Numerical control machine tool cutter life prediction method and device and numerical control machine tool

Abstract

The invention discloses a numerical control machine tool cutter service life prediction method and device and a numerical control machine tool, and relates to the technical field of numerical control machine tools, and the method comprises the steps: collecting cutter operation characteristic data in real time in the operation process of the numerical control machine tool, and the cutter operation characteristic data comprises multiple types of characteristic data; inputting the tool operation characteristic data into a tool life prediction model, and outputting to obtain a first tool life prediction result; and obtaining a processed finished product image after material processing, performing identification in combination with a preset target finished product image to obtain a processing similarity, and performing classification in combination with the first tool life prediction result to obtain a tool life prediction result. The technical problem that the service life of the numerical control machine tool cutter is not accurately predicted in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of numerically controlled machine tools, and in particular to a method and device for predicting tool life of numerically controlled machine tools and a numerically controlled machine tool. Background Art

[0002] When the tool of a CNC machine tool is worn out, it needs to be replaced. In the prior art, the tool condition is generally checked and recorded regularly. When the tool is damaged or about to be damaged (that is, when the wear is serious), the tool is replaced. The existing tool wear detection has high labor costs, and the reliability of high-end tools is relatively strong, and the possibility of sudden damage is relatively low. However, other tools have the problem of sudden damage. If they are not identified and replaced in time, it will cause material damage or even damage to the machine tool.

[0003] Some existing technologies have proposed to judge tool damage or predict tool life based on the current change of the motor driving the tool, but the prediction accuracy of a single data is low, and a large amount of data is required for training, which may cause overfitting of the model. Therefore, there is a technical problem of inaccurate prediction of tool life for CNC machine tools in the existing technology. Summary of the invention

[0004] The present application provides a method and device for predicting the tool life of a CNC machine tool, and a CNC machine tool, which are used to solve the technical problem of inaccurate prediction of the tool life of a CNC machine tool existing in the prior art.

[0005] In view of the above problems, the present application provides a method and device for predicting tool life of a CNC machine tool and a CNC machine tool.

[0006] In a first aspect, the present application provides a method for predicting tool life of a CNC machine tool, the method comprising: During the operation of the CNC machine tool, tool operation feature data is collected in real time, wherein the tool operation feature data includes multiple types of feature data; Inputting the tool operation characteristic data into a tool life prediction model, and outputting a first tool life prediction result; The image of the finished product after material processing is obtained, combined with the preset target finished product image, the processing similarity is identified, and the tool life prediction result is obtained by classification in combination with the first tool life prediction result.

[0007] In a second aspect, the present application provides a device for predicting tool life of a CNC machine tool, the device comprising: The tool data acquisition module is used to collect tool operation characteristic data in real time during the operation of the CNC machine tool, wherein the tool operation characteristic data includes multiple types of characteristic data; A first life prediction module, used for inputting the tool operation characteristic data into a tool life prediction model, and outputting a first tool life prediction result; The life prediction classification module is used to obtain the image of the finished product after material processing, combine it with the preset target finished product image, identify and obtain the processing similarity, and combine it with the first tool life prediction result to classify and obtain the tool life prediction result.

[0008] In a third aspect, the present application provides a CNC machine tool, comprising the CNC machine tool tool life prediction device of the second aspect.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages: The present application proposes a method for predicting tool life of CNC machine tools, which collects tool operation feature data of multimodal data during the operation of the CNC machine tool, then trains a tool life prediction model based on the multimodal data, inputs and predicts the current tool operation feature data, obtains a first tool life prediction result, then obtains a finished product image after material processing, combines the preset target finished product image, identifies the processing similarity, and classifies the tool life prediction result in combination with the first tool life prediction result. The present application identifies the quality of material processing through a lightweight model, then combines the tool life predicted by multimodal data, and finally classifies and predicts to obtain the tool life prediction result. With a small amount of data, the turning tool life is identified, achieving the technical effect of improving the accuracy of tool life prediction and improving the processing quality of CNC machine tools. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A schematic diagram of a process flow of a method for predicting tool life of a CNC machine tool provided in an embodiment of the present application; Figure 2 A schematic diagram of the architecture of a tool life prediction model in a CNC machine tool tool life prediction method provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a tool life prediction device for a CNC machine tool provided in an embodiment of the present application.

