CN109822398A - A kind of numerically-controlled machine tool breaking detection system and method based on deep learning - Google Patents
A kind of numerically-controlled machine tool breaking detection system and method based on deep learning Download PDFInfo
- Publication number
- CN109822398A CN109822398A CN201910228970.XA CN201910228970A CN109822398A CN 109822398 A CN109822398 A CN 109822398A CN 201910228970 A CN201910228970 A CN 201910228970A CN 109822398 A CN109822398 A CN 109822398A
- Authority
- CN
- China
- Prior art keywords
- image
- breaking
- cutter
- module
- machine tool
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to cracking of cutter intelligent measurement fields, and specifically disclose a kind of numerically-controlled machine tool breaking detection system and method based on deep learning, it utilizes the video of Tool in Cutting workpiece in image data acquiring module photograph cutting process, the image in video is extracted using pre-processing image data module, and the image of extraction is positioned, cutting and normalized, treated image is received using the edge calculations module for being integrated with breaking arbiter, and breaking is obtained using convolutional neural networks forward inference trained in advance and differentiates result, differentiate that result realizes the control of lathe according to breaking using lathe alarm module.Automatic, real-time, the accurate monitoring of cutting tool state in numerical-controlled machine tool machining process can be achieved in the present invention, has high degree of automation, easy to implement, high accuracy for examination.
Description
Technical field
The invention belongs to cracking of cutter intelligent measurement fields, more particularly, to a kind of numerical control machine based on deep learning
Bed breaking detection system and method.
Background technique
As numerically-controlled machine tool is quickly popularized used in the manufacturing industry, monitored with Intelligentized method, Maintenance CNC lathe
Lasting, the health operation of process, improving numerically-controlled machine tool productivity becomes the important topic in intelligence manufacture field.Machine tool
The real-time detection of fracture is the major issue for needing to solve in numerical control processing, especially in " unmanned factory " or " people's multimachine "
It under production model, if machine tool is broken, needs to find and replaced in time, is otherwise currently processing destruction
With the workpiece of subsequent feeding, generate waste product or defect ware, seriously affect processing efficiency, waste material, improve time and material at
This.
Since every machine has special messenger to be responsible in conventional processes, during machine tool processing, worker can be timely
It was found that the generation of cracking of cutter, and cutter is replaced in time.However, being integrated with automatic material travelling bogie, automatic charging & discharging machine
Device people, workpiece quality are detected etc. automatically in the unmanned factory of equipment, and the real-time detection of cracking of cutter, which becomes, guarantees that process is held
The continuous key being normally carried out.
The existing automatic breaking detection technique huge number of numerically-controlled machine tool is based primarily upon probe displacement, cutting force variation, function
The sensor signals such as rate signal intensity can play preferable detection effect under conditions of restriction, but there is also sensor peaces
Dress is complicated, is easier to affected by noise, the problems such as accuracy in detection is not high.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of numerical control machines based on deep learning
Bed breaking detection system and method realize cutter in numerical-controlled machine tool machining process using image data combination convolutional neural networks
Automatic, real-time, the accurate monitoring of state has high degree of automation, easy to implement, high accuracy for examination.
To achieve the above object, according to one aspect of the present invention, a kind of numerically-controlled machine tool based on deep learning is proposed
Breaking detection system comprising:
Image data acquiring module, for shooting the video of Tool in Cutting workpiece in cutting process;
Pre-processing image data module, for extracting the image in video, and the image of extraction is positioned, cut and
Normalized, then will treated image transmitting into edge calculations module;
It is integrated with the edge calculations module of breaking arbiter, for receiving treated image, and utilizes breaking arbiter
In trained in advance convolutional neural networks forward inference obtain breaking and differentiate result;
Lathe alarm module, for differentiating that result realizes the control of lathe according to breaking.
As it is further preferred that the image that the convolutional neural networks are exported with pre-processing image data module is defeated
Enter, differentiates that result is output with cutter comprising convolutional layer, pond layer and full articulamentum, wherein convolutional layer and pond layer are used for
The feature of input picture is extracted, full articulamentum is used to whether fracture progress to cutter shown in image according to the characteristics of image of extraction
Classification.
As it is further preferred that the positioning position that specially search cutter occurs in the picture comprising manually
Positioning and automatic positioning;Described cut is specially the image being cut out centered on cutter according to the tool position oriented;Institute
State normalized be specially by each pixel point value of the image cut out subtract the mean value of whole image slices vegetarian refreshments again divided by
Variance.
As it is further preferred that the automatic positioning is realized by template matching method or convolutional neural networks method, wherein
Template matching method refers to that the image template that window sliding is carried out in original image using the knife handle image of prior typing matches, matching degree
Highest position is knife handle position, and fixed bias is then recycled to find position of tool tip, and coordinate is cut out in determination;Convolutional Neural net
Network method refers to a large amount of knife handle images one image characteristics extraction device based on convolutional neural networks of training, the image characteristics extraction
Device can judge that the picture of input is the probability of knife handle, then do window sliding in original image using the extractor, find out each
Window picture is the probability of knife handle, is knife handle position at maximum probability, then fixed bias is recycled to find position of tool tip, really
Surely coordinate is cut out.
