Method and equipment for monitoring wear state of tool under unknown working condition and medium

Abstract

The invention discloses an unknown working condition tool wear state monitoring method and device and a medium. The method comprises the steps that a source domain input sequence corresponding to a tool is constructed; inputting the source domain input sequence as a training sample into a preset wear monitoring model, and adaptively extracting wear degradation features in the training sample based on a multi-attention mechanism carried by a feature extractor in the wear monitoring model; calculating a domain difference metric loss for evaluating a feature distribution difference between the wear degradation features; calculating classification loss between the wear degradation characteristics; according to the domain difference measurement loss and the classification loss, comprehensive loss corresponding to the wear monitoring model is determined, iteration is carried out on the wear monitoring model according to the comprehensive loss until training is completed, and the wear monitoring model with generalization is obtained; and through the wear monitoring model, predicting a tool wear state under an unknown cutting working condition.

Description

Technical Field

[0001] The present application relates to the field of intelligent manufacturing technology, and specifically to a method, device and medium for monitoring the wear status of a tool under unknown working conditions. Background Art

[0002] As the main executor of cutting tasks, the performance of cutting tools directly affects the machining quality of workpieces and the production efficiency of enterprises. In the actual processing of products, cutting tools will inevitably wear out under the influence of various factors such as mechanical, thermal and chemical factors. Excessive wear not only reduces the surface quality of workpieces, but also leads to increased cutting loads, which has a negative impact on the machining accuracy of machine tools and even brings safety risks in extreme cases.

[0003] In order to deal with the above problems, two common tool replacement strategies are used in actual processing. One strategy is to replace the tool by experienced personnel based on subjective judgment of on-site factors such as cutting sound and vibration. This strategy places high demands on the processing experience of on-site personnel and is relatively random. Another strategy is to use batch tools to conduct a large number of tests to establish a statistical analysis model for tool wear durability, and then formulate a periodic tool replacement strategy based on the model. However, due to the differences between individual tools and the variability of the cutting environment, there is a deviation between the actual tool life and the theoretical derivation value, which will cause certain problems. If the tool is replaced too late, it is easy to fail, affecting the processing quality and bringing certain safety risks. If it is replaced too early, it will not be fully utilized, which will lead to increased production costs. Therefore, from the perspective of reducing costs, increasing efficiency and ensuring safety, it is necessary to conduct more intelligent and accurate tool wear monitoring during the processing process.

[0004] With the rapid development of technologies such as the Industrial Internet of Things and advanced sensors, intelligent perception on the machine tool side has gradually become prominent, and the way to obtain tool wear monitoring data during processing has also been continuously expanded. Data-driven methods have gradually become one of the main ways to implement tool wear monitoring. Among them, many methods use artificial intelligence algorithms such as deep learning to automatically extract features from a large amount of monitoring data and establish a mapping relationship model with tool wear. However, the cutting conditions in the actual processing process are complex and changeable, and the monitoring data often show characteristics such as data distribution offset and invisibility. When dealing with the above scenarios, the wear modeling method based on deep learning will greatly reduce the performance of wear state monitoring due to the large data distribution differences in the deployment of monitoring scenarios and model training stages. At the same time, the existing wear monitoring models ignore the temporal correlation between features and the cross-channel characteristics between multi-dimensional features when performing automatic feature extraction.

[0005] In addition, the wear monitoring method based on transfer learning relaxes the independent and identically distributed restrictions in deep learning. The existing methods use monitoring data of historical working conditions and target new working conditions to jointly train the model based on feature, parameter or model transfer, or domain adaptation, which improves its wear state monitoring performance under variable cutting conditions. The above methods still need to obtain monitoring data under the target new working conditions to participate in training during the model training stage. However, in actual industrial scenarios, data in many cutting conditions are invisible during the model training process, and even the cutting conditions are unknown, which seriously limits the wear state monitoring performance of the above transfer learning methods. Summary of the invention

[0006] In order to solve the above problems, this application proposes a method for monitoring tool wear status in unknown working conditions, including:

[0007] Generate a tool wear data set according to the monitoring signal and tool wear state generated by tool cutting under historical working conditions, and map the tool wear data set to a corresponding state space, so as to construct a source domain input sequence corresponding to the tool according to the tool wear data set in the state space;

[0008] Inputting the source domain input sequence as a training sample into a preset wear monitoring model, and adaptively extracting wear degradation features in the training sample based on a multi-attention mechanism of a feature extractor in the wear monitoring model;

[0009] Performing metric alignment on the wear degradation features in different historical working conditions to calculate the metric distance between the wear degradation features, and calculating the domain difference metric loss for evaluating the feature distribution difference between the wear degradation features according to the metric distance;

[0010] Inputting the wear degradation features into a parameter-sharing wear state classifier to obtain a corresponding wear state classification result, and calculating the classification loss between the wear degradation features for the wear state classification result;

[0011] Determine the comprehensive loss corresponding to the wear monitoring model according to the domain difference metric loss and the classification loss, iterate the wear monitoring model according to the comprehensive loss until the training is completed, and obtain the wear monitoring model with generalization;

[0012] The wear monitoring model is used to predict the tool wear state under unknown cutting conditions.

[0013] In one implementation of the present application, a tool wear data set is generated based on the monitoring signal generated by tool cutting under historical working conditions and the tool wear state, specifically including:

[0014] Collect monitoring signals generated by tool cutting under historical working conditions, and detect the wear band width corresponding to the back face of the tool through a preset measuring device, so as to determine the tool wear state corresponding to the tool according to the wear band width;

[0015] According to the tool wear state, generating a corresponding wear state label, and associating the wear state label with the monitoring signal;

[0016] Preprocessing the monitoring signal, and performing sliding window segmentation on the preprocessed monitoring signal to obtain a plurality of segmented signals;

[0017] For the segmented signal, feature extraction is performed on the data of each channel in the segmented signal, and the extracted features are spliced ​​according to the timing direction of the monitoring signal to generate a corresponding multi-channel feature time series;

[0018] The multi-channel feature time series is normalized to obtain a tool wear data set.

[0019] In one implementation of the present application, based on the multi-attention mechanism of the feature extractor in the wear monitoring model, the wear degradation features in the training sample are adaptively extracted, specifically including:

[0020] The feature extractor includes a Transformer encoder, a channel attention layer, and an attention statistics pool module;

[0021] Based on the Transformer encoder in the feature extractor, adaptively assigning corresponding first feature weights to the feature matrix of multiple channels in the source domain input sequence; wherein the first feature weight refers to the weight of the global eigenvalue in the feature matrix;

[0022] Based on the channel attention layer, a squeezing operation and an excitation operation are respectively performed on the feature matrix through a global average pool and a gate mechanism to calibrate feature channel weights corresponding to multiple channels in the feature matrix;

[0023] Multiplying the feature channel weights by the feature matrix corresponding to the channel to obtain a feature map based on the calibrated feature channel weights;

[0024] Based on the attention statistics pool module, by capturing the importance of features in each channel at different times, the second feature weight corresponding to each feature in the channel is adjusted, and the feature map is reduced in dimension to map the feature map to the hidden layer feature space to obtain the wear and degradation feature; wherein the second feature weight refers to the weights corresponding to multiple features in a single channel along the time series direction in the feature map.

[0025] In one implementation of the present application, the wear degradation features in different historical working conditions are metrically aligned to calculate the metric distance between the wear degradation features. The specific calculation method is:

[0026]

[0027] In the formula, represents the p-order DSWD calculation, For the corresponding Wasserstein metric calculation, Const is a regularization constant greater than zero that satisfies the condition in the unit sphere The set of probability measures r on , satisfy From the unit sphere arrive The set of all Borel measurable functions of , is the Radon transform operator, f b is a Borel measurable function, is the gradient of the deep neural network with parameter r, λ Const is the regularization hyperparameter in model training.

[0028] In one implementation of the present application, based on the metric distance, a domain difference metric loss for evaluating the feature distribution difference between the wear degradation features is calculated, and the specific calculation method is:

[0029]

[0030] Where M represents the number of source domains, and They are the wear degradation characteristics under different historical working conditions.

[0031] In one implementation of the present application, the classification loss between the wear degradation features is calculated for the wear state classification result, and the specific calculation method is:

[0032]

[0033] Among them, F φ (x i ) is the wear degradation characteristic, y i For wear status label, T θ is the wear state classifier.

[0034] In one implementation of the present application, the comprehensive loss corresponding to the wear monitoring model is determined according to the domain difference measurement loss and the classification loss, specifically including:

[0035] Determining a loss function weight parameter corresponding to the domain difference metric loss;

[0036] The domain difference metric loss is weighted by the loss function weight parameter, and the weighted domain difference metric loss and the classification loss are added to obtain the comprehensive loss corresponding to the wear monitoring model.

[0037] In one implementation of the present application, before reducing the dimension of the feature map, the method further includes:

[0038] Performing a residual connection on the feature map and the feature matrix corresponding to the source domain input sequence to obtain the feature map output by the channel attention layer;

[0039] The residual connection is expressed as:

[0040] F mid =X+f SE (f TE (X)

[0041] In the formula, f SE 、f TE They represent the squeeze excitation operation and the Transformer encoder operation respectively, and X is the feature matrix.

[0042] The present application provides a device for monitoring tool wear in unknown working conditions, the device comprising:

[0043] at least one processor;

[0044] and, a memory communicatively coupled to the at least one processor;

[0045] Wherein, the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method for monitoring tool wear status in an unknown working condition as described in any of the above items.

[0046] The embodiment of the present application provides a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured as follows:

[0047] A method for monitoring tool wear status in unknown working conditions as described in any of the above items.

[0048] The unknown working condition tool wear state monitoring method proposed in this application can bring the following beneficial effects:

[0049] Based on the feature extractor of multi-attention mechanism, the temporal and cross-channel correlations between extracted features are enhanced. In addition, the wear monitoring model does not need to access the target cutting condition data during the construction process. It directly uses multiple historical conditions (i.e., source domain) data to learn generalized representations, and performs cross-domain distribution alignment of features between source domains through metric alignment, thereby effectively generalizing the monitoring model to unknown cutting scenario tasks, effectively improving the performance of tool wear state recognition under unknown cutting conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0051] Figure 1 A schematic flow chart of a method for monitoring tool wear status in an unknown working condition provided in an embodiment of the present application;

[0052] Figure 2 A schematic diagram of a multi-source domain generalization method integrating multi-attention mechanism feature extraction provided in an embodiment of the present application;

[0053] Figure 3 A schematic diagram of an experimental device provided in an embodiment of the present application;

[0054] Figure 4 A schematic diagram of the structure of a feature extractor provided in an embodiment of the present application;

[0055] Figure 5 A schematic diagram of the architecture of a wear monitoring model provided in an embodiment of the present application;

[0056] Figure 6 A schematic diagram of a DSWD measurement principle provided in an embodiment of the present application;

[0057] Figure 7 A schematic diagram of the structure of a tool wear status monitoring device for unknown working conditions provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only 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 the present application.

[0059] By using deep learning or transfer learning, the wear monitoring model is trained using sensor monitoring data, achieving better monitoring performance in multiple cutting conditions or variable cutting conditions, but there are still great limitations. The above method requires the use of data from the target monitoring conditions to participate in model training, and when automatically extracting features from multi-channel sensor data, it ignores the temporal correlation between features and the cross-channel characteristics between multi-dimensional features, resulting in poor monitoring performance in unknown cutting conditions where data cannot be obtained, and the generalization performance needs to be improved.

[0060] The embodiment of the present application proposes a multi-source domain generalization method that integrates multi-attention mechanism feature extraction, which is used for tool wear status monitoring under unknown cutting conditions. Based on the idea of ​​multi-source domain generalization, a generalized wear monitoring method that can obtain domain-invariant representations is constructed. This method is based on the deep model of the multi-attention mechanism, enhances the temporal and cross-channel correlations between extracted features, and does not need to access the target cutting condition (ie, target domain) data during the model construction process. It directly uses multiple historical conditions (ie, source domain) data to learn generalized representations, and uses the distributed sliced ​​Wasserstein distance (DSWD) to align the features between source domains across domains, thereby effectively generalizing the wear monitoring model to unknown cutting scene tasks.

[0061] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0062] like Figure 1 As shown, a method for monitoring tool wear status in an unknown working condition provided by an embodiment of the present application includes:

[0063] S101: Generate a tool wear data set based on the monitoring signal and tool wear state generated by tool cutting under historical working conditions, and map the tool wear data set to the corresponding state space, so as to construct a source domain input sequence corresponding to the tool according to the tool wear data set in the state space.

[0064] Figure 2 A schematic diagram of a multi-source domain generalization method for integrating multi-attention mechanism feature extraction provided in an embodiment of the present application. Figure 2 As shown, given and are the model input space and wear state space respectively, The number of tool wear states in the space is C, and X and Y represent the corresponding random variables. The tool wear data set obtained under a certain cutting condition is defined as D, which represents The joint distribution P on XY The number of tool wear data sets obtained under M historical working conditions constitutes the source domain exist Target domain It consists of a tool wear dataset under unknown cutting conditions. For example, the source domain The dataset is formally expressed as Where L is the source domain The number of samples, x i,l Input space for the model The characteristic sequence of i,l Space for wear label When identifying the tool wear state under changing cutting conditions, the marginal distributions of the source domain and the target domain are usually different, such as Assume that the label space of each domain is the same, that is, The present invention aims to use only the source domain D S The sample is used as input, and a generalized feature extractor F is trained by integrating the multi-source domain generalization method of the multi-attention mechanism. φ : and classifier T θ : After model training, F φ It can map the target domain data under unknown cutting conditions into the hidden feature space with domain-invariant representation. At the same time T θ Achieve reliable identification of tool wear status.

[0065] Based on this, a tool wear data set is generated according to the monitoring signals and tool wear status generated by tool cutting under historical working conditions, and the tool wear data set is mapped to the corresponding state space. The state space here refers to the model input space and the wear state space. In this way, according to the tool wear data set in the state space, the source domain input sequence corresponding to the tool is constructed.

[0066] The tool wear dataset can be obtained by Figure 3The experimental device shown in the figure is obtained. On processing equipment such as CNC machine tools, sensor equipment such as dynamometers and accelerometers are installed to monitor the corresponding multi-channel cutting force, vibration and other monitoring signals in real time online during the cutting process. These signals are amplified by charge amplifiers and transmitted to the data acquisition module, and finally transmitted to the time series database for storage. In addition, the observation of the tool wear state adopts an offline measurement method, using preset measuring equipment such as optical microscopes, based on standards such as GB / T 16460-2016 end mill life standard or other documents, and the wear band width of the tool back face is used as the basis for dividing the tool wear state, so as to generate corresponding wear state labels according to the wear band width, store the wear state labels in the time series database and associate and match the monitoring signals. After obtaining the above-mentioned original monitoring signals and wear state labels, a series of preprocessing such as format / unit conversion, data outlier detection, missing value processing, and noise reduction are performed. Then, the preprocessed multi-channel monitoring signals are segmented by sliding windows to obtain several segmented signals. For the segmented signals, the data of each channel is extracted in the time domain, frequency domain, and time-frequency domain. Then, each type of extracted features is spliced ​​along the time series direction of the monitoring signal to generate a multi-channel feature time series, and the time series is normalized and other operations are performed to form a tool wear data set.

[0067] S102: Input the source domain input sequence as a training sample into a preset wear monitoring model, and adaptively extract the wear degradation features in the training sample based on the multi-attention mechanism of the feature extractor in the wear monitoring model.

[0068] Constructing a practical latent feature space intrinsically related to tool wear status from a multi-channel feature sequence is the basis for tool wear status monitoring under cutting conditions. Therefore, the embodiment of the present application proposes a wear monitoring model with multiple attention mechanisms, focusing on extracting high-level features from both time and cross-channel aspects and enhancing the feature correlation with the wear status. The wear monitoring model includes a feature extractor, the structure of which is as follows Figure 4 As shown in Figure 1, the feature extractor consists of a Transformer encoder, a channel attention layer consisting of a squeeze and excitation module, and an Attentive Statistics Pooling (ASP) module. In order to effectively identify the main features related to the wear state in a global perspective, the Transformer encoder is selected as the basic component of the feature extractor, which can establish global temporal correlations for all input sequence features in parallel.

[0069] Specifically, based on the Transformer encoder, the corresponding first feature weights are assigned to the feature matrix of multiple channels in the source domain input sequence. The feature matrix is ​​composed of features of multiple channels, and the first feature weight refers to the weight of the global eigenvalue in the feature matrix. In addition, a large number of statistical features obtained by different sensors usually contribute to the identification of tool wear states to varying degrees. Therefore, a channel attention layer is used after the Transformer encoder to highlight those feature channels that contribute more to the identification of wear states. In the channel attention layer, the feature matrix is ​​squeezed and excited respectively through the global average pooling and gate mechanism to calibrate the feature channel weights. The feature channel weight refers to the weight corresponding to each channel in the feature matrix. The calibrated feature channel weights are multiplied by the feature matrix corresponding to the source domain input sequence to obtain a feature map based on the calibrated feature weights. It should be noted that after obtaining the feature map of the calibrated feature weights, it is necessary to perform residual connection on the feature map and the feature matrix corresponding to the source domain input sequence, so as to avoid the performance degradation caused by network deepening through residual connection. The residual connection is expressed as: F mid =X+f SE (f TE (X)

[0070] In the formula, f SE 、f TE They represent the squeeze excitation operation and the Transformer encoder operation respectively, and X is the feature matrix.

[0071] If the output features are processed by GAP or global max pooling (GMP) at the end of the feature extractor, key information may be lost. Simply flattening the output features into the hidden feature space may result in too much redundant information, thus limiting the accuracy of wear state recognition. At the same time, a single input data sample contains the complete cut-in information of the time series from cut-in to cut-out. Since the multi-channel features in the sample are extracted sequentially along the time series direction, the importance of channel features at different times for wear state recognition should be different. Therefore, the ASP module is used at the end of the feature extractor for dimensionality reduction. This module uses the attention mechanism and the attention statistical pooling layer in turn to calculate the weighted average and weighted standard deviation vectors. Based on this, the ASP module can capture the importance of features in each channel at different times, thereby reducing the redundant dimensions of the hidden feature space while adjusting the second feature weight corresponding to each feature in each channel. The second feature weight refers to the weights corresponding to multiple features in a single channel along the time series direction. Through the above operations, the feature extractor establishes a hidden feature space that is highly correlated with tool wear. By mapping the feature map to the hidden feature space, the wear degradation feature can be obtained. Among them, the hidden layer feature space can be expressed as:

[0072] F φ =f ASP (F mid )

[0073] In the formula, f ASP Represents an attention statistics pooling operation.

[0074] S103: performing metric alignment on the wear degradation features in different historical working conditions to calculate the metric distance between the wear degradation features, and calculating the domain difference metric loss for evaluating the feature distribution difference between the wear degradation features based on the metric distance.

[0075] Based on the feature extractor based on the multi-attention mechanism, the embodiment of the present application proposes a wear monitoring model, whose architecture is as follows: Figure 5 As shown. The training samples of multiple historical working conditions are used as input, and the wear degradation features are extracted using a feature extractor based on a multi-attention mechanism. The input is mapped to the feature space through parameter sharing during model training. Due to the differences in the dynamic response characteristics of the tool during cutting caused by different cutting conditions, the distribution of the monitoring data extracted under different working conditions is significantly different. In order to reduce the feature differences between multiple working conditions in the feature space, the domain-invariant representation between cutting conditions can be obtained, such as Figure 6 A schematic diagram of the DSWD measurement principle is shown. DSWD is used in the feature space to measure and align the wear degradation features between various working conditions, and the domain difference measurement loss is used as one of the loss functions. The above purpose is achieved by minimizing the loss function in model training.

[0076] Among them, the specific calculation method of the measurement can be expressed as:

[0077]

[0078] In the formula, represents the p-order DSWD calculation, For the corresponding Wasserstein metric calculation, Const is a regularization constant greater than zero that satisfies the condition in the unit sphere The set of probability measures r on , satisfy From the unit sphere arrive The set of all Borel measurable functions of , is the Radon transform operator, f b is a Borel measurable function, is the gradient of the deep neural network with parameter r, λ Const is the regularization hyperparameter in model training.

[0079] Through DSWD, feature distribution differences reduce redundant information when measuring, construct a hidden feature space with domain-invariant representation, and increase effective information, which can identify more valuable hidden feature distribution information. The domain difference measurement loss calculation formula can be expressed as:

[0080]

[0081] Where M represents the number of source domains, and They are the wear degradation characteristics under different historical working conditions.

[0082] S104: inputting the wear degradation features into the parameter-sharing wear state classifier to obtain the corresponding wear state classification results, and calculating the classification loss between the wear degradation features for the wear state classification results.

[0083] The wear degradation characteristics of each historical working condition are respectively sent to the parameter-sharing wear state classifier to obtain the corresponding wear state classification results. At the same time, according to the wear state classification results, the classification loss obtained by the cross entropy function is calculated. By minimizing this loss during training, the accuracy of wear state recognition is improved. The classification recognition loss calculation formula is as follows:

[0084]

[0085] Among them, F φ (x i ) is the wear degradation characteristic, y i For wear status label, T θ is the wear state classifier.

[0086] S105: Determine the comprehensive loss corresponding to the wear monitoring model according to the domain difference measurement loss and the classification loss, iterate the wear monitoring model according to the comprehensive loss until the training is completed, and obtain a generalized wear monitoring model.

[0087] Both domain difference metric loss and classification loss are loss functions used to measure the wear monitoring model. The comprehensive loss corresponding to the wear monitoring model is determined according to these two loss values. The wear monitoring model can be iterated. By minimizing the above comprehensive loss, a trained wear monitoring model with domain-invariant representation can be obtained.

[0088] Specifically, the loss function weight parameter corresponding to the domain difference metric loss is determined, the domain difference metric loss is weighted by the loss function weight parameter, and the weighted domain difference metric loss and classification loss are added to obtain the comprehensive loss corresponding to the wear monitoring model. The comprehensive loss can be expressed as:

[0089] Ltotal =L ce +βL DSWD

[0090] In the formula, β is the weight parameter of the loss function.

[0091] S106: Predicting the tool wear state under unknown cutting conditions through the wear monitoring model.

[0092] The wear monitoring model can predict the tool wear state under unknown cutting conditions. The real-time tool monitoring signal collected in the unknown cutting condition is input into the wear monitoring model, and the wear monitoring model will identify the tool wear state and output it.

[0093] The above are embodiments of the method proposed in this application. Based on the same idea, some embodiments of this application also provide devices and non-volatile computer storage media corresponding to the above methods.

[0094] Figure 7 The schematic diagram of the structure of a tool wear state monitoring device for unknown working conditions provided in the embodiment of the present application is shown in FIG. Figure 7 As shown, including:

[0095] at least one processor; and,

[0096] at least one processor is communicatively connected to a memory; wherein,

[0097] The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor so that the at least one processor can execute a method for monitoring the wear state of a tool under unknown working conditions as described in any one of the above items.

[0098] The embodiment of the present application provides a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured as follows:

[0099] A method for monitoring tool wear status in unknown working conditions as described in any of the above items.

[0100] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0101] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0102] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0103] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0104] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0106] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0107] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0108] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0109] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0110] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for monitoring tool wear status in unknown working conditions, characterized in that: The method comprises: Generate a tool wear data set according to the monitoring signal and tool wear state generated by tool cutting under historical working conditions, and map the tool wear data set to a corresponding state space, so as to construct a source domain input sequence corresponding to the tool according to the tool wear data set in the state space; Inputting the source domain input sequence as a training sample into a preset wear monitoring model, and adaptively extracting wear degradation features in the training sample based on a multi-attention mechanism of a feature extractor in the wear monitoring model; Performing metric alignment on the wear degradation features in different historical working conditions to calculate the metric distance between the wear degradation features, and calculating the domain difference metric loss for evaluating the feature distribution difference between the wear degradation features according to the metric distance; Inputting the wear degradation features into a parameter-sharing wear state classifier to obtain a corresponding wear state classification result, and calculating the classification loss between the wear degradation features for the wear state classification result; Determine the comprehensive loss corresponding to the wear monitoring model according to the domain difference metric loss and the classification loss, iterate the wear monitoring model according to the comprehensive loss until the training is completed, and obtain the wear monitoring model with generalization; The wear monitoring model is used to predict the tool wear state under unknown cutting conditions.

2. The method for monitoring tool wear status in unknown working conditions according to claim 1, characterized in that: According to the monitoring signals and tool wear status generated by tool cutting under historical working conditions, a tool wear data set is generated, including: Collect monitoring signals generated by tool cutting under historical working conditions, and detect the wear band width corresponding to the back face of the tool through a preset measuring device, so as to determine the tool wear state corresponding to the tool according to the wear band width; According to the tool wear state, generating a corresponding wear state label, and associating the wear state label with the monitoring signal; Preprocessing the monitoring signal, and performing sliding window segmentation on the preprocessed monitoring signal to obtain a plurality of segmented signals; For the segmented signal, feature extraction is performed on the data of each channel in the segmented signal, and the extracted features are spliced ​​according to the timing direction of the monitoring signal to generate a corresponding multi-channel feature time series; The multi-channel feature time series is normalized to obtain a tool wear data set.

3. The method for monitoring tool wear status in unknown working conditions according to claim 1, characterized in that: Based on the multi-attention mechanism of the feature extractor in the wear monitoring model, the wear degradation features in the training samples are adaptively extracted, specifically including: The feature extractor includes a Transformer encoder, a channel attention layer, and an attention statistics pool module; Based on the Transformer encoder in the feature extractor, adaptively assigning corresponding first feature weights to the feature matrix of multiple channels in the source domain input sequence; wherein the first feature weight refers to the weight of the global eigenvalue in the feature matrix; Based on the channel attention layer, a squeezing operation and an excitation operation are respectively performed on the feature matrix through a global average pool and a gate mechanism to calibrate feature channel weights corresponding to multiple channels in the feature matrix; Multiplying the feature channel weights by the feature matrix corresponding to the channel to obtain a feature map based on the calibrated feature channel weights; Based on the attention statistics pool module, by capturing the importance of features in each channel at different times, the second feature weight corresponding to each feature in the channel is adjusted, and the feature map is reduced in dimension to map the feature map to the hidden layer feature space to obtain the wear and degradation feature; wherein the second feature weight refers to the weights corresponding to multiple features in a single channel along the time series direction in the feature map.

4. The method for monitoring tool wear status in unknown working conditions according to claim 1, characterized in that: The wear degradation features in different historical working conditions are metrically aligned to calculate the metric distance between the wear degradation features. The specific calculation method is: In the formula, represents the p-order DSWD calculation, For the corresponding Wasserstein metric calculation, Const is a regularization constant greater than zero that satisfies the condition in the unit sphere The set of probability measures r on , satisfy From the unit sphere arrive The set of all Borel measurable functions of , is the Radon transform operator, f b is a Borel measurable function, is the gradient of the deep neural network with parameter r, λ Const is the regularization hyperparameter in model training.

5. The method for monitoring tool wear status in unknown working conditions according to claim 4, characterized in that: According to the metric distance, a domain difference metric loss for evaluating the feature distribution difference between the wear degradation features is calculated, and the specific calculation method is: Where M represents the number of source domains, and They are the wear degradation characteristics under different historical working conditions.

6. The method for monitoring tool wear status in unknown working conditions according to claim 1, characterized in that: According to the wear state classification result, the classification loss between the wear degradation characteristics is calculated, and the specific calculation method is: Among them, F φ (x i ) is the wear degradation characteristic, y i For wear status label, T θ is the wear state classifier.

7. The method for monitoring tool wear status in unknown working conditions according to claim 1, characterized in that: Determining the comprehensive loss corresponding to the wear monitoring model according to the domain difference measurement loss and the classification loss specifically includes: Determining a loss function weight parameter corresponding to the domain difference metric loss; The domain difference metric loss is weighted by the loss function weight parameter, and the weighted domain difference metric loss and the classification loss are added to obtain the comprehensive loss corresponding to the wear monitoring model.

8. The method for monitoring tool wear status in unknown working conditions according to claim 2, characterized in that: Before reducing the dimension of the feature map, the method further includes: Performing a residual connection on the feature map and the feature matrix corresponding to the source domain input sequence to obtain the feature map output by the channel attention layer; The residual connection is expressed as: F mid =X+f SE (f TE (X)) In the formula, f SE 、f TE They represent the squeeze excitation operation and the Transformer encoder operation respectively, and X is the feature matrix.

9. A tool wear status monitoring device for unknown working conditions, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; Wherein, the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method for monitoring tool wear status in unknown working conditions as described in any one of claims 1-8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured to: A method for monitoring tool wear status in unknown working conditions as described in any one of claims 1 to 8.

Images