Method and system for detecting state of intersection hole boring cutter

Abstract

The invention discloses a method and system for detecting the state of an intersection hole boring cutter, and belongs to the technical field of industrial big data. Detection is carried out based on multi-modal data, and the abrasion bandwidth of the rear tool face of the tool and the abrasion ratio of the front tool face image of the tool are correspondingly obtained through collected machining time sequence data in the one-tool machining process and the front tool face image of the tool after one-tool machining are completed and serve as detection indexes of the intersection point hole boring tool state. The machining precision condition is represented through the abrasion bandwidth, and when the abrasion bandwidth is large, it is indicated that the rear tool face of the tool is abraded greatly, and the machining precision is affected; the surface roughness condition is represented through the abrasion ratio, and when the abrasion ratio is large, it is indicated that the front tool face of the tool is abraded greatly, and the surface roughness of a workpiece is affected. According to the method, the requirements for machining precision and surface roughness are comprehensively considered, and accurate detection of the state of the intersection hole boring cutter is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial big data, and more specifically, relates to a method and system for detecting the state of an intersection hole boring tool. Background Art

[0002] Manufacturing is the pillar of the national economy and an important manifestation of the country's comprehensive national strength. As the forefront of the manufacturing industry, the aviation industry plays a leading and guiding role in the country's economic construction and national defense security. It has an extremely important strategic position and comprehensively reflects the strength of the country's manufacturing industry and the level of scientific and technological modernization. As one of the main products of the aviation industry, the level of aircraft manufacturing has become an important indicator for measuring the level of the aviation industry due to its high quality requirements and high technical difficulty. The aircraft manufacturing process can be roughly divided into four stages: blank manufacturing, parts processing, component assembly and testing. As a key link that directly determines the final quality and performance of aircraft products, component assembly has become the core bottleneck of efficiency and quality in the aircraft production process due to its complex coordination relationship, large number of parts and connectors involved, and huge workload.

[0003] The wing-body intersection holes are located on both sides of the aircraft fuselage, used to connect the aircraft fuselage and the wings. They are key high-value-added parts for aircraft assembly and connection. The quality of their processing directly affects the accuracy of the aircraft product appearance, the balance of the wings, and the maneuverability of the aircraft. In the existing wing-body intersection hole processing methods, a conservative trial cutting strategy is generally adopted, and the process of preheating trial cutting, tool adjustment processing, aperture measurement, and updating aperture is implemented. The total number of boring tools in the processing process reaches more than 200, and the processing time accounts for more than 30%. In this process, real-time detection of the intersection hole boring tool status is of great significance.

[0004] Since the processing of the intersection hole between the wing and the fuselage generally requires precision boring, the processing environment is a weak rigid clamping state of the component and there is no cutting fluid, which places extremely high demands on the processing equipment and process capabilities of the production unit, especially the processing accuracy and surface roughness. Among them, too low processing accuracy will lead to the inability of the wing and the fuselage to match, and too high surface roughness will lead to a reduced life of the matching parts between the wing and the fuselage. However, the existing tool status detection methods often only start from the perspective of processing accuracy, and use data-driven models to analyze time series data to detect the wear status of the tool. It is unable to pay full attention to the processing accuracy and surface roughness requirements, and thus is unable to accurately detect the intersection hole boring tool status. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a method and system for detecting the state of an intersection hole boring tool, so as to solve the technical problem that the prior art cannot accurately detect the state of an intersection hole boring tool.

[0006] In order to achieve the above-mentioned purpose, in a first aspect, the present invention provides a method for detecting the state of an intersection hole boring tool, comprising: after each cut of a workpiece is processed, based on the collected processing timing data during the cut and the image of the tool rake face after the cut is completed, performing the following detection operations:

[0007] The processing time series data is input into the flank wear bandwidth detection model to obtain the wear bandwidth of the tool flank; wherein the processing time series data includes: tool cutting force signal, tool vibration signal and machine tool motor current signal; the flank wear bandwidth detection model is a deep learning model;

[0008] Detect the wear area in the tool rake face image, and calculate the area ratio of the wear area to the tool rake face image as the wear ratio of the tool rake face;

[0009] When the obtained wear bandwidth is greater than a preset bandwidth threshold or the obtained wear ratio is greater than a preset ratio, it is determined that the intersection hole boring tool is currently in a damaged state; otherwise, it is determined that the intersection hole boring tool is currently in an undamaged state;

[0010] The position and viewing angle of the image acquisition device relative to the tool are the same each time the image of the tool rake face is acquired.