As it is further preferred that the edge calculations module includes hardware module and software module, wherein the hardware
Module includes computer and router equipped with image processor or tensor processor, and the computer is for providing convolutional Neural
The computing capability of network forward inference, router are image data acquiring module, pre-processing image data for establishing local area network
Transmission method is provided between module, edge calculations module and lathe alarm module;The software module is operate in edge calculations
Program in module for receiving the image of pre-processing image data module offer, and is sent to the convolution of breaking arbiter
Forward inference is carried out in neural network, and the lathe alarm module run in numerically-controlled machine tool is transferred to after obtaining a result.
As it is further preferred that the lathe alarm module is operate in the software module of digital control system, it is used to connect
Receive the breaking that obtains of edge calculations module differentiate as a result, and result manipulation lathe is differentiated according to breaking, be if breaking differentiates result
Breaking then manipulates lathe hard stop, and shows breaking alarm on the human-computer interaction interface of digital control system, if breaking differentiates knot
Fruit is non-breaking, then does not deal with.
It is another aspect of this invention to provide that a kind of numerically-controlled machine tool knife-breaking detecting method based on deep learning is provided,
Include the following steps:
S1 shoots the video of Tool in Cutting workpiece in cutting process;
S2 extracts the image in video, and is positioned, cut and normalized to the image of extraction;
By treated, image inputs in convolutional neural networks trained in advance to obtain breaking differentiation result S3;
S4 differentiates that result realizes the control of lathe according to breaking.
As it is further preferred that being trained using following steps to convolutional neural networks:
1) the unbroken sample of acquisition cutter and cracking of cutter sample, and to the unbroken sample of cutter and cracking of cutter sample into
Row image preprocessing is to construct training set;
2) convolutional neural networks structure is constructed, will be instructed in the sample input convolutional neural networks structure in training set
Practice, distribution of the cracking of cutter probability about input picture is obtained with training.
As it is further preferred that cracking of cutter sample acquires in the following way: being carried out using normal cutter to workpiece
Then cutting fractures cutter since at sampled point using with model when being cut to shutdown at sampled point and removing normal cutter
The case where having run remaining cutting path, being broken at the sampled point when simulating Tool in Cutting, changes one after fractureing cutter
The cracking of cutter sample that the image in video in the section time is used as training.
As it is further preferred that the unbroken sample of cutter acquires in the following way: using normal cutter to workpiece into
Row is primary completely normally to be cut, and each frame image in video is as the unbroken sample of cutter.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, mainly have below
Technological merit:
1. technical solution proposed by the present invention directly sentences method for distinguishing using image, relative to other knife-breaking detecting methods
More direct, more simply, easy to spread, stability is high.
2. technical solution proposed by the present invention uses advanced convolutional neural networks algorithm, the accurate of breaking detection is improved
Rate.
3. technical solution proposed by the present invention introduces the information processing scheme of edge calculations, solves numerically-controlled machine tool and calculate energy
Power lacks bring function restriction, and edge calculations module can equally extend the other function module needed for numerically-controlled machine tool
It calculates, and effectively controls the input cost of system.
Detailed description of the invention
Fig. 1 is the schematic diagram of the numerically-controlled machine tool breaking detection system of the invention based on deep learning;
Fig. 2 is the flow chart of the numerically-controlled machine tool knife-breaking detecting method of the invention based on deep learning;
Fig. 3 a is the cracking of cutter positive sample feature schematic diagram that the convolutional neural networks of the embodiment of the present invention input;
Fig. 3 b is the unbroken negative sample feature schematic diagram of cutter that the convolutional neural networks of the embodiment of the present invention input;
Fig. 4 is that the data of the embodiment of the present invention acquire sampling point distributions schematic diagram, wherein (a)~(c) is cut mark in workpiece
Positive sampling point distributions figure, (d)~(f) are sampling point distributions figure of the cut mark on the left of workpiece, and (g)~(i) is cut mark in work
The subsequent sampling point distributions figure of part;
Fig. 5 is the convolutional neural networks structural schematic diagram of the breaking arbiter of the embodiment of the present invention;
Fig. 6 is the positive and negative sample image after the embodiment of the present invention is cut out, wherein (a) is cracking of cutter positive sample, it (b) is knife
Has unbroken negative sample;
Fig. 7 is that real-time detection on numerically-controlled machine tool of the embodiment of the present invention goes out effect picture shown by cracking of cutter, wherein
(a) image when normally cutting for cutter (b) is cracking of cutter moment image, (c) has been broken for cutter, and system is examined in real time
Measure breaking and in the image of digital control system interface box prompt.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
As shown in Figure 1, a kind of numerically-controlled machine tool breaking detection system based on deep learning provided in an embodiment of the present invention,
Including image data acquiring module, pre-processing image data module, edge calculations module and lathe alarm module, wherein image
Data acquisition module is used to shoot the video of Tool in Cutting workpiece in cutting process;Pre-processing image data module is for mentioning
The image in video is taken, and the image of extraction is pre-processed, including positioning, cutting and normalized, then will be located in advance
Image transmitting after reason is into edge calculations module;Edge calculations module is integrated with breaking arbiter, which will
For the pretreated image transmitting received into breaking arbiter, which utilizes convolutional Neural net trained in advance
Network carries out forward inference to obtain breaking and differentiate as a result, and breaking is differentiated that result is transmitted in lathe alarm module;Lathe report
Alert module is used to receive breaking and differentiates as a result, and differentiating that result realizes the control of lathe according to breaking.