[0012] In the accompanying drawings, the components represented by the reference numerals are described as follows: Tool data collection module 11, first life prediction module 12, life prediction classification module 13. DETAILED DESCRIPTION

[0013] The present application provides a method and device for predicting the tool life of a CNC machine tool, and a CNC machine tool, so as to solve the technical problem of inaccurate prediction of the tool life of a CNC machine tool existing in the prior art.

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction 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. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0015] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0016] Embodiment 1, as Figure 1 As shown, the present application provides a method for predicting tool life of a CNC machine tool, the method comprising: S100: During the operation of the CNC machine tool, real-time collection of tool operation feature data, wherein the tool operation feature data includes multiple types of feature data; In the embodiment of the present application, during the operation of the CNC machine tool, tool operation feature data is collected in real time, wherein multiple types of feature data are collected, tool life is predicted based on multi-modal data, prediction accuracy is improved, and overfitting of training is avoided.

[0017] Step S100 in the method provided in the embodiment of the present application includes: During the operation of the CNC machine tool, auxiliary light is used to collect material images in real time, wherein the auxiliary light is irradiated onto the material to form a pattern of a preset shape; Real-time acquisition of spindle current signals of tool operation; Collect audio signals of lane operation in real time; The material image, the spindle current signal and the audio signal are aligned to obtain tool operation characteristic data, wherein the acquisition frequency of the spindle current signal and the audio signal is greater than the acquisition frequency of the material image.

[0018] In the embodiment of the present application, during the operation of the CNC machine tool, auxiliary light is used in real time to capture material images, and the auxiliary light is used to improve the quality and consistency of image capture.

[0019] For example, during the machining of a turning tool (rough and fine turning tool), a light beam is used to irradiate the material to form a pattern on the material. When the turning tool processes the material, the formed pattern also changes due to changes in the material structure. Then, the material image of the material is collected to reflect the wear condition of the turning tool.

[0020] Among them, auxiliary light is used to form a preset pattern on the surface of the material. Compared with the single photoelectric, the formed light pattern has a larger deformation amplitude when the material is deformed by turning tool processing, and the subsequent recognition accuracy is higher.

[0021] For example, the preset pattern can match the finished shape of the material. For example, if the finished surface of the material is a plane, the preset pattern is a straight line. If the finished surface is a conical surface, the preset pattern can be an inclined straight line. The preset image can be set according to the trend of the material shape change, and can also be a curve, etc.

[0022] Exemplarily, the frame rate of capturing material images is 30, that is, 30 material images are captured in 1 second.

[0023] Furthermore, during the operation, the spindle current signal and audio signal of the tool operation are collected in real time, for example, the spindle current size of the tool operation is collected, and the audio signal of the tool turning process is collected through a microphone.

[0024] The acquisition frequency of the spindle current signal and the audio signal is greater than the acquisition frequency of the material image. For example, the sampling frequency of the current signal data and the audio signal is 200 Hz, that is, 200 spindle current signals and audio signals are collected in 1 second.

[0025] Furthermore, the collected material image, spindle current signal and audio signal are aligned to obtain the material image, spindle current signal and audio signal of the tool turning tool at the same time as the tool operation feature data.

[0026] The step of "aligning the material image, the spindle current signal and the audio signal to obtain tool operation feature data" in the method provided in the embodiment of the present application includes: Acquiring an exposure time interval of the material image; Selecting the middle moment in the exposure time interval as the alignment timestamp; According to the acquisition frequency of the material image, the spindle current signal and the audio signal are divided to obtain a spindle current signal segment sequence and an audio signal segment sequence; The spindle current signal segment and the audio signal segment under the alignment timestamp are indexed to obtain the aligned spindle current signal and the aligned audio signal, which are combined with the material image as tool operation feature data.