[0011] Further preferably, the flank wear bandwidth detection model comprises: an MCNN network, an attention mechanism module, a GRU network and a perceptron connected in sequence;

[0012] Among them, the MCNN network is used to extract features of different scales of processing time series data;

[0013] The attention mechanism module is used to calculate the weights of features of different scales of the processed time series data based on the attention mechanism, and perform weighted summation of features of different scales to obtain fused features;

[0014] The GRU network is used to extract the temporal features of the fusion features;

[0015] The perceptron is used to map the time series features to the corresponding tool flank wear bandwidth.

[0016] Further preferably, the GRU network is a BiGRU network.

[0017] Further preferably, the flank wear bandwidth detection model is trained in the following manner:

[0018] Obtain a first training set; the first training set includes: processing time series data of different tools in different tool processing processes, and the corresponding label is the wear bandwidth of the tool back face actually measured;

[0019] Input each processing time series data in the first training set into the flank wear bandwidth detection model to obtain the corresponding tool flank wear bandwidth detection result;

[0020] The flank wear bandwidth detection model is trained by minimizing the difference loss between the detection results of the tool flank wear bandwidth and the corresponding label.

[0021] Further preferably, an image segmentation model is used to detect the worn area in the tool rake face image.

[0022] Further preferably, the image segmentation model comprises: an image encoder, a hint encoder and a mask decoder;

[0023] The image encoder is used to encode the tool rake face image into an image embedding vector;

[0024] The cue encoder is used to encode the foreground prior information of the tool rake face image into a cue embedding vector;

[0025] The mask decoder is used to perform mask processing on the image embedding vector and the prompt embedding vector to obtain a mask image of the wear area in the tool rake face image, and further obtain the wear area in the tool rake face image.

[0026] Further preferably, the above image segmentation model is trained in the following manner:

[0027] Obtain a second training set; the second training set includes: tool rake face images of different tools after each cutting, and the corresponding labels are real mask images of the wear area in the tool rake face images;

[0028] Input each tool rake face image in the second training set into the image segmentation model to obtain a mask image of the corresponding wear area;

[0029] The image segmentation model is trained by minimizing the difference loss between the mask image of the worn area and the corresponding label.

[0030] In a second aspect, the present invention provides a system for detecting the state of an intersection hole boring tool, comprising: a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it executes the method for detecting the state of an intersection hole boring tool provided in the first aspect of the present invention.

[0031] In a third aspect, the present invention provides a monitoring system for the state of an intersection hole boring tool, comprising:

[0032] The data acquisition module is used to collect the processing time sequence data of each cutting process of the workpiece, and the image of the tool rake face after the cutting process is completed; wherein the processing time sequence data includes: tool cutting force signal, tool vibration signal and machine tool motor current signal;

[0033] The state detection module is used to execute the intersection hole boring tool state detection method provided by the first aspect of the present invention.

[0034] In a fourth aspect, the present invention further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is executed by a processor, the device where the storage medium is located is controlled to execute the method for detecting the state of an intersection hole boring tool provided in the first aspect of the present invention.

[0035] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0036] 1. The present invention provides a detection method for the state of an intersection hole boring tool. The detection is based on multimodal data. Through the collected processing timing data during a single-cut processing and the image of the tool front face after the single-cut processing, the wear bandwidth of the tool back face and the wear ratio of the tool front face image are respectively obtained as detection indicators of the intersection hole boring tool state; the wear bandwidth is used to characterize the processing accuracy. When the wear bandwidth is large, it indicates that the tool back face is worn more, affecting the processing accuracy; the wear ratio is used to characterize the surface roughness. When the wear ratio is large, it indicates that the tool front face is worn more, affecting the surface roughness of the workpiece. The present invention takes into account the processing accuracy and surface roughness requirements in all aspects, and realizes accurate detection of the state of the intersection hole boring tool.