Specifically, image data acquiring module includes camera, bracket, lighting apparatus and camera control interface, wherein camera
For shooting the video image in lathe process;Camera for being fixed in lathe and being directed at cutter by bracket;Illumination is set
It is ready for use on and illuminates inside lathe, especially cutter and workpiece, it is reflective more tight since workpiece and cutter are usually all smooth metals
Weight, thus need using scattering light source;Camera control interface is for directly controlling camera shooting action in computer program
Programming interface.
Further, pre-processing image data module is schemed one by one for extracting from the video that camera takes
Picture, and the pretreatment before input breaking arbiter is carried out to image.The image for inputting breaking arbiter is centered on cutter
One block of image generally comprises the part workpiece near knife handle lower end, completed tool and point of a knife, and pretreatment process includes positioning,
It cuts out, normalization etc..
Wherein, the purpose of positioning is the position searching for cutter in picture and occurring, and provides coordinate to cut out operation.Positioning point
To manually locate and being automatically positioned, manual positioning by observation, artificially finds out position of the cutter in image coordinate system, defeated
Enter to program, the tool position situation constant relative to camera position suitable for process;Automatic positioning passes through template
It is realized with method or convolutional neural networks, need to fix using in original image with tool position relationship, feature is obvious and has uniqueness
Characteristics of image.For this sentences knife handle, template matching method uses the knife handle image of prior typing, (i.e. to be pre-treated in original image
Image) on carry out window sliding image template matching, the highest position of matching degree is knife handle position, then recycles knife
The fixed bias of handle and point of a knife finds position of tool tip, in order to which coordinate is cut out in determination;Convolutional neural networks method is i.e. with a large amount of knives
Handle image trains an image characteristics extraction device based on convolutional neural networks, which can judge that the picture of input is one
The probability of a knife handle, the extractor do window sliding in original image, and the picture for finding out each window is the probability of knife handle, and probability is most
General goal is knife handle position, then the fixed bias of knife handle and point of a knife is recycled to find position of tool tip, in order to which seat is cut out in determination
Mark.And the tool position referred to according to orienting is cut out, the image being cut out centered on cutter, image size is differentiated by breaking
The input requirements of device determine.Normalization operation is that the value of each pixel of image to be processed is subtracted whole image slices to be processed
The mean value of vegetarian refreshments is made again divided by the variance of whole image slices vegetarian refreshments to be processed with realizing the normalization of image with this with image
Data are that the breaking arbiter based on convolutional neural networks of input is easier to training and differentiates.
Further, the function of breaking arbiter is to judge whether cutter fractures by convolutional neural networks algorithm.Volume
Product neural network is a kind of neural network type for being good at the data such as processing image, audio, digital signal, distinctive convolution
The operations such as layer, pond layer are suitble to extract local feature in the data such as image, audio, digital signal, by local feature in sky
Between combination on position and level, understand the relationship inside data between different dimensions and element, data carried out to realize
The functional requirement of the tasks such as classification, detection, segmentation.The convolutional neural networks model of breaking arbiter of the invention is with image data
The image of preprocessing module output is input, and network structure is made of multilayer convolutional layer, pond layer, full articulamentum, and network passes through
Convolutional layer, pond layer extract characteristics of image, full articulamentum according to the characteristics of image of extraction to the cutter in image whether fracture into
Row classification, is divided into fracture and unbroken two class.In order to promote classification accuracy and training speed, present invention preferably uses VGG,
The model parameter that the classics convolutional neural networks structure such as GoogLeNet, ResNet, DenseNet and pre-training are completed carries out image
Then feature extraction is classified using the methods of full articulamentum or support vector machines.