[0027] In the embodiment of the present application, the three types of data are aligned using the material image as the alignment reference.

[0028] When auxiliary light is used to collect material images, exposure and image collection are performed when each material image is collected. There is an exposure time interval, and the exposure time interval includes a time length. For example, the exposure time interval of the first material image is 0-0.033 seconds, and the exposure time interval of the second material image is 0.033-0.066 seconds. The middle moment in the exposure time interval is selected as the alignment timestamp for data alignment. For example, 0.016 seconds is selected as the alignment timestamp of the first material image.

[0029] When performing data alignment, it is necessary to ensure that the frequencies of the material image, the spindle current signal, and the audio signal are consistent. Therefore, the data of the spindle current signal and the audio signal are divided according to the acquisition frequency of the material image. Exemplarily, the spindle current signal and the audio signal acquired within 1 second are divided into 30 segments to obtain a spindle current signal segment sequence and an audio signal segment sequence. The spindle current signal segment sequence includes 30 spindle current signal segments, and the audio signal segment sequence includes 30 audio signal segments.

[0030] According to the alignment timestamp, the spindle current signal segment and the audio signal segment under the alignment timestamp are indexed, that is, the spindle current signal segment and the audio signal segment with the alignment timestamp in the corresponding time period are used as the aligned spindle current signal and the aligned audio signal of the material image.

[0031] Integrate the material image, the spindle current signal and the audio signal to obtain the tool operation characteristic data after alignment.

[0032] The step of "dividing the spindle current signal and the audio signal according to the acquisition frequency of the material image to obtain a spindle current signal segment sequence and an audio signal segment sequence" in the method provided in the embodiment of the present application includes: According to the acquisition frequency of the material image, configure the division time step; The spindle current signal and the audio signal are divided according to the division time step to obtain a spindle current signal segment sequence and an audio signal segment sequence, wherein there is an overlapping portion between two adjacent spindle current signal segments and audio signal segments.

[0033] In the embodiment of the present application, when the spindle current signal and the audio signal are divided, there are overlapping parts between two adjacent spindle current signal segments and audio signal segments, so as to facilitate the prediction of tool life based on time sequence and improve accuracy.

[0034] Therefore, according to the acquisition frequency of the material image, when configuring the division time step, the division time step is greater than 1 second divided by the average time step of the acquisition frequency of the material image. For example, the average time step of 1 second divided by the acquisition frequency of the material image is 1 second divided by 30=0.033 seconds, and the actual division time step is set to 0.05 seconds. In this way, when the spindle current signal and the audio signal are divided according to the division time step, there are overlapping parts in the two adjacent spindle current signal segments and audio signal segments.

[0035] Exemplarily, when the spindle current signal and the audio signal are divided according to the division time step, 0-0.05 seconds is the first segment of the spindle current signal and the audio signal, and 0.033-0.083 is the second segment of the spindle current signal and the audio signal. In this way, the spindle current signal segment sequence and the audio signal segment sequence of the acquisition frequency of the material image are divided. For example, the spindle current signal segment sequence and the audio signal segment sequence include 30 spindle current signal segments and audio signal segments, and there are overlapping parts in two adjacent spindle current signal segments and audio signal segments.

[0036] S200: inputting the tool operation characteristic data into a tool life prediction model, and outputting a first tool life prediction result; In the embodiment of the present application, tool operation characteristic data is acquired during collection and processing, and is input into a trained tool life prediction model to perform a preliminary tool life prediction, and a first tool life prediction result is obtained as an output.

