] Using the data training set to perform model training on the pre-built original prediction model to obtain an optimized prediction model;

[0009] The data in the data test set, the position corresponding to each data and the time corresponding to each data are respectively converted to obtain a tool wear characteristic value sequence, a data position characteristic sequence and a data time characteristic sequence;

[0010] The optimization prediction model is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence to obtain a prediction result.

[0011] Optionally, preprocessing the original tool wear data set to obtain a data training set and a data test set includes the following steps:

[0012] The original tool wear data set is standardized to obtain a standardized data set. The standardized processing algorithm is:

[0013]

[0014] Among them, x represents the original data of tool wear, represents the average value of the original tool wear data, σ represents the standard deviation of the original tool wear data, and z represents the standardized value of the tool wear data;

[0015] The standardized data set is divided into the data training set and the data test set.

[0016] Optionally, the original prediction model is trained using the data training set to obtain an optimized prediction model, comprising the steps of:

[0017] Determine the target prediction value;

[0018] Based on the relationship between the target predicted value and the actual predicted value, a model loss function is constructed;

[0019] Input the data in the data training set into the original prediction model in order of data sequence, and calculate the actual prediction value sequence;

[0020] The corresponding model loss values ​​are calculated respectively through the actual prediction value sequence and the constructed model loss function to obtain a model loss value sequence;

[0021] The parameters in the original prediction model are adjusted based on the model loss value sequence until the model loss value sequence converges, and the model corresponding to the model loss value sequence is the optimized prediction model.

[0022] Optionally, the original prediction model is trained using the data training set to obtain an optimized prediction model, further comprising the steps of:

[0023] Compare the convergence value of the model loss value sequence with the preset target convergence value: if the convergence value of the model loss value sequence is greater than the preset target convergence value, adjust the parameters in the model and continue training, otherwise terminate the training.

[0024] Optionally, data conversion is performed on the data in the data test set, the position corresponding to each data, and the time corresponding to each data to obtain a tool wear characteristic value sequence, a data position characteristic sequence, and a data time characteristic sequence, including the steps of:

[0025] Using a one-dimensional convolution function to perform convolution processing on the data test set to obtain a tool wear characteristic value sequence;

[0026] Using a coding function to perform position coding on the positions corresponding to each data in the data test set to obtain a data position feature sequence;

[0027] The time corresponding to each data in the data test set is encoded using a fully connected layer to obtain a data time feature sequence.

[0028] Optionally, the optimization prediction model is used to calculate and process the tool wear characteristic value sequence, the data position characteristic sequence and the data time characteristic sequence to obtain a prediction result, including the steps of:

[0029] Using the encoder in the optimization prediction model, the tool wear characteristic value sequence, the data position characteristic sequence and the data time characteristic sequence are calculated and processed to obtain the encoder output result;

[0030] Using the decoder in the optimization prediction model, the tool wear characteristic value sequence, the data position characteristic sequence, the data time characteristic sequence and the encoder output result are calculated and processed to obtain the decoder output result;

[0031] The decoder output result is processed using the connection layer function in the optimized prediction model to obtain a prediction result.

[0032] Optionally, the encoder in the optimization prediction model includes a first encoding layer, a first convolutional layer, a second encoding layer, a second convolutional layer, a third encoding layer and an encoding normalization layer. The first encoding layer, the first convolutional layer, the second encoding layer, the second convolutional layer, the third encoding layer and the encoding normalization layer calculate and process the tool wear feature value sequence, the data position feature sequence and the data time feature sequence in turn to obtain the encoder output result.

[0033] Optionally, the decoder in the optimization prediction model includes a first decoding layer, a second decoding layer and a decoding normalization layer, and the first decoding layer, the second decoding layer and the decoding normalization layer sequentially calculate and process the tool wear characteristic value sequence, the data position characteristic sequence, the data time characteristic sequence and the encoder output result to obtain the decoder output result.

[0034] The present application also provides a tool wear state prediction device, the prediction device comprising:

[0035] A data acquisition module, used to acquire original tool wear data set;

[0036] A preprocessing module, used for preprocessing the original tool wear data set to obtain a data training set and a data test set;

[0037] A model training module, used to perform model training on the original prediction model using the data training set to obtain an optimized prediction model;

[0038] A data conversion module, used to convert the data in the data test set, the position corresponding to each data, and the time corresponding to each data, to obtain a tool wear characteristic value sequence, a data position characteristic sequence, and a data time characteristic sequence;

[0039] The prediction module is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence using the optimization prediction model to obtain a prediction result.

[0040] The present application also provides a storage medium having computer executable instructions stored thereon. When the computer executable instructions are executed by a computing device, the storage medium is used to execute any one of the above-mentioned tool wear state prediction methods.

[0041] The beneficial effects of the present invention are as follows:

[0042] By preprocessing the original tool wear data set, a data training set and a data test set are obtained. The pre-built original prediction model is trained using the data training set to obtain an optimized prediction model. The tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence are calculated and processed using the optimized prediction model to obtain the prediction result, thereby effectively improving the tool processing efficiency. It can also complete the prediction of a complete processing cycle under the tool wear state, thereby improving the accuracy of tool wear data prediction and ensuring product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0045] Figure 1 It is a flow chart of a tool wear state prediction method of the present invention.

[0046] Figure 2 It is a prediction model diagram in a tool wear state prediction method of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0048] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0050] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0051] Please refer to the attached Figure 1 and attached Figure 2 The tool wear state prediction method according to the present application is described in detail.

[0052] The method comprises:

[0053] Step S10, obtaining an original tool wear data set.

[0054] Step S20, preprocessing the original tool wear data set to obtain a data training set and a data test set. Optionally, preprocessing the original tool wear data set also includes the steps of: regularizing the data in the original tool wear data set, supplementing missing data, and deleting abnormal data. Optionally, the data regularization of the tool wear data set is to integrate data from different time periods together, and unify data from different rows and columns into rows or columns, so as to facilitate data input, and it is not easy to have data errors, and it is easier to check whether the data is abnormal, so that the final training result is more accurate. Among them, data missing supplementation is to timely supplement the missing tool wear data, so as to avoid insufficient data input into the training model, so that the trained original prediction model predicts the tool wear state more accurately. In addition, data abnormality deletion is to delete the tool wear data if the difference is too large or too small, or to obtain new data again, so as to avoid inaccurate data input into the training model, so as to avoid inaccurate prediction of the tool wear state by the trained original prediction model.

[0055] Optionally, correlation analysis is performed on the preprocessed tool wear data to determine the tool wear characteristics, that is, data analysis of tool wear in different processing cycles is determined to determine the current tool wear characteristics, so that the final tool wear characteristics are input into the original training model.

[0056] Optionally, preprocessing the original tool wear data set to obtain a data training set and a data test set includes the following steps:

[0057] The original tool wear data set is standardized to obtain a standardized data set. The standardized processing algorithm is:

[0058]

[0059] Among them, x represents the original data of tool wear, represents the average value of the original tool wear data, σ represents the standard deviation of the original tool wear data, and z represents the standardized value of the tool wear data;

[0060] The standardized data set is divided to obtain the data training set and the data test set. The preprocessed tool wear data is then used in the subsequent input prediction model to avoid the existence of dimensions in the tool wear data (i.e., the size distribution of the tool wear data is consistent), which affects the accuracy of the prediction result of the tool wear state for the next use. Optionally, all standardized data are divided into 8:2 ratios, 8 of which are training sets and 2 are test sets. The original prediction model is trained with 8 sets of standardized training sets of tool wear data. When the tool wear data standardization converges, the trained original prediction model is tested with 2 sets as test sets. If the test is qualified, there is no need to retrain the prediction model. If the test is unqualified, the original prediction model is retrained.

[0061] Step S30, using the data training set to train the pre-built original prediction model to obtain an optimized prediction model. Optionally, the original prediction model can be an RD-Informer model, and the RD-Informer model is trained using a plurality of pre-configured training sets to obtain an optimized prediction model, wherein each of the training samples is pre-processed tool wear data. During the training process, if the original data is directly input into the model for training, it is likely to cause the gradient to disappear or explode during the calculation process, making it difficult for the model to converge. Therefore, the training sample is pre-processed tool wear data to avoid the existence of dimensions in the data, which affects the final prediction result, and obtain accurate tool wear prediction.

[0062] In one embodiment, the original prediction model is trained using the data training set to obtain an optimized prediction model, comprising the steps of:

[0063] Determine the target prediction value;

[0064] Based on the relationship between the target predicted value and the actual predicted value, a model loss function is constructed;

[0065] Input the data in the data training set into the original prediction model in order of data sequence, and calculate the actual prediction value sequence;

[0066] The corresponding model loss values ​​are calculated respectively through the actual prediction value sequence and the constructed model loss function to obtain a model loss value sequence;

[0067] The parameters in the original prediction model are adjusted based on the model loss value sequence until the model loss value sequence converges, and the model corresponding to the model loss value sequence is the optimized prediction model. Optionally, it converges to 0.

[0068] Optionally, by passing the input data (x t ,y t) to the original training model F θ In (y|x) (the training model formula that has been calculated), in the case of dropout twice, it is considered that the sample passes through two different models, namely the target prediction value and the actual prediction value, and the prediction output is and Then the loss is calculated as follows:

[0069]

[0070]

[0071] Loss = L mse +αL mserd

[0072] Where D(·) represents the MSE calculation, x and y represent the target prediction value and the actual prediction value respectively, and F represents a complete encoder-decoder model. θ (y|x) is the predicted value. By judging whether the loss value converges, it is determined whether the training model is qualified. Of course, the trained model needs to be tested. This application trains and verifies in the training model. The hyperparameter settings of the model are shown in Table 1 below:

[0073] Table 1

[0074]

[0075] Among them, the input step size represents the length of two processing periods, the prediction step size represents the next processing cycle, epoch is the number of iterations of model training, batch-size is the number of samples input to the model for each training, dropout represents the specified drop probability of 0.01, and randomly discards the output of the previous layer, the optimizer represents the network parameters used to update and calculate the network parameters that affect the model training and model output, so that they approach or reach the optimal value, thereby minimizing (or maximizing) the loss function. The present invention adopts Adam adaptive moment estimation, the initial learning rate represents the learning rate set at the beginning of the training model, and the learning rate is the step size of the weight update. The activation function represents allowing the neural network to learn nonlinear functions, and GELU is the Gaussian error linear unit activation function.

[0076] In one embodiment, the original prediction model is trained using the data training set to obtain an optimized prediction model, further comprising the steps of:

[0077] Compare the convergence value of the model loss value sequence with the preset target convergence value: if the convergence value of the model loss value sequence is greater than the preset target convergence value, adjust the parameters in the model and continue training, otherwise terminate the training. Optionally, the preset target convergence value is 0.048, preferably 0, and the trained model is tested, as shown in Table 2 below:

[0078] Table 2

[0079] epoch Train Loss Test Loss 1 0.116 0.065 2 0.060 0.060 3 0.042 0.054 4 0.036 0.052 5 0.030 0.051 6 0.028 0.048

[0080] Epoch represents the number of iterations of model training, Train Loss represents the training loss value, and Test Loss represents the test loss.

[0081] Step S40, data conversion is performed on the data in the data test set, the position corresponding to each data, and the time corresponding to each data to obtain a tool wear characteristic value sequence, a data position characteristic sequence, and a data time characteristic sequence. Further, the tool wear characteristic value sequence, the data position characteristic sequence, and the data time characteristic sequence are added to obtain the tool wear data value.

[0082] Optionally, data conversion is performed on the data in the data test set, the position corresponding to each data, and the time corresponding to each data to obtain a tool wear characteristic value sequence, a data position characteristic sequence, and a data time characteristic sequence, including the steps of:

[0083] The data test set is convolved using a one-dimensional convolution function to obtain a tool wear feature value sequence. Optionally, the one-dimensional convolution kernel is 3×1. The tool wear feature value sequence is the preprocessed tool wear data x t Perform a one-dimensional convolution operation, that is, the input data from C in Dimension is mapped to d dimension, and the one-dimensional convolution function is set as value embedding, then its output can be expressed as v out =value_embedding(x t ).

[0084] The encoding function is used to positionally encode the positions corresponding to each data in the data test set to obtain a data position feature sequence. Specifically, in the model, all elements are processed together. This approach can speed up the model training, but it will ignore the order of data elements. For time series data, the order between data is particularly important. In order to ensure that the order is not lost, the prediction model encodes each input position. The position encoding formula is as follows:

[0085] PE (pos,2i) =sin(pos / 100002i / d )

[0086] PE (pos,2i+1) =cos(pos / 10000 2i / d )

[0087] Among them, PE(pos,2i) and PE(pos,2i+1) represent even positions and odd positions respectively, pos represents the position of the element in the sequence, and i represents the dimension;

[0088] Let the position encoding function be position_embedding, then the final output can be expressed as:

[0089] p out =position_embedding(x t ).

[0090] The fully connected layer is used to encode the time corresponding to each data in the data test set to obtain the data time feature sequence. Specifically, for time series data, time information plays a very important role. For example, months, Sundays, holidays, etc. will have a certain impact on related activities. The specific method is to use a fully connected layer to encode the input timestamp x mark Mapped to d dimension to obtain time features.

[0091] Let the temporal encoding function be temporal_embedding, then its output can be expressed as:

[0092] t out =temporal_embedding(x mark ),

[0093] Then, the final tool wear data processing input can be expressed as:

[0094] x embedding =v out +p out +t out .

[0095] Step S50, using the optimization prediction model to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence to obtain a prediction result. Optionally, the optimization prediction model includes an encoder and a decoder, and the encoder encodes and calculates the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence, and inputs the final result of the encoder processing into the decoder for decoding processing to obtain the next tool wear prediction result.

[0096] Optionally, the encoder in the optimization prediction model is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence to obtain the encoder output result. Specifically, the encoder includes: a first encoding layer, a first convolutional layer, a second encoding layer, a second convolutional layer, a third encoding layer and an encoding normalization layer, namely, attn_layers:EncoderLayer layer, conv_layers:ConvLayer layer, attn_layers:EncoderLayer layer, conv_layers:ConvLayer layer, attn_layers:EncoderLayer layer and norm:LayerNorm layer, the first encoding layer, the first convolutional layer, the second encoding layer, the second convolutional layer, the third encoding layer and the encoding normalization layer sequentially calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence to obtain the encoder output result, that is, the first encoding layer is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence to obtain the first encoding layer output result;

[0097] The first convolution layer is used to perform calculation processing on the output result of the first encoding layer to obtain the output result of the first convolution layer;

[0098] The second coding layer is used to perform calculation processing on the output result of the first convolutional layer to obtain a second coding layer output result;

[0099] The second convolution layer is used to perform calculation processing on the output result of the second encoding layer to obtain the output result of the second convolution layer;

[0100] The third coding layer is used to perform calculation processing on the output result of the second convolutional layer to obtain the output result of the third coding layer;

[0101] The encoding normalization layer is used to perform calculation processing on the output result of the third encoding layer to obtain the encoder output result.

[0102] Specifically, in order to accurately and efficiently model the relationship between time series, the ProbSparse self-attention operation is used in both the encoder and decoder of the optimized prediction model. Specifically, the input encoded data is processed through the attn_layers:EncoderLayer layer and multiplied by W Q , W K and W V Three weight matrices, namely x embedding =v out +p out +t out Multiply by W Q , WK and W V Three weight matrices, Q (query), K (key) and V (value). ProbSparse self-attention first randomly samples K and obtains Then calculate each q i ∈Q About Sparse coefficients of a matrix:

[0103]

[0104] Among them, Q represents the query vector of tool wear points, K represents the “queried” vector, V represents the content vector (tool wear feature vector), and L K represents the length of K(key), d represents the dimension of the input of the encoded tool wear feature value, and q i represents the i-th row in Q, k j represents the j-th row in K, is in transposed form. Q(query) and K(key) are vectors for calculating tool wear Attention weights, and V represents the vector of tool wear features.

[0105] Specifically, by converting the i， ) is processed, in M(q i , ), select top-u queries Here u=clnL Q (c is a hyperparameter, L Q represents the length of Q(query)), then the process of ProbSparse self-attention can be expressed as:

[0106]

[0107] A(Q, K, V) is further calculated and processed in the encoder, and the final result is input into the decoder.

[0108] The decoder in the optimization prediction model is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence, data time characteristic sequence and the encoder output result to obtain the decoder output result. Optionally, the decoder in the optimization prediction model includes a first decoding layer, a second decoding layer and a decoding normalization layer, and the first decoding layer, the second decoding layer and the decoding normalization layer sequentially calculate and process the tool wear characteristic value sequence, data position characteristic sequence, data time characteristic sequence and the encoder output result to obtain the decoder output result, that is, the decoder includes layers:DecoderLayer layer, layers:DecoderLayer layer, norm:LayerNorm layer and Projection:Linear layer, and the connection layer (Linear layer) function in the optimization prediction model is used to calculate and process the decoder output result to obtain the prediction result.

[0109] Specifically, A(Q, K, V) is further calculated and processed in the encoder, and the final result is input to the decoder, that is, the input of the decoder can be expressed as in, is the start token, which is exactly the same as the encoder input. is a placeholder reserved for the prediction sequence, and its value is all zero. The decoder model is De, which is known. Input to De, then the predicted output can be expressed as: Finally, the prediction results are obtained.

[0110] In one embodiment, the encoder is used to capture the telecoupling relationship between the original tool data, and the decoder is used to achieve tool wear prediction for a long sequence.

[0111] The present application also provides a tool wear state prediction device, the prediction device comprising:

[0112] A data acquisition module, used to acquire original tool wear data set;

[0113] A preprocessing module, used for preprocessing the original tool wear data set to obtain a data training set and a data test set;

[0114] A model training module, used to perform model training on the original prediction model using the data training set to obtain an optimized prediction model;

[0115] A data conversion module, used to convert the data in the data test set, the position corresponding to each data, and the time corresponding to each data, to obtain a tool wear characteristic value sequence, a data position characteristic sequence, and a data time characteristic sequence;

[0116] The prediction module is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence using the optimization prediction model to obtain a prediction result.

[0117] The present application also provides a storage medium having computer executable instructions stored thereon. When the computer executable instructions are executed by a computing device, the storage medium is used to execute any one of the above-mentioned tool wear state prediction methods.

[0118] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A tool wear state prediction method, characterized in that: The prediction method comprises the steps of: Obtaining the original tool wear data set; Preprocessing the original tool wear data set to obtain a data training set and a data test set; Using the data training set to perform model training on the pre-built original prediction model to obtain an optimized prediction model; The data in the data test set, the position corresponding to each data and the time corresponding to each data are respectively converted to obtain a tool wear characteristic value sequence, a data position characteristic sequence and a data time characteristic sequence; The optimization prediction model is used to calculate and process the tool wear characteristic value sequence, the data position characteristic sequence and the data time characteristic sequence to obtain a prediction result; in, The data in the data test set, the position corresponding to each data and the time corresponding to each data are respectively converted to obtain a tool wear characteristic value sequence, a data position characteristic sequence and a data time characteristic sequence, including the steps of: Using a one-dimensional convolution function to perform convolution processing on the data test set to obtain a tool wear characteristic value sequence; Using a coding function to perform position coding on the positions corresponding to each data in the data test set to obtain a data position feature sequence; The time corresponding to each data in the data test set is encoded using a fully connected layer to obtain a data time feature sequence.

2. The tool wear state prediction method according to claim 1, characterized in that: Preprocessing the original tool wear data set to obtain a data training set and a data test set includes the following steps: The original tool wear data set is standardized to obtain a standardized data set. The standardized processing algorithm is: Among them, x represents the original data of tool wear, represents the average value of the original tool wear data, σ represents the standard deviation of the original tool wear data, and z represents the standardized value of the tool wear data; The standardized data set is divided into the data training set and the data test set.

3. The tool wear state prediction method according to claim 1, characterized in that: The original prediction model is trained using the data training set to obtain an optimized prediction model, comprising the steps of: Determine the target prediction value; Based on the relationship between the target predicted value and the actual predicted value, a model loss function is constructed; Input the data in the data training set into the original prediction model in order of data sequence, and calculate the actual prediction value sequence; The corresponding model loss values ​​are calculated respectively through the actual prediction value sequence and the constructed model loss function to obtain a model loss value sequence; The parameters in the original prediction model are adjusted based on the model loss value sequence until the model loss value sequence converges, and the model corresponding to the model loss value sequence is the optimized prediction model.

4. The tool wear state prediction method according to claim 3, characterized in that: The original prediction model is trained using the data training set to obtain an optimized prediction model, further comprising the steps of: Compare the convergence value of the model loss value sequence with the preset target convergence value: if the convergence value of the model loss value sequence is greater than the preset target convergence value, adjust the parameters in the model and continue training, otherwise terminate the training.

5. The tool wear state prediction method according to claim 1, characterized in that: The optimization prediction model is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence to obtain a prediction result, including the steps of: Using the encoder in the optimization prediction model, the tool wear characteristic value sequence, the data position characteristic sequence and the data time characteristic sequence are calculated and processed to obtain the encoder output result; Using the decoder in the optimization prediction model, the tool wear characteristic value sequence, the data position characteristic sequence, the data time characteristic sequence and the encoder output result are calculated and processed to obtain the decoder output result; The decoder output result is processed using the connection layer function in the optimized prediction model to obtain a prediction result.

6. The tool wear state prediction method according to claim 5, characterized in that: The encoder in the optimization prediction model includes a first encoding layer, a first convolutional layer, a second encoding layer, a second convolutional layer, a third encoding layer and an encoding normalization layer. The first encoding layer, the first convolutional layer, the second encoding layer, the second convolutional layer, the third encoding layer and the encoding normalization layer calculate and process the tool wear characteristic value sequence, the data position characteristic sequence and the data time characteristic sequence in turn to obtain the encoder output result.

7. The tool wear state prediction method according to claim 5, characterized in that: The decoder in the optimization prediction model includes a first decoding layer, a second decoding layer and a decoding normalization layer. The first decoding layer, the second decoding layer and the decoding normalization layer sequentially calculate and process the tool wear characteristic value sequence, the data position characteristic sequence, the data time characteristic sequence and the encoder output result to obtain the decoder output result.

8. A tool wear state prediction device, characterized in that: The prediction device comprises: A data acquisition module, used to acquire original tool wear data set; A preprocessing module, used for preprocessing the original tool wear data set to obtain a data training set and a data test set; A model training module is used to perform model training on the original prediction model using the data training set to obtain an optimized prediction model; The data conversion module is used to convert the data in the data test set, the position corresponding to each data, and the time corresponding to each data, to obtain a tool wear characteristic value sequence, a data position characteristic sequence, and a data time characteristic sequence. in, The data in the data test set, the position corresponding to each data and the time corresponding to each data are respectively converted to obtain a tool wear characteristic value sequence, a data position characteristic sequence and a data time characteristic sequence, including the steps of: Using a one-dimensional convolution function to perform convolution processing on the data test set to obtain a tool wear characteristic value sequence; Using a coding function to perform position coding on the positions corresponding to each data in the data test set to obtain a data position feature sequence; Using a fully connected layer to encode the time corresponding to each data in the data test set to obtain a data time feature sequence; The prediction module is used to calculate and process the tool wear characteristic value sequence, data position characteristic sequence and data time characteristic sequence using the optimization prediction model to obtain a prediction result.

9. A storage medium, characterized in that: Computer executable instructions are stored thereon, and when the computer executable instructions are executed by a computing device, they are used to execute any one of the above method claims 1-7.

Images