More specifically, the function of edge calculations module is that additional computing capability is provided for numerically-controlled machine tool, for lathe
Real-time judge is carried out to cracking of cutter situation in cutting process, stable, efficient computing resource is provided.The meter of convolutional neural networks
Calculation needs to occupy a large amount of computing resource, and training and reasoning are usually in the calculating that high performance graphics processor (GPU) is housed
It is carried out on machine, and does not have the calculating that extra computing resource supports neural network on the computer of numerically-controlled machine tool usually, thus
Need to provide for numerical control machine lathe be exclusively used in breaking detection etc. be related to the intelligent Applications such as neural network assistant edge calculate
Module.Edge calculations module of the invention is made of hardware module and software module, and hardware module is by being equipped with image processor
(GPU) or the computer of tensor processor (TPU, NPU etc.) and router composition, GPU on computer etc. is for providing convolution
The computing capability of neural network forward inference, router are camera, numerically-controlled machine tool, edge calculations module for establishing local area network
Between transmission method is provided;Software module is operate in the program in edge calculations module, for etc. image data to be received it is pre-
The picture that processing module transmits, the convolutional neural networks for being sent to breaking arbiter carry out forward inference, pass after obtaining a result
Return the lathe alarm module for running on numerically-controlled machine tool.
In addition, lathe alarm module is operate in the software module of digital control system, function is to receive edge calculations module
The breaking passed back differentiates as a result, if breaking differentiates that result is breaking, calls Machine-Tool Control interface, manipulates lathe hard stop,
And breaking alarm is shown on the human-computer interaction interface of digital control system, to notify maintenance personnel to carry out cutter changing, if breaking is sentenced
Other result is not breaking, then does not deal with.
As shown in Fig. 2, the present invention also provides a kind of numerically-controlled machine tool knife-breaking detecting method based on deep learning comprising
Following steps:
S1 shoots the video of Tool in Cutting workpiece in cutting process;
S2 extracts the image in video, and is positioned, cut and normalized;
By treated, image inputs in convolutional neural networks trained in advance to obtain breaking differentiation result S3;
S4 differentiates that result realizes the control of lathe according to breaking.
Specifically, being trained using following steps to convolutional neural networks:
1) the unbroken sample of cutter (the unbroken image of cutter) and cracking of cutter sample (cracking of cutter image) are acquired, and right
The unbroken sample of cutter and cracking of cutter sample carry out image preprocessing to construct training set;
2) convolutional neural networks structure is constructed, will be instructed in the sample input convolutional neural networks structure in training set
Practice, distribution of the cracking of cutter probability about input picture is obtained with training.
Fig. 3 a and Fig. 3 b are respectively cracking of cutter positive sample and the unbroken negative sample spy of cutter of convolutional neural networks input
Schematic diagram is levied, it can be seen from the figure that can indicate that the characteristics of image of cracking of cutter is concentrated mainly at two: first is that the length of cutter
It spends, the cutter length after breaking is shorter than normal cutter, and cutter and workpiece do not contact;Second is that tool position and cutter are on workpiece
Revolution mark end not in same position, this is because cutter is normally to cut before breaking, the path passed by can be left
Revolution mark, after cracking of cutter, cutter and workpiece are not contacted, although the tool marks still in feed, workpiece no longer change.
Fig. 4 is that data acquire sampling point distributions schematic diagram, wherein (a)~(c) is cut mark in the positive sampled point of workpiece point
Cloth, (d)~(f) are sampling point distributions of the cut mark on the left of workpiece, and (g)~(i) is cut mark in the subsequent sampling point distributions of workpiece.
In the actual processing process, cutter may be broken in any position in cutting path, in order to make the characteristics of image near point of a knife
It is more significant, only take the image near point of a knife as the input of convolutional neural networks, thus other than cutter, knife handle feature, make
Different location on cutting path of workpiece and lathe characteristics of image for background be it is different, as shown in Figure 4.Due to volume
The essence of product neural network even depth learning algorithm is learning objective functional value (i.e. the probability value P of cracking of cutter) about input number
According to the distribution of (input picture), thus for the particular process process of specific workpiece, theoretically needed in training set comprising all
Position breaking and the sample for not having breaking, but consider operability, in practical operation, cutter path is subjected to even partition, often
Every 30~50mm, one sampled point is set, for example, in the process of milling cuboid shown in Fig. 4, in milling path
Three sampled points the case where (left before only depicting in figure, rear three faces) is respectively taken on four edges.
In data acquisition, in order to acquire the positive sample of cracking of cutter, first new workpiece is carried out using normal cutter
Cutting, when be cut at sampled point shut down, remove normal cutter, change the cutter that fractures of same model into, from sampled point walk
Complete remaining cutting path changes one after fractureing cutter the case where fracture at the sampled point when simulating Tool in Cutting as a result,
The image in video in the section time can be used as the breaking positive sample that training uses.The unbroken negative sample of cutter in order to obtain,
It needs to carry out primary complete normal cutting to new workpiece with normal cutter, each frame image in video can be used as non-breaking
Negative sample.But due in video consecutive frame image it is too similar, in order to avoid network generate over-fitting, to normal process
Video take a frame to be put into training set every 3~8 frames.