[0037] Among them, the construction steps of the tool life prediction model include: According to the tool operation data in the historical time, the sample tool operation feature data set is processed and obtained, and each set of sample tool operation feature data is annotated to obtain the sample first tool life prediction result set as the supervised training data set, wherein each sample first tool life prediction result includes the annotated information of the early stage of tool wear, the middle stage of tool wear and the late stage of tool wear; Constructing a material image feature extraction path, a current signal feature extraction path, an audio signal feature extraction path, a fusion module and a classification module in the tool life prediction model; The supervised training data set is used to perform supervised training, verification and testing on the tool life prediction model until convergence, thereby obtaining a tool life prediction model.

[0038] like Figure 2As shown, in the embodiment of the present application, based on the tool operation data in the historical time, multimodal data related to the tool life is first collected and processed to form a sample tool operation feature data set. Specifically, the sample tool operation feature data set includes three types of key data: image data formed during the material processing process, current signal data of the spindle motor, and audio signal data during the processing process. Alignment processing is performed through the steps in the aforementioned content to obtain a sample tool operation feature data set.

[0039] After obtaining the sample tool operation feature data set, each set of sample tool operation feature data is annotated by combining the tool life annotation standard. The annotation content includes three stages of tool wear, namely, the early stage of tool wear (the tool performance is basically intact, only slight surface wear is shown, and the tool sharpness is basically maintained), the middle stage of tool wear (the tool performance has declined, the surface wear is more obvious, and the processing quality begins to fluctuate), and the late stage of tool wear (the tool performance has seriously declined, the degree of wear is significant, and there is a risk of tool breakage). Annotation can be based on manual annotation or measurement of tool wear. Finally, the first tool life prediction result set of the sample is obtained as the supervised training data set of the model.

[0040] like Figure 2 As shown, the network structure of the tool life prediction model is constructed, and the network structure of the tool life prediction model includes a material image feature extraction path, a current signal feature extraction path, an audio signal feature extraction path, a fusion module and a classification module.

[0041] Among them, the material image feature extraction path is built based on a 2D convolutional neural network, and features are extracted from the material image and flattened into a one-dimensional vector. The current signal feature extraction path and the audio signal feature extraction path are built based on the first and second GRU neural networks, and features are extracted from the current signal and audio signal, and output as a one-dimensional vector. The fusion module is built based on a DNN, specifically including a fully connected layer and a Dropout layer. The classification module is built based on the third GRU neural network and a softmax activation function. The output categories of the tool life prediction model include the early stage of tool wear, the middle stage of tool wear, and the late stage of tool wear. Among them, the output of GRU at each moment includes historical information. In order to better integrate historical information, it also includes a splicing layer, which is set after the material image feature extraction path, the current signal feature extraction path, and the audio signal feature extraction path, respectively. The output of one full second is spliced ​​and then sent to the fusion module for fusion. The amount of data after fusion is large, so the dropout parameter is set to 0.3.

[0042] The supervised training data set is used to conduct supervised training on the tool life prediction model. During the training process, the model parameters are adjusted by back propagation and iterative optimization according to the output of the tool life prediction model, and the recognition accuracy of the model for the tool life stage is gradually improved. Model training also includes verification and testing steps to evaluate the generalization ability and prediction performance of the model. The loss function in the training process adopts cross entropy loss, and the optimizer is updated using the Adam algorithm until the model converges on the verification set, for example, the accuracy reaches 90%, and the final tool life prediction model is obtained.

[0043] Based on the trained tool life prediction model, the currently collected tool operation feature data is input into the tool life prediction model, and the first tool life prediction result is output, which includes one of the early stage of tool wear, the middle stage of tool wear or the late stage of tool wear.

[0044] S300: Acquire a finished product image after material processing, combine it with a preset target finished product image, identify and obtain processing similarity, and combine it with the first tool life prediction result to obtain a tool life prediction result through classification.

[0045] Furthermore, in order to improve the accuracy of tool life prediction analysis, tool life analysis is also performed based on the similarity between the material after processing and the standard.