[0037] 2. Furthermore, the detection method of the intersection hole boring tool state provided by the present invention, the back tool wear bandwidth detection model includes: an MCNN network, an attention mechanism module, a GRU network and a perceptron connected in sequence. After extracting features of different scales of the processing time series data through the MCNN network, weights are assigned to features of different scales through the attention mechanism module, and fused according to the weights. Finally, the time series features of the fused features are further extracted through the GRU network, thereby realizing deep feature extraction from the spatial scale to the time series dimension, and improving the detection accuracy of the wear bandwidth of the back tool face of the tool.

[0038] 3. Furthermore, considering that during the boring process, the current state information of the tool will be affected by historical data and future operations at the same time, and therefore has a great relationship with the previous and next moment information, in the intersection hole boring tool state detection method provided by the present invention, the back tool wear bandwidth detection model adopts the BiGRU network, which utilizes a bidirectional loop structure to connect the front and rear hidden layers with opposite propagation directions to the same output layer, so that it can process the input sequence from front to back and from back to front, thereby better capturing the contextual information in the sequence and further improving the detection accuracy of the wear bandwidth of the back tool face.

[0039] 4. Furthermore, in the detection method of the intersection hole boring tool state provided by the present invention, the image segmentation model includes: an image encoder, a prompt encoder and a mask decoder; the image of the tool front face is encoded into an image embedding vector by the image encoder, and the foreground prior information of the tool front face image is encoded into a prompt embedding vector by the prompt encoder, and the segmentation mask is predicted based on the image embedding vector and the prompt embedding vector, which makes full use of the prior information and the original image information, and can realize the segmentation of the tool front face image more flexibly and accurately.

[0040] 5. The present invention provides a method for detecting the state of an intersection hole boring tool, which can directly deploy a back tool wear bandwidth detection model locally for use without relying on cloud computing, reducing dependence on network transmission and meeting the requirements for real-time monitoring and fault diagnosis during the processing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A flow chart of a method for detecting the state of an intersection hole boring tool provided by an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of the structure of a flank wear bandwidth detection model provided by an embodiment of the present invention;

[0043] Figure 3 A schematic diagram for verifying the prediction results of the flank wear bandwidth detection model provided by an embodiment of the present invention;

[0044] Figure 4 A schematic diagram of a monitoring system for the state of an intersection hole boring tool provided by an embodiment of the present invention;

[0045] Figure 5 A system architecture diagram of the hardware platform provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0047] In order to achieve the above-mentioned object, in a first aspect, the present invention provides a method for detecting the state of an intersection hole boring tool, such as Figure 1 As shown, it includes: after each cutting of the workpiece, based on the collected processing timing data during the cutting process and the image of the tool rake face after the cutting process is completed, the following detection operations are performed:

[0048] Inputting the processing time series data into the flank wear bandwidth detection model to obtain the wear bandwidth of the tool flank; wherein the processing time series data includes: tool cutting force signal, tool vibration signal and machine tool motor current signal;

[0049] Detect the wear area in the tool rake face image, and calculate the area ratio of the wear area to the tool rake face image as the wear ratio of the tool rake face;

[0050] When the obtained wear bandwidth is greater than a preset bandwidth threshold or the obtained wear ratio is greater than a preset ratio, it is determined that the intersection hole boring tool is currently in a damaged state and needs to be replaced; otherwise, it is determined that the intersection hole boring tool is currently in an undamaged state;

[0051] The position and viewing angle of the image acquisition device relative to the tool are the same each time the image of the tool rake face is acquired.

[0052] It should be noted that, generally, during the machining of the wing-body intersection hole, if the back tool wear value is greater than 0.1 mm or the ratio of the front tool wear area to the tool area exceeds 0.037, it is considered that the tool should be replaced. In an optional embodiment, the preset bandwidth threshold and the preset ratio are 0.1 mm and 0.037, respectively.

[0053] It should be noted that the flank wear bandwidth detection model is a deep learning model, which can be a CNN, LSTM, GRU, Transformer and other models.

[0054] Preferably, in an optional implementation, the flank wear bandwidth detection model comprises: an MCNN network, an attention mechanism module, a GRU network and a perceptron connected in sequence;

[0055] Among them, the MCNN network is used to extract features of different scales of processing time series data;

[0056] The attention mechanism module is used to calculate the weights of features of different scales of the processed time series data based on the attention mechanism, and perform weighted summation of features of different scales to obtain fused features;

[0057] The GRU network is used to extract the temporal features of the fusion features;

[0058] The perceptron is used to map the time series features to the corresponding tool flank wear bandwidth.

[0059] Specifically, the MCNN network includes multiple parallel feature extraction layers, such as Figure 2 In the illustrated implementation, MCNN includes an input layer, a feature extraction layer, and a pooling layer connected in sequence; the feature extraction layer includes three parallel feature extraction layers; the input layer includes a convolutional layer, a BN layer, and a pooling layer connected in sequence; each parallel feature extraction layer includes a convolutional layer, a BN layer, and an activation layer connected in sequence; the convolution kernel sizes of the convolutional layers in the three parallel feature extraction layers are 3*3, 5*5, and 7*7, respectively. In an optional implementation, the convolution layer uses a two-dimensional convolutional layer Conv2d, the pooling layer uses an average pooling layer AvePool, and the activation layer uses a rectified linear unit ReLU. It should be noted that this is only one implementation, and the pooling layer can also use a maximum pooling layer, a global pooling layer, etc., and the activation layer can also use Leaky ReLU, Sigmoid, etc., which are not limited here.

[0060] In this embodiment, three convolution kernel sizes of 3, 5, and 7 are selected. Since the input data is very long, large convolution kernels and pooling layers are applied to the input layer to learn global features. The number of convolution kernels in the feature extraction layer is greater than the number of convolution kernels in the input layer; in this embodiment, the number of convolution kernels in the input layer is 16, and each convolution layer in the feature extraction layer doubles to 32. Finally, three average pooling operations are used for fast downsampling.

[0061] Among them, the convolution layer with a larger convolution kernel is used to obtain a wide receptive field and global features, and the convolution layer with a smaller size is used to learn high-level features; adding a batch normalization layer (BN layer) after each convolution layer can reduce internal covariate shift.

[0062] In this embodiment, a fully connected layer is provided between the MCNN network and the attention mechanism module, and two dropout layers are placed in the fully connected layer to avoid the problem of overfitting.

[0063] The receptive field of the convolution kernel is locally limited, and it usually takes multiple layers of stacking to associate different parts of the image. To solve this problem, a method is designed to implement the attention mechanism through query, key and value by introducing autonomous prompts.1 ,a 2 ,…,a n} is the vectorized input vector, q, k, and v are query, key, and value vectors respectively. Query is an autonomous prompt, which is a feature vector of subjective consciousness; Key is an involuntary prompt, which refers to the prominent feature information vector of an object; and Value is a feature vector representing the object itself. From this perspective, the attention mechanism is to achieve the distribution of attention weights to Value through the convergence of attention between Query and Key, and generate the final output result. 1 As an example, the attention mechanism between a vector and other vectors, 1 The q vector of the input vector is multiplied by the k vector of other input vectors and passed through the softmax layer to obtain a 1 The matching value of other input vectors is multiplied by the weighted v vector of each input vector to obtain the output vector. This process can be described by the formula

[0064]

[0065] Among them, Q, K, V are the matrix representations of q, k, v of the above input vectors, d k It is a scaling factor, which makes the dot product result fall in the area with larger softmax function gradient, thereby accelerating the training process of the model.

[0066] On the basis of the above, it is also necessary to positionally encode each input vector so that the network can learn the positional relationship between the inputs. Position encoding can be implemented in many ways. In this implementation, the classic sin and cos encoding method in the original transformer is used, and its formula is as follows:

[0067]

[0068]

[0069] Among them, PE represents position encoding, pos represents the position of the input vector (i.e., token) to be requested in the entire input vector, the position encoding is a vector (embedding), i represents the index of each element in the vector, and d model The dimension of the vector representing the positional encoding.