Fig. 5 is the convolutional neural networks topology example figure of breaking arbiter.The input of network is individual pretreated figure
As data, middle layer is that hidden layers, the outputs such as convolution, Chi Hua, full connection are the probability value P for representing cracking of cutter.Pass through input
Sample in training set is trained convolutional neural networks, obtains convolutional neural networks known to parameter, training knot with training
Pretreated tool image need to only be inputted in convolutional neural networks known to the parameter i.e. after beam when carrying out breaking detection
Cracking of cutter probability value P can be obtained, in general, determining cracking of cutter if P >=0.5;If P < 0.5, it is determined as that cutter is unbroken,
I.e. convolutional neural networks known to the parameter can be judged according to the image of input cutter be fracture or it is unbroken.It needs to illustrate
, in order to guarantee the stability determined, wrong report is reduced, in actual use, utilizes convolutional neural networks pair known to parameter
The picture extracted in the cutting video acquired in real time carries out breaking differentiation, when continuous 5~10 picture is determined as breaking, just leads to
Know that lathe is shut down, otherwise, does not regard as true breaking, lathe works on.Classical image classification net is used in the present invention
Network ResNet-V1-50 (K.He, X.Zhang, S.Ren, and J.Sun.Deep residual learning for image
Recognition.In CVPR, 2016.) structure, and migrated the network parameter completed based on ImageNet pre-training and carried out
Network parameter initialization, segmentation carry out four-wheel training, and more open block participate in training every time from back to front, use
Momentum algorithm optimizes network, and learning rate is set as 0.05, and momentum weight is set as 0.9.Specifically how utilize training set
Convolutional neural networks are trained, are the prior arts, this will not be repeated here.
Before sample data enters neural metwork training, a series of images pretreatment, processing method and step are carried out
Method in S2 is consistent, that is, is positioned, cut and normalized.Specifically:
1) it positions
Original image navigates to knife handle position using template matching method, navigates to position of tool tip after biasing is fixed downwards;
2) it cuts
Centered on point of a knife point, it is cut out the image of 140 × 140 pixel sizes;
Picture is subjected to Arbitrary Rotation around center, rotation angle range is preferably [- 4 °, 4 °];
By image cropping to 128 × 128, cutting out center X, Y-axis with respect to original image off-centring range is [- 3,3] mm;
Random to change brightness of image, difference range is [- 3,3], i.e., increases a number simultaneously to pixel each in image,
The range of the number is the arbitrary integer between -3~3;
Random to change picture contrast, the amount of change is set according to actual needs, obtains institute as shown in fig. 6, cutting
The cracking of cutter positive sample and the unbroken negative sample of cutter needed;
3) it normalizes
Each pixel in image is subtracted to the mean value of whole image, then divided by standard deviation, realizes the normalization of image
Processing.
The image preprocessing process of training data effectively improves the generalization ability of network, after the completion of network training, in reality
In use, the image acquired in real time carries out a breaking prediction every 10 frames, the pretreatment before prediction is then directly with point of a knife on border
It is cut out 128 × 128 image centered on point and normalizes.
In embodiments of the present invention, training set includes 11940 positive sample pictures, and 6355 negative sample pictures have been trained
At network comprising 5750 positive samples, 98% accuracy rate is obtained on the verifying collection of 3325 negative samples, is being equipped with
The calculating speed that 25fps is obtained in the edge calculations module of one piece of Intel i7-6700 CPU meets machine tool break detection
Requirement of real-time.
Currently, technology of the present invention is in the 9 type intelligent numerical control of Central China of Wuhan Huazhong Numerical Control Co., Ltd.'s exploitation
It is applied successfully on lathe, Fig. 7 is shown by the present invention cracking of cutter that real-time detection goes out on 9 type intelligent numerical control machine tool of Central China
Effect picture, wherein (a) is cutter image when normally cutting, and (b) is cracking of cutter moment image, (c) be broken for cutter and
System real-time detection is to breaking and the image that is prompted at digital control system interface with box.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (10)
1. a kind of numerically-controlled machine tool breaking detection system based on deep learning characterized by comprising
Image data acquiring module, for shooting the video of Tool in Cutting workpiece in cutting process;
Pre-processing image data module for extracting the image in video, and positions the image of extraction, is cut and normalizing
Change processing, then will treated image transmitting into edge calculations module;
It is integrated with the edge calculations module of breaking arbiter, for receiving treated image, and using pre- in breaking arbiter
First trained convolutional neural networks forward inference obtains breaking and differentiates result;
Lathe alarm module, for differentiating that result realizes the control of lathe according to breaking.