[0046] Get the finished product image after material processing, combine it with the preset target finished product image of the expected processing (i.e. the finished product image after standard processing), identify the similarity between the two. If the tool life is good and the wear is less, the finished product image should have a high similarity with the target finished product image. On the contrary, if the tool wear is more serious, the similarity is lower. Identify the similarity between the two, obtain the processing similarity, combine it with the first tool life prediction result, classify the final tool life prediction result, and obtain the tool life prediction result.

[0047] By combining the similarity between the finished product image after processing and the target finished product image, tool life prediction classification can be performed to improve the accuracy and reliability of tool life prediction analysis.

[0048] Step S300 in the method provided in the embodiment of the present application includes: After the CNC machine tool completes material processing, the processed product image is acquired, and the target finished product image preset for material processing is obtained; Based on the twin network, a finished product similarity recognition network is constructed; The processed finished product image and the target finished product image are input into the finished product similarity recognition network, image features are extracted and the Euclidean distance is calculated to obtain the processing similarity.

[0049] In the embodiment of the present application, after the CNC machine tool completes material processing, in order to further evaluate the processing quality and tool status, the system collects an image of the processed product and obtains a preset target finished product image before processing. These two images reflect the actual shape and ideal shape of the processed product, respectively, providing a basis for the quantitative evaluation of processing quality.

[0050] Based on the Siamese Network, a finished product similarity recognition network is constructed, which includes two convolutional neural network paths and can extract features from the input finished product images.

[0051] The currently acquired processed finished product image and the target finished product image are input into the finished product similarity recognition network, the image features are extracted, and the Euclidean distance between the two image features is calculated to obtain the processing similarity between the processed finished product image and the target finished product image.

[0052] The relationship between the threshold of the Euclidean distance and the processing similarity can be preset. The processing similarity includes, for example, a low similarity or a high similarity. If the Euclidean distance is less than the threshold of the Euclidean distance, the similarity is low. If the Euclidean distance is greater than the threshold of the Euclidean distance, the similarity is high. In this way, the processing similarity between the processed finished product image and the target finished product image is identified. The processing similarity includes one of the two classification results of low similarity and high similarity.

[0053] Furthermore, step S300 in the method provided in the embodiment of the present application further includes: Construct tool life classifier; The first tool life prediction result and the processing similarity are input into the tool life classifier to obtain the tool life prediction result.

[0054] In an embodiment of the present application, a tool life classifier is constructed to perform final tool life prediction classification based on the first tool life prediction result and the processing similarity.

[0055] Among them, the tool life classifier is constructed, including: Obtaining a sample first tool life prediction result set at the early stage of tool wear, the middle stage of tool wear, and the late stage of tool wear, and collecting a sample processing similarity set; Traversing the combination of the sample first tool life prediction result set and the sample processing similarity set, and marking the sample tool life prediction result for each combination to obtain the sample tool life prediction result set; A tool life classifier is constructed by using the sample first tool life prediction result set, the sample processing similarity set and the sample tool life prediction result set.

[0056] In the embodiment of the present application, a sample first tool life prediction result set of the early stage of tool wear, the middle stage of tool wear and the late stage of tool wear is first obtained, that is, the sample first tool life prediction result includes one of the early stage of tool wear, the middle stage of tool wear and the late stage of tool wear. The sample first tool life prediction result set is used as one of the classification data for classifying and obtaining the tool life prediction result.

[0057] A sample processing similarity set is further obtained. Exemplarily, the sample processing similarities include two types, namely, low similarity and high similarity.

[0058] Each sample first tool life prediction result and each sample processing similarity in the sample first tool life prediction result set and the sample processing similarity set are traversed and combined to obtain multiple combinations, such as six combinations: early stage of tool wear-low similarity, mid-stage of tool wear-low similarity, late stage of tool wear-low similarity, early stage of tool wear-high similarity, mid-stage of tool wear-high similarity and late stage of tool wear-high similarity.

[0059] The sample tool life prediction results are labeled for each combination, wherein the sample tool life prediction results include early stage, middle stage and late stage. Each combination is labeled, for example, the sample tool life prediction result corresponding to the early stage of tool wear-low similarity is late stage, the sample tool life prediction result corresponding to the middle stage of tool wear-low similarity is late stage, the sample tool life prediction result corresponding to the late stage of tool wear-low similarity is late stage, the sample tool life prediction result corresponding to the early stage of tool wear-high similarity is early stage, the sample tool life prediction result corresponding to the middle stage of tool wear-high similarity is middle stage, and the sample tool life prediction result corresponding to the late stage of tool wear-high similarity is middle stage. In this way, a set of sample tool life prediction results is obtained.

[0060] A tool life classifier is constructed using the sample first tool life prediction result set, the sample processing similarity set and the sample tool life prediction result set. Specifically, a mapping relationship between the combination of the sample first tool life prediction result set, the sample processing similarity set and the sample tool life prediction result set is constructed to obtain the tool life classifier.

[0061] After the first tool life prediction result is obtained by prediction and the processing similarity is obtained by identification, the two are input into the tool life classifier to obtain the tool life prediction result corresponding to the combination of the two, and the prediction and classification of tool life are completed. The tool life prediction result is specifically one of the early, middle and late stages. When the tool life prediction result is the late stage, the tool can be replaced.

[0062] In summary, the embodiments of the present application have at least the following technical effects: The embodiment of the present application proposes a method for predicting tool life of CNC machine tools. During the operation of the CNC machine tool, tool operation characteristic data of multimodal data is collected, and then a tool life prediction model is trained based on the multimodal data. The current tool operation characteristic data is input for prediction to obtain a first tool life prediction result. Then, the image of the finished product after material processing is obtained, and the processing similarity is obtained by combining with the preset target finished product image. The tool life prediction result is obtained by classification in combination with the first tool life prediction result. The present application identifies the quality of material processing through a lightweight model, and then combines the tool life predicted by multimodal data, and finally classifies and predicts to obtain the tool life prediction result. With a small amount of data, the turning tool life is identified, and the technical effect of improving the accuracy of tool life prediction and improving the processing quality of CNC machine tools is achieved.

[0063] Embodiment 2, as Figure 3 As shown, based on the same inventive concept as the method for predicting tool life of a CNC machine tool provided in the first embodiment, the embodiment of the present invention further provides a device for predicting tool life of a CNC machine tool, comprising: The tool data acquisition module 11 is used to collect tool operation characteristic data in real time during the operation of the CNC machine tool, wherein the tool operation characteristic data includes multiple types of characteristic data; A first life prediction module 12, used for inputting the tool operation characteristic data into a tool life prediction model, and outputting a first tool life prediction result; The life prediction classification module 13 is used to obtain the image of the finished product after material processing, identify and obtain the processing similarity in combination with the preset target finished product image, and classify and obtain the tool life prediction result in combination with the first tool life prediction result.

[0064] In one embodiment, the tool data collection module 11 is further used for: During the operation of the CNC machine tool, auxiliary light is used to collect material images in real time, wherein the auxiliary light is irradiated onto the material to form a pattern of a preset shape; Real-time acquisition of spindle current signals of tool operation; Collect audio signals of lane operation in real time; The material image, the spindle current signal and the audio signal are aligned to obtain tool operation characteristic data, wherein the acquisition frequency of the spindle current signal and the audio signal is greater than the acquisition frequency of the material image.

[0065] The material image, the spindle current signal and the audio signal are aligned to obtain tool operation characteristic data, including: Acquiring an exposure time interval of the material image; Selecting the middle moment in the exposure time interval as the alignment timestamp; According to the acquisition frequency of the material image, the spindle current signal and the audio signal are divided to obtain a spindle current signal segment sequence and an audio signal segment sequence; The spindle current signal segment and the audio signal segment under the alignment timestamp are indexed to obtain the aligned spindle current signal and the aligned audio signal, which are combined with the material image as tool operation feature data.

[0066] Wherein, the spindle current signal and the audio signal are divided according to the acquisition frequency of the material image to obtain a spindle current signal segment sequence and an audio signal segment sequence, including: According to the acquisition frequency of the material image, configure the division time step; The spindle current signal and the audio signal are divided according to the division time step to obtain a spindle current signal segment sequence and an audio signal segment sequence, wherein there is an overlapping portion between two adjacent spindle current signal segments and audio signal segments.

[0067] In one embodiment, the first life prediction module 12 is further used for: According to the tool operation data in the historical time, the sample tool operation feature data set is processed and obtained, and each set of sample tool operation feature data is annotated to obtain the sample first tool life prediction result set as the supervised training data set, wherein each sample first tool life prediction result includes the annotated information of the early stage of tool wear, the middle stage of tool wear and the late stage of tool wear; Constructing a material image feature extraction path, a current signal feature extraction path, an audio signal feature extraction path, a fusion module and a classification module in the tool life prediction model; The supervised training data set is used to perform supervised training, verification and testing on the tool life prediction model until convergence, thereby obtaining a tool life prediction model.

[0068] In one embodiment, the life prediction classification module 13 is further used for: After the CNC machine tool completes material processing, the processed product image is acquired, and the target finished product image preset for material processing is obtained; Based on the twin network, a finished product similarity recognition network is constructed; The processed finished product image and the target finished product image are input into the finished product similarity recognition network, image features are extracted and the Euclidean distance is calculated to obtain the processing similarity.

[0069] In one embodiment, the life prediction classification module 13 is further used for: Construct tool life classifier; The first tool life prediction result and the processing similarity are input into the tool life classifier to obtain the tool life prediction result.

[0070] Among them, building a tool life classifier includes: Obtaining a sample first tool life prediction result set at the early stage of tool wear, the middle stage of tool wear, and the late stage of tool wear, and collecting a sample processing similarity set; Traversing the combination of the sample first tool life prediction result set and the sample processing similarity set, and marking the sample tool life prediction result for each combination to obtain the sample tool life prediction result set; A tool life classifier is constructed by using the sample first tool life prediction result set, the sample processing similarity set and the sample tool life prediction result set.

[0071] Embodiment three, based on the same inventive concept of a CNC machine tool tool life prediction device provided in Embodiment two, the embodiment of the present application also provides a CNC machine tool, which includes a CNC machine tool tool life prediction device provided in Embodiment two, and also includes any structure of a CNC machine tool in the prior art.

[0072] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0073] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0074] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for predicting tool life of a CNC machine tool, characterized in that: The method comprises: During the operation of the CNC machine tool, tool operation feature data is collected in real time, wherein the tool operation feature data includes multiple types of feature data; Inputting the tool operation characteristic data into a tool life prediction model, and outputting a first tool life prediction result; The image of the finished product after material processing is obtained, combined with the preset target finished product image, the processing similarity is identified, and the tool life prediction result is obtained by classification in combination with the first tool life prediction result.

2. The method for predicting tool life of a CNC machine tool according to claim 1, characterized in that: During the operation of the CNC machine tool, the tool operation characteristic data is collected in real time, including: During the operation of the CNC machine tool, auxiliary light is used to collect material images in real time, wherein the auxiliary light is irradiated onto the material to form a pattern of a preset shape; Real-time acquisition of spindle current signals of tool operation; Collect audio signals of lane operation in real time; The material image, the spindle current signal and the audio signal are aligned to obtain tool operation characteristic data, wherein the acquisition frequency of the spindle current signal and the audio signal is greater than the acquisition frequency of the material image.

3. The method for predicting tool life of a CNC machine tool according to claim 2, characterized in that: The material image, spindle current signal and audio signal are aligned to obtain tool operation feature data, including: Acquiring an exposure time interval of the material image; Selecting the middle moment in the exposure time interval as the alignment timestamp; According to the acquisition frequency of the material image, the spindle current signal and the audio signal are divided to obtain a spindle current signal segment sequence and an audio signal segment sequence; The spindle current signal segment and the audio signal segment under the alignment timestamp are indexed to obtain the aligned spindle current signal and the aligned audio signal, which are combined with the material image as tool operation feature data.

4. The method for predicting tool life of a CNC machine tool according to claim 3, characterized in that: According to the acquisition frequency of the material image, the spindle current signal and the audio signal are divided to obtain a spindle current signal segment sequence and an audio signal segment sequence, including: According to the acquisition frequency of the material image, configure the division time step; The spindle current signal and the audio signal are divided according to the division time step to obtain a spindle current signal segment sequence and an audio signal segment sequence, wherein there is an overlapping portion between two adjacent spindle current signal segments and audio signal segments.

5. The method for predicting tool life of a CNC machine tool according to claim 4, characterized in that: The steps to build a tool life prediction model include: According to the tool operation data in the historical time, the sample tool operation feature data set is processed and obtained, and each set of sample tool operation feature data is annotated to obtain the sample first tool life prediction result set as the supervised training data set, wherein each sample first tool life prediction result includes the annotated information of the early stage of tool wear, the middle stage of tool wear and the late stage of tool wear; Constructing a material image feature extraction path, a current signal feature extraction path, an audio signal feature extraction path, a fusion module and a classification module in the tool life prediction model; The supervised training data set is used to perform supervised training, verification and testing on the tool life prediction model until convergence, thereby obtaining a tool life prediction model.

6. The method for predicting tool life of a CNC machine tool according to claim 1, characterized in that: Obtain the finished product image after material processing, combine it with the preset target finished product image, and identify the processing similarity, including: After the CNC machine tool completes material processing, the finished product image is acquired, and the preset target finished product image of the material processing is obtained; Based on the twin network, a finished product similarity recognition network is constructed; The processed finished product image and the target finished product image are input into the finished product similarity recognition network, image features are extracted and the Euclidean distance is calculated to obtain the processing similarity.

7. The method for predicting tool life of a CNC machine tool according to claim 1, characterized in that: Combined with the first tool life prediction result, the tool life prediction result is obtained by classification, including: Construct tool life classifier; The first tool life prediction result and the processing similarity are input into the tool life classifier to obtain the tool life prediction result.

8. The method for predicting tool life of a CNC machine tool according to claim 7, characterized in that: Building a tool life classifier involves: Obtaining a sample first tool life prediction result set at the early stage of tool wear, the middle stage of tool wear, and the late stage of tool wear, and collecting a sample processing similarity set; Traversing the combination of the sample first tool life prediction result set and the sample processing similarity set, and marking the sample tool life prediction result for each combination to obtain the sample tool life prediction result set; A tool life classifier is constructed by using the sample first tool life prediction result set, the sample processing similarity set and the sample tool life prediction result set.

9. A tool life prediction device for a CNC machine tool, characterized in that: Used to implement the method for predicting tool life of a CNC machine tool according to any one of claims 1 to 8, the device comprises: The tool data acquisition module is used to collect tool operation feature data in real time during the operation of the CNC machine tool, wherein the tool operation feature data includes multiple types of feature data; A first life prediction module, used for inputting the tool operation characteristic data into a tool life prediction model, and outputting a first tool life prediction result; The life prediction classification module is used to obtain the image of the finished product after material processing, combine it with the preset target finished product image, identify and obtain the processing similarity, and combine it with the first tool life prediction result to classify and obtain the tool life prediction result.

10. A numerically controlled machine tool, characterized in that: It includes the tool life prediction device for CNC machine tools as described in claim 9.

Images