[0070] The GRU network uses the state information h of the previous moment t-1 and the input information x at the current time t tSelective retention and utilization are used to obtain tool state degradation information in the data. Generally speaking, in neural networks, information is transmitted from front to back; but in the boring process, the current state information of the tool will be affected by historical data and future operations at the same time, so it has a greater relationship with the previous and later information. Therefore, in an optional implementation, it is preferred to use a BiGRU network. The BiGRU network uses a bidirectional loop structure to connect the two hidden layers with opposite propagation directions to the same output layer, so that it can process the input sequence from front to back and from back to front, thereby better capturing the contextual information in the sequence.

[0071] The update memory formula is as follows:

[0072]

[0073] Among them, z t The closer it is to 1, the higher the h t-1 For the current moment h t The smaller the impact of the calculation; t The closer it is to 0, the more hidden the candidate is at the current moment. The less memory you have, the more you remember about the state just before.

[0074] In an optional implementation manner, the above-mentioned flank wear bandwidth detection model is trained by the following method:

[0075] Obtain a first training set; the first training set includes: processing time series data of different tools in different tool processing processes, and the corresponding label is the wear bandwidth of the tool back face actually measured;

[0076] Input each processing time series data in the first training set into the flank wear bandwidth detection model to obtain the corresponding tool flank wear bandwidth detection result;

[0077] The flank wear bandwidth detection model is trained by minimizing the difference loss between the detection results of the tool flank wear bandwidth and the corresponding label.

[0078] In an optional embodiment, data collection is performed in a horizontal boring process, the tool used is a triangular boring tool, and the workpiece used is a titanium alloy test piece. Vibration, cutting force and current signals are collected by sensors, and then through the AD conversion module, high-quality discrete data of boring processing is finally obtained. The image of the front cutting edge of the tool is obtained by an industrial camera. All signals are collected at a frequency of 10kHz for 50s. The data set is randomly divided into a training set, a validation set and a test set, accounting for 80%, 10% and 10% of the data set respectively. The model is initialized with random weights using the xavier initialization technique, and the model is optimized using the Adam optimizer.

[0079] After training, MAE, MSE, and R 2 The performance of the model is evaluated by the evaluation indicators such as MAE, MSE, R 2 The evaluation indicators reached 6.03, 9.01, and 9.04 respectively. Figure 3 shown.

[0080] It should be noted that there are many ways to detect the wear area in the tool rake face image, such as threshold segmentation or Grabcut algorithm.

[0081] In an optional implementation, an image segmentation model is used to detect the wear area in the tool rake face image, wherein the image segmentation model may be a SAM model, a threshold segmentation model, or the like.

[0082] In addition, image segmentation algorithms can also be used to detect the wear area in the tool rake face image, such as the Grabcut algorithm.

[0083] In an optional embodiment, the image segmentation model includes: an image encoder, a hint encoder, and a mask decoder;

[0084] The image encoder is used to encode the tool rake face image into an image embedding vector;

[0085] The hint encoder is used to encode the foreground prior information of the tool rake face image into a hint embedding vector; wherein the foreground prior information can be a mask matrix, a bounding box or a marking point for identifying the foreground area (or background area), or a text indicating the location of the foreground area (or background area);

[0086] The mask decoder is used to perform mask processing on the image embedding vector and the prompt embedding vector to obtain a mask image of the wear area in the tool rake face image, and further obtain the wear area in the tool rake face image.

[0087] At this time, the ratio of the number of non-zero pixels in the mask image of the wear area to the number of all pixels in the mask image can be calculated as the wear ratio of the tool rake face.

[0088] Preferably, the image encoder is based on a visual transformer pre-trained with visual attention, for encoding the tool rake face image into a graph embedding vector;

[0089] The hint encoder encodes background points, masks, bounding boxes or text into embedding vectors in real time, which is used to generate hint embedding vectors of hints (i.e., positions, texts, etc.) and is fed into the mask decoder together with the image embedding vectors.

[0090] A lightweight mask decoder predicts a segmentation mask based on the embedding vectors from the image encoder and the hint encoder, which maps the image embedding, hint embedding, and output label to the mask and outputs it. In an optional implementation, the mask decoder uses immediate self-attention and cross-attention in both directions (from hint embedding to image embedding and back again). This layout enhances the dataset and allows the model to learn and improve over time, making it efficient and flexible.

[0091] It should be noted that the above-mentioned image segmentation model can directly adopt the existing pre-trained image segmentation model, or can further fine-tune the pre-trained image segmentation model before use, or can be directly trained. Among them, the existing pre-trained image segmentation model has been trained with more than one million images and billions of masks, and its data set has been carefully planned to cover a wide range of fields, objects and scenes, so that the model can be well generalized to different tasks, so that it can return effective mask segmentation for any prompt. Generally speaking, the prompt is the segmentation task, which can be information such as foreground, background points, text, etc. Under the required working conditions of the present invention, it indicates the content to be segmented in the image, that is, the wear area.

[0092] In an optional implementation manner, the above image segmentation model is trained in the following manner:

[0093] Obtain a second training set; the second training set includes: tool rake face images of different tools after each cutting, and the corresponding labels are real mask images of the wear area in the tool rake face images;

[0094] Input each tool rake face image in the second training set into the image segmentation model to obtain a mask image of the corresponding wear area;

[0095] The image segmentation model is trained by minimizing the difference loss between the mask image of the worn area and the corresponding label.

[0096] It should be noted that the above training method can be used for both fine-tuning and training.

[0097] The method for detecting the state of an intersection hole boring tool provided by the present invention can be applied in boring processing scenarios and can be deployed on a laptop computer at a production site to monitor the health state of the tool in real time.

[0098] In a second aspect, the present invention provides a system for detecting the state of an intersection hole boring tool, comprising: a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it executes the method for detecting the state of an intersection hole boring tool provided in the first aspect of the present invention.

[0099] The related technical solution is the same as the method for detecting the state of the intersection hole boring tool provided in the first aspect of the present invention, and will not be described in detail here.

[0100] In a third aspect, the present invention provides a monitoring system for the state of an intersection hole boring tool, such as Figure 4 As shown, including:

[0101] The data acquisition module is used to collect the processing time sequence data of each cutting process of the workpiece, and the image of the tool rake face after the cutting process is completed; wherein the processing time sequence data includes: tool cutting force signal, tool vibration signal and machine tool motor current signal;

[0102] The state detection module is used to execute the intersection hole boring tool state detection method provided by the first aspect of the present invention.

[0103] In an optional embodiment, the data acquisition module is used to collect multi-source data during the processing process, including: a timing signal acquisition module and an image acquisition device; wherein the timing signal acquisition module is used to collect the processing timing data of each cutting process of the workpiece; the image acquisition device is used to collect the tool image after each cutting process is completed.

[0104] Specifically, Figure 5 As shown, the timing signal acquisition module includes: sensors, main control module, analog-to-digital converter; the sensors include: torque sensor, acceleration sensor and current sensor; among them, the torque sensor is installed on the workpiece to collect cutting force signals; the acceleration sensor is installed on the tool handle to collect vibration signals; the current sensor is installed on the machine tool motor to collect current signals. The main control module is used to send start / end signals to each sensor to ensure the synchronization of collected data; the analog-to-digital converter is used to convert the analog signal sent back by the sensor into a digital signal, which is convenient for the computer to read and save. The image acquisition device includes a camera, such as an industrial array camera.

[0105] The related technical solution is the same as the method for detecting the state of the intersection hole boring tool provided in the first aspect of the present invention, and will not be described in detail here.

[0106] In a fourth aspect, the present invention further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is executed by a processor, the device where the storage medium is located is controlled to execute the method for detecting the state of an intersection hole boring tool provided in the first aspect of the present invention.

[0107] The related technical solution is the same as the method for detecting the state of the intersection hole boring tool provided in the first aspect of the present invention, and will not be described in detail here.

[0108] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting the state of an intersection hole boring tool, characterized in that: Including: After each cut of the workpiece is processed, based on the collected processing timing data during the cut and the image of the tool rake face after the cut is completed, the following detection operations are performed: The processing time series data is input into the flank wear bandwidth detection model to obtain the wear bandwidth of the tool flank; the processing time series data includes: tool cutting force signal, tool vibration signal and machine tool motor current signal; the flank wear bandwidth detection model is a deep learning model; Detecting the worn area in the tool rake face image, and calculating the area ratio of the worn area to the tool rake face image as the wear ratio of the tool rake face; When the wear bandwidth is greater than a preset bandwidth threshold or the wear ratio is greater than a preset ratio, it is determined that the intersection hole boring tool is currently in a damaged state; otherwise, it is determined that the intersection hole boring tool is currently in an undamaged state; The position and viewing angle of the image acquisition device used to acquire the image of the front cutting edge of the tool each time are the same relative to the tool.

2. The method for detecting the state of an intersection hole boring tool according to claim 1, characterized in that: The flank wear bandwidth detection model comprises: an MCNN network, an attention mechanism module, a GRU network and a perceptron connected in sequence; The MCNN network is used to extract features of different scales of the processing time series data; The attention mechanism module is used to calculate the weights of the features of different scales of the processed time series data based on the attention mechanism, and perform weighted summation on the features of different scales to obtain fusion features; The GRU network is used to extract the temporal features of the fusion features; The perceptron is used to map the time series features into corresponding tool flank wear bandwidths.

3. The method for detecting the state of an intersection hole boring tool according to claim 2, characterized in that: The GRU network is a BiGRU network.

4. The method for detecting the state of an intersection hole boring tool according to any one of claims 1 to 3, characterized in that: The flank wear bandwidth detection model is trained in the following way: Obtain a first training set; the first training set includes: processing time series data of different tools in different tool processing processes, and the corresponding label is the wear bandwidth of the tool back face actually measured; Input each processing time series data in the first training set into the flank wear bandwidth detection model to obtain a detection result of the wear bandwidth of the corresponding tool flank; The flank wear bandwidth detection model is trained by minimizing the difference loss between the detection result of the wear bandwidth of the tool flank and the corresponding label.

5. The method for detecting the state of an intersection hole boring tool according to claim 1, characterized in that: An image segmentation model is used to detect the worn area in the tool rake face image.

6. The method for detecting the state of an intersection hole boring tool according to claim 5, characterized in that: The image segmentation model includes: an image encoder, a hint encoder and a mask decoder; The image encoder is used to encode the tool rake face image into an image embedding vector; The hint encoder is used to encode the foreground prior information of the tool rake face image into a hint embedding vector; The mask decoder is used to perform mask processing on the image embedding vector and the prompt embedding vector to obtain a mask image of the wear area in the tool rake face image, and further obtain the wear area in the tool rake face image.

7. The method for detecting the state of an intersection hole boring tool according to claim 5 or 6, characterized in that: The image segmentation model is trained in the following way: Acquire a second training set; the second training set includes: tool rake face images of different tools after each cutting, and the corresponding labels are real mask images of the wear area in the tool rake face images; Inputting each tool rake face image in the second training set into the image segmentation model to obtain a mask image of the corresponding wear area; The image segmentation model is trained by minimizing the difference loss between the mask image of the worn area and the corresponding label.

8. A detection system for intersection hole boring tool status, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for detecting the state of an intersection hole boring tool according to any one of claims 1 to 7 of the present invention is executed.

9. A monitoring system for the state of an intersection hole boring tool, characterized in that: include: The data acquisition module is used to collect the processing time sequence data of each cutting process of the workpiece, and the image of the tool rake face after the cutting process is completed; wherein the processing time sequence data includes: tool cutting force signal, tool vibration signal and machine tool motor current signal; the image acquisition device used for collecting the tool rake face image each time has the same position and viewing angle relative to the tool; A state detection module is used to execute the intersection hole boring tool state detection method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed by a processor, the device where the storage medium is located is controlled to execute the method for detecting the state of an intersection hole boring tool according to any one of claims 1 to 7.

Images