2. the numerically-controlled machine tool breaking detection system based on deep learning as described in claim 1, which is characterized in that the convolution
Neural network is input with the image that pre-processing image data module exports, and differentiates that result is output with cutter comprising convolution
Layer, pond layer and full articulamentum, wherein convolutional layer and pond layer are used to extract the feature of input picture, and full articulamentum is used for basis
The characteristics of image of extraction is classified to whether cutter shown in image fractures.
3. the numerically-controlled machine tool breaking detection system based on deep learning as described in claim 1, which is characterized in that the positioning
The specially position that search cutter occurs in the picture comprising manual positioning and automatic positioning;Described cut is specially basis
The tool position oriented is cut out the image centered on cutter;The normalized is specially the image that will be cut out
Each pixel point value subtracts the mean value of whole image slices vegetarian refreshments again divided by variance.
4. the numerically-controlled machine tool breaking detection system based on deep learning as claimed in claim 3, which is characterized in that described automatic
It is located through template matching method or convolutional neural networks method is realized, wherein template matching method refers to the knife handle using prior typing
Image carries out the image template matching of window sliding in original image, and the highest position of matching degree is knife handle position, then sharp again
Position of tool tip is found with fixed bias, coordinate is cut out in determination;Convolutional neural networks method refers to a large amount of knife handle images training one
Image characteristics extraction device based on convolutional neural networks, the image characteristics extraction device can judge that the picture of input is the general of knife handle
Then rate does window sliding using the extractor in original image, find out the probability that each window picture is knife handle, at maximum probability
As knife handle position, then recycles fixed bias to find position of tool tip, and coordinate is cut out in determination.
5. the numerically-controlled machine tool breaking detection system based on deep learning as described in claim 1, which is characterized in that the edge
Computing module includes hardware module and software module, wherein the hardware module includes handling equipped with image processor or tensor
The computer and router of device, the computer are used for providing the computing capability of convolutional neural networks forward inference, router
It is image data acquiring module, pre-processing image data module, edge calculations module and lathe alarm module in establishing local area network
Between transmission method is provided;The software module is operate in the program in edge calculations module, pre- for receiving image data
The image that processing module provides, and be sent in the convolutional neural networks of breaking arbiter and carry out forward inference, it obtains a result
It is transferred to the lathe alarm module run in numerically-controlled machine tool afterwards.
6. the numerically-controlled machine tool breaking detection system as described in any one in claim 1-5 based on deep learning, which is characterized in that
The lathe alarm module is operate in the software module of digital control system, is used to receive the breaking that edge calculations module obtains and sentences
Not as a result, and according to breaking differentiate result manipulate lathe, if breaking differentiate result be breaking, manipulate lathe hard stop, and
Breaking alarm is shown on the human-computer interaction interface of digital control system, if breaking differentiates that result is non-breaking, is not dealt with.
7. a kind of numerically-controlled machine tool knife-breaking detecting method based on deep learning, which comprises the steps of:
S1 shoots the video of Tool in Cutting workpiece in cutting process;
S2 extracts the image in video, and is positioned, cut and normalized to the image of extraction;
By treated, image inputs in convolutional neural networks trained in advance to obtain breaking differentiation result S3;
S4 differentiates that result realizes the control of lathe according to breaking.
8. the numerically-controlled machine tool knife-breaking detecting method based on deep learning as claimed in claim 7, which is characterized in that using as follows
Step is trained convolutional neural networks:
1) the unbroken sample of acquisition cutter and cracking of cutter sample, and figure is carried out to the unbroken sample of cutter and cracking of cutter sample
As pretreatment is to construct training set;
2) convolutional neural networks structure is constructed, will be trained in the sample input convolutional neural networks structure in training set, with
Training obtains distribution of the cracking of cutter probability about input picture.
9. the numerically-controlled machine tool knife-breaking detecting method based on deep learning as claimed in claim 8, which is characterized in that cracking of cutter
Sample acquires in the following way: being cut using normal cutter workpiece, shuts down and remove just at sampled point when being cut to
Then normal cutter has brought into operation remaining cutting path from sampled point using with the model cutter that fractures, has simulated Tool in Cutting
When the case where being broken at the sampled point, change image in the video in a period of time after fractureing cutter and used as training
Cracking of cutter sample.
10. the numerically-controlled machine tool knife-breaking detecting method based on deep learning as claimed in claim 8, which is characterized in that cutter is not
Fracture sample acquires in the following way: primary complete normal cutting is carried out to workpiece using normal cutter, it is every in video
One frame image is as the unbroken sample of cutter.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910228970.XA CN109822398B (en) | 2019-03-25 | 2019-03-25 | Numerical control machine tool broken cutter detection system and method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910228970.XA CN109822398B (en) | 2019-03-25 | 2019-03-25 | Numerical control machine tool broken cutter detection system and method based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109822398A true CN109822398A (en) | 2019-05-31 |
CN109822398B CN109822398B (en) | 2020-06-02 |
Family
ID=66872225
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910228970.XA Active CN109822398B (en) | 2019-03-25 | 2019-03-25 | Numerical control machine tool broken cutter detection system and method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109822398B (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110405540A (en) * | 2019-07-09 | 2019-11-05 | 华中科技大学 | A kind of artificial intelligence breaking detection system and method |
CN110561192A (en) * | 2019-09-11 | 2019-12-13 | 大连理工大学 | Deep hole boring cutter state monitoring method based on stacking self-encoder |
CN110687871A (en) * | 2019-10-17 | 2020-01-14 | 深圳市鹏云汇科技有限公司 | Efficient intelligent cutter WMS system |
CN111300144A (en) * | 2019-11-25 | 2020-06-19 | 上海大学 | Automatic detection method for tool wear state based on image processing |
CN111390648A (en) * | 2020-03-31 | 2020-07-10 | 江麓机电集团有限公司 | Turning tool abrasion judging method based on antagonistic neural network |
CN112683193A (en) * | 2020-11-06 | 2021-04-20 | 西安交通大学 | Cutter type distinguishing and geometric parameter detecting method and system based on machine vision |
CN113478540A (en) * | 2021-09-08 | 2021-10-08 | 江苏三维智能制造研究院有限公司 | Intelligent equipment system and method thereof |
CN113909995A (en) * | 2021-09-30 | 2022-01-11 | 上汽通用五菱汽车股份有限公司 | Cutter operation analysis system |
CN113941901A (en) * | 2020-07-17 | 2022-01-18 | 智能云科信息科技有限公司 | Machine tool cutter monitoring method and device and electronic equipment |
CN114147541A (en) * | 2021-12-22 | 2022-03-08 | 深圳职业技术学院 | Image acquisition system and method for bottom surface of cutter in numerical control machine tool |
CN114589358A (en) * | 2022-03-22 | 2022-06-07 | 尹亚军 | Cutting device for numerical control machine tool based on big data |
CN115365889A (en) * | 2022-09-17 | 2022-11-22 | 杭州鹏润电子有限公司 | Method, system and storage medium for detecting knife breaking |
CN115847187A (en) * | 2023-02-27 | 2023-03-28 | 成都大金航太科技股份有限公司 | Real-time monitoring system for deep and narrow groove turning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104117876A (en) * | 2013-04-28 | 2014-10-29 | 郑州大学 | Method and device for checking whether cutting tool is broken or abraded or not |
CN107584334A (en) * | 2017-08-25 | 2018-01-16 | 南京航空航天大学 | A kind of complex structural member numerical control machining cutter status real time monitor method based on deep learning |
CN107909081A (en) * | 2017-10-27 | 2018-04-13 | 东南大学 | The quick obtaining and quick calibrating method of image data set in a kind of deep learning |
CN108319962A (en) * | 2018-01-29 | 2018-07-24 | 安徽大学 | A kind of Tool Wear Monitoring method based on convolutional neural networks |
CN109262368A (en) * | 2018-09-13 | 2019-01-25 | 成都数之联科技有限公司 | A kind of tool failure determination method |
KR20190013344A (en) * | 2017-08-01 | 2019-02-11 | 사단법인 캠틱종합기술원 | An Intelligent CNC machine control system for smart monitering, smart diagnosis and smart control by using the physical cutting characteristic map in which the cutting characteristics are mapped in accordance to cutting location in terms of cutting time on working coordinate |
-
2019
- 2019-03-25 CN CN201910228970.XA patent/CN109822398B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104117876A (en) * | 2013-04-28 | 2014-10-29 | 郑州大学 | Method and device for checking whether cutting tool is broken or abraded or not |
KR20190013344A (en) * | 2017-08-01 | 2019-02-11 | 사단법인 캠틱종합기술원 | An Intelligent CNC machine control system for smart monitering, smart diagnosis and smart control by using the physical cutting characteristic map in which the cutting characteristics are mapped in accordance to cutting location in terms of cutting time on working coordinate |
CN107584334A (en) * | 2017-08-25 | 2018-01-16 | 南京航空航天大学 | A kind of complex structural member numerical control machining cutter status real time monitor method based on deep learning |
CN107909081A (en) * | 2017-10-27 | 2018-04-13 | 东南大学 | The quick obtaining and quick calibrating method of image data set in a kind of deep learning |
CN108319962A (en) * | 2018-01-29 | 2018-07-24 | 安徽大学 | A kind of Tool Wear Monitoring method based on convolutional neural networks |
CN109262368A (en) * | 2018-09-13 | 2019-01-25 | 成都数之联科技有限公司 | A kind of tool failure determination method |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110405540A (en) * | 2019-07-09 | 2019-11-05 | 华中科技大学 | A kind of artificial intelligence breaking detection system and method |
CN110405540B (en) * | 2019-07-09 | 2020-07-10 | 华中科技大学 | Artificial intelligence broken cutter detection system and method |
CN110561192A (en) * | 2019-09-11 | 2019-12-13 | 大连理工大学 | Deep hole boring cutter state monitoring method based on stacking self-encoder |
CN110687871A (en) * | 2019-10-17 | 2020-01-14 | 深圳市鹏云汇科技有限公司 | Efficient intelligent cutter WMS system |
CN111300144A (en) * | 2019-11-25 | 2020-06-19 | 上海大学 | Automatic detection method for tool wear state based on image processing |
CN111390648B (en) * | 2020-03-31 | 2021-12-28 | 江麓机电集团有限公司 | Turning tool abrasion judging method based on antagonistic neural network |
CN111390648A (en) * | 2020-03-31 | 2020-07-10 | 江麓机电集团有限公司 | Turning tool abrasion judging method based on antagonistic neural network |
CN113941901A (en) * | 2020-07-17 | 2022-01-18 | 智能云科信息科技有限公司 | Machine tool cutter monitoring method and device and electronic equipment |
CN113941901B (en) * | 2020-07-17 | 2024-04-23 | 智能云科信息科技有限公司 | Machine tool cutter monitoring method, machine tool cutter monitoring device and electronic equipment |
CN112683193A (en) * | 2020-11-06 | 2021-04-20 | 西安交通大学 | Cutter type distinguishing and geometric parameter detecting method and system based on machine vision |
CN113478540A (en) * | 2021-09-08 | 2021-10-08 | 江苏三维智能制造研究院有限公司 | Intelligent equipment system and method thereof |
CN113909995A (en) * | 2021-09-30 | 2022-01-11 | 上汽通用五菱汽车股份有限公司 | Cutter operation analysis system |
CN114147541A (en) * | 2021-12-22 | 2022-03-08 | 深圳职业技术学院 | Image acquisition system and method for bottom surface of cutter in numerical control machine tool |
CN114589358A (en) * | 2022-03-22 | 2022-06-07 | 尹亚军 | Cutting device for numerical control machine tool based on big data |
CN115365889A (en) * | 2022-09-17 | 2022-11-22 | 杭州鹏润电子有限公司 | Method, system and storage medium for detecting knife breaking |
CN115847187A (en) * | 2023-02-27 | 2023-03-28 | 成都大金航太科技股份有限公司 | Real-time monitoring system for deep and narrow groove turning |
CN115847187B (en) * | 2023-02-27 | 2023-05-05 | 成都大金航太科技股份有限公司 | Real-time monitoring system for deep and narrow groove turning |
Also Published As
Publication number | Publication date |
---|---|
CN109822398B (en) | 2020-06-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109822398A (en) | A kind of numerically-controlled machine tool breaking detection system and method based on deep learning | |
CN105044122B (en) | A kind of copper piece surface defect visible detection method based on semi-supervised learning model | |
CN101696945B (en) | On-line detection method of machine vision system to photovoltaic glass flaws | |
CN108805040A (en) | It is a kind of that face recognition algorithms are blocked based on piecemeal | |
CN2932377Y (en) | Flow-type machine visual detector | |
CN103984951B (en) | A kind of Magnetic testing defect inspection method and system | |
CN112184648A (en) | Piston surface defect detection method and system based on deep learning | |
CN101852768A (en) | Workpiece flaw identification method based on compound characteristics in magnaflux powder inspection environment | |
CN101696943B (en) | On-line detection method of machine vision system to medical surgical knife flaws | |
CN109085178B (en) | Defect fingerprint accurate online monitoring and feedback method for additive manufacturing | |
You et al. | Machine vision based adaptive online condition monitoring for milling cutter under spindle rotation | |
CN110728166A (en) | Intelligent monitoring integrated system for confirming track by face recognition | |
CN108931621B (en) | Zinc ore grade soft measurement method based on process texture characteristics | |
CN101696877A (en) | On-line detection method of machine vision system to spring verticality | |
Chen et al. | Research and implementation of robot path planning based on computer image recognition technology | |
Xu et al. | A defect inspection for explosive cartridge using an improved visual attention and image-weighted eigenvalue | |
CN113793385A (en) | Method and device for positioning fish head and fish tail | |
Ramôa et al. | Real-Time 3D door detection and classification on a low-power device | |
CN113333329A (en) | Cigarette defect detection system based on deep learning | |
CN112748116A (en) | Medical gauze surface defect online detection method and device | |
Li et al. | A computational model of vision attention for inspection of surface quality in production line | |
CN116309375B (en) | Method for detecting double-sided defects of solid wood plate and determining intelligent processing coordinates | |
CN114062366B (en) | Online detection method and system for winnowing stem-removing quality in tobacco shred production | |
CN106815922B (en) | A kind of paper money discrimination method and system based on cell phone application and cloud platform | |
Wang et al. | A deep learning-based method for vehicle licenseplate recognition in natural scene |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |