CN110363254A - Part category method for quickly identifying and identification device for crankshaft flexible production line - Google Patents
Part category method for quickly identifying and identification device for crankshaft flexible production line Download PDFInfo
- Publication number
- CN110363254A CN110363254A CN201910737919.1A CN201910737919A CN110363254A CN 110363254 A CN110363254 A CN 110363254A CN 201910737919 A CN201910737919 A CN 201910737919A CN 110363254 A CN110363254 A CN 110363254A
- Authority
- CN
- China
- Prior art keywords
- image
- gray level
- level image
- crankshaft
- point
- 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
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 238000012706 support-vector machine Methods 0.000 claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 14
- 230000004927 fusion Effects 0.000 claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims abstract description 12
- 238000004891 communication Methods 0.000 claims description 42
- 238000001514 detection method Methods 0.000 claims description 35
- 230000000007 visual effect Effects 0.000 claims description 33
- 238000012545 processing Methods 0.000 claims description 16
- 230000009466 transformation Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 4
- 239000008186 active pharmaceutical agent Substances 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000009434 installation Methods 0.000 claims description 3
- 210000000988 bone and bone Anatomy 0.000 claims 1
- 230000000875 corresponding effect Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000006698 induction Effects 0.000 description 2
- 230000001939 inductive effect Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003754 machining Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000005622 photoelectricity Effects 0.000 description 1
- 230000008929 regeneration Effects 0.000 description 1
- 238000011069 regeneration method Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000002195 synergetic effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000007514 turning Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/06—Recognition of objects for industrial automation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention is the part category method for quickly identifying and identification device for crankshaft flexible production line, the step of this method, is: the part image of identification to be sorted is obtained by industrial camera, background operation is removed using image subtraction algorithm to the image of acquisition, workpiece portion is extracted;Burn into expansion, opening operation and closed operation are carried out to workpiece image, obtain gray level image and the sample workpiece gray level image library of identification part to be sorted;The extraction of framework characteristic is carried out using gray level image;The training that AlexNet network model is carried out using gray level image, after network training is good, obtaining one with framework characteristic matrix there is a b of same column to tie up gray level image feature vector;Two parameters are merged by algorithm;The feature of fusion is input to support vector machines to be trained, completes classification.The precision of Classification and Identification is improved, classification speed is fast.
Description
Technical field
It is specifically a kind of to be used for crankshaft flexible production line the present invention relates to flexible production line part category identification technology field
Part category method for quickly identifying and identification device.
Background technique
Important part one of of the crankshaft as automobile engine, manufacturing quality and productivity directly affect auto output and
Overall performance.Various since its structure is complicated, traditional manufacturing technology and processing technology changes, to meet high-volume homology
Column crankshaft product, modified product, regeneration product and new product fast and efficient production, crankshaft flexible production line come into being.It is flexible
Production line generally presses process flow and arranges a series of high speed compound machining centers and efficient special purpose machine tool, and by automatic conveying device
It is connected with manipulator truss, manipulator grabs part and identifies that its type is transported by identification device with being responsible for quickly, efficiently and accurately
To corresponding processing stations.The advantages that machine vision is with non-contact, high reliablity, high degree of automation, is widely used in producing
In the identification of line part image.
The document of patent No. ZL201210297994.9 discloses the Workpiece recognition device on a kind of automatic production line,
The cylinder to connect with air control unit is installed, the piston rod to match with cylinder is connected with movable plate, and movable plate connects in rack
Guiding mechanism, movable plate are equipped with sliding sleeve, are set with spring on the shifting axle to match with sliding sleeve, the one of shifting axle is terminated with
Shifting axle is pressed to impact head side by impact head, spring, and the another of shifting axle is terminated with the induction to match with first close to switch
Head;Shifting axle is multiple groups.After workpiece is sent to certain position by drum-type work transfer device, cylinder movement, drive movable plate to
Workpiece direction is mobile, contacts the impact head of shifting axle front end with workpiece, and the shifting axle after compression is mobile relative to sliding sleeve, bullet
Spring is compressed, and for the inductive head of mobile shaft rear end and first close to switch mutual induction, multiple inductive heads issue multiple signals, control
System can recognize workpiece.The device overall structure is simple, flexible movements, and identification is accurate.But the device recognition speed is slower,
It is not able to satisfy the requirement of real-time, reduces the production efficiency of production line.
The document of patent No. ZL201621015282.3 discloses a kind of part case segmentation identification based on deep learning
Method, including part image acquisition, image preprocessing construct the data set by pretreated image construction;Build part knowledge
Other platform identifies part by Mask R-CNN network, obtains recognition result;Parts Recognition probability threshold value is set, is known
It is to identify successfully that other probability, which is higher than threshold determination, otherwise, it is determined that being recognition failures, it is quasi- to calculate identification for record identification error image
True rate is completed Parts Recognition and image segmentation, and is warned to the wrong part of identification.This method recognition accuracy is high, identification
Efficiency is higher, still, the backbone network of the MaskR-CNN network in this method it is usually used be ResNet101, including
There is a computing module more than 100, arithmetic speed can be relatively slow;Mask R-CNN is related to the data exchange between multiple CPU and GPU, drop
Low system speed and stability.
Summary of the invention
In view of the deficiencies of the prior art, the technical issues of present invention intends to solve is to provide a kind of for crankshaft flexible production
The part category method for quickly identifying and identification device of line.
The technical solution that the present invention solves the technical problem is to provide a kind of part kind for crankshaft flexible production line
The step of class method for quickly identifying, this method, is:
Step 1: obtaining the part image of identification to be sorted by industrial camera, image subtraction is used to the image of acquisition
Algorithm is removed background operation, and workpiece portion is extracted, i.e., sets by the image with crankshaft part and in advance
Subtraction acquisition workpiece image is done without the image of crankshaft part;The existing part image for choosing crankshaft flexible production line simultaneously carries out
Background operation is removed, sample workpiece image library is obtained, the sample size in sample workpiece image library is N, and N is the integer greater than 800;
Step 2: the workpiece image to the first step carries out burn into expansion, opening operation and closed operation, identification to be sorted is obtained
The gray level image of part and sample workpiece gray level image library;
Step 3: carrying out the extraction of framework characteristic using the gray level image in sample workpiece gray level image library:
(1) the Euclidean distance transformation and projective transformation of bianry image (gray level image) are calculated:
One width bianry image I=X ∪ S, X is prospect shape, and S is background, and the Euclidean distance of image I converts DsFor prospect shape
Each point x (x on shape X1,x2) arrive background S shortest distance Ds(x):
Wherein,
Y is the point on background S, and coordinate is (y1, y2),
Projective transformation Ps are as follows:
PS(x)=y ∈ S | d (x, y)=DS(x)}
(2) δ is calculatedx
δxThe radius of the maximum shape distortion for the skeleton branches starting point being connected with x can be regarded as.It enables:
Nx=μ ∈ I | d (x, μ)=1 }
NxFor the direct neighbor point of point x on image.The Euclidean distance of the point x and its consecutive points μ on image are calculated first
Convert Ds(x) and Ds(μ), finds subpointWithTie pointWithObtain line segment Zx,μ, z Zx,μOn point.In order to
It is further reduced calculation amount, is approximately line segment Z by point zx,μMidpoint.DzFor point z to the Euclidean distance of background S, with point z it is
The heart is not r with the radius of the background S greatest circle intersectedx,μ, it is calculated by following formula:
rx,μ=sup { Dz|z∈Zx,μ}
Sup refers to supremum,
Enabling m=(x+ μ)/2 is the midpoint of line segment [x, μ], so as to calculate δx:
ConditionIt can be to avoid the offline section of consecutive points μ ratio xOn perpendicular bisector it is close.
(3) skeleton drawing is obtained
By meeting condition δxThe skeleton drawing MA of skeletal points composition prospect shape X of >=δ a series of, the formula of skeleton drawing MA is such as
Under:
MAδ(x)=x ∈ X | δx≥δ}
Wherein δ is the parameter for being used to eliminate false branch.
Skeleton drawing is expressed as the matrix of an a × b with framework characteristic matrix, and a, b are the integer greater than zero.
Step 4: carrying out the training of AlexNet network model using the gray level image in sample workpiece gray level image library;
Establish the AlexNet network model of deep learning, including 3 convolutional layers and 2 full articulamentums, first convolutional layer
Convolution kernel size be 3 × 3, the convolution kernel size of second convolutional layer is 7 × 7, and the convolution kernel size of third convolutional layer is
13 × 13, the corresponding nonlinear activation function ReLU of each convolution kernel;
LRN (local acknowledgement's normalization) layer in AlexNet network model is removed, the calculating time is reduced and memory disappears
Consumption;Second is that reducing by Dropout layers of number, one Dropout layers are only used in a network, because Dropout will increase convergence
Required the number of iterations, after Dropout layers are added to the last one full articulamentum;Using MAXPooling (maximum pond), enable
The overlapping of pond layer, can reduce over-fitting;
The gray level image in sample workpiece gray level image library is converted 150 × 150 sizes by the characteristics of according to crankshaft image
Image, the input as network;AlexNet network model is instructed with the gray level image in sample workpiece gray level image library
Practice, after network training is good, feature extraction is carried out to gray level image by trained network, obtains one and framework characteristic square
There is battle array a b of same column to tie up gray level image feature vector;
Step 5: Fusion Features Training Support Vector Machines
The gray level image feature of the framework characteristic of third step and the 4th step is subjected to Fusion Features, i.e., by gray level image feature
Vector is accordingly added in the last line in the skeleton drawing eigenmatrix of third step, obtains the matrix of (a+1) × b dimension, will melt
Feature after conjunction, which is input in support vector machines, is trained support vector machines, obtains trained support vector machines (SVM);
Step 6: quickly identification
By the gray level image of the identification part to be sorted of second step according in the skeletal extraction and the 4th step in third step
AlexNet network model carries out gray level image feature extraction, then after carrying out two Fusion Features, is input to trained supporting vector
Machine is classified, output element classification results.
Quick identification of this method for crankshaft on production line, identifies crankshaft part by Fusion Features, knows
Rate is not estimated can achieve 95% or more.
The quick identification device of a kind of part category for crankshaft flexible production line, which is characterized in that the device includes view
Examine gauge head, photoelectric limit switch, communication system, mechanical arm;Communication system is connected by RS232 interface with production lane controller
It connects;Truss is made of three crossbeams and a vertical beam, and three crossbeams are connected perpendicular on the crossbeam of I-shaped centre in I-shaped
Visual examination gauge head is installed in beam, vertical beam lower section, and mechanical arm is mounted on vertical beam, and mechanical arm can be moved up and down with truss or so;
The visual examination gauge head includes image capture module, image processing board and visual examination gauge head power supply, and visual examination gauge head power supply is
Entire visual examination gauge head power supply, visual examination gauge head one end are fixed on the vertical beam of truss by two screws, and other end installation image is adopted
Collect module;Described image acquisition module includes Digital CCD Camera, telecentric lens and LED annular light source;Digital CCD Camera and remote
The connection of heart camera lens;The interface of Digital CCD Camera connects image processing board;Image processing board using DSP as core processor, with
The logic control of FPGA realization circuit;FPGA carry bluetooth and kilomega network interface chip, the kilomega network interface chip of carry to
Complete the communication with Digital CCD Camera;FPGA is interconnected by parallel port and DSP, while DSP and FPGA is sent by serial ports and instructed;
The CCD camera is located at right above crankshaft part;
The crankshaft part is placed in detection station;The detection station includes detection station platform and is symmetricly set on
Two support V-grooves and two fixed V-grooves on detection station platform;The lower part of detection station platform is installed on the slide rail,
Detection station platform can move back and forth on that track under the drive of the motor;Two support V-grooves are respectively positioned on two fixed V-types
The outside of slot;
Photoelectric limit switch is installed, the photoelectric limit switch is used on the crossbeam of the interorbital immediately below CCD camera
Judge detection station whether in-position;
The communication system includes arm processor, communication conversion chip, bluetooth;The communication of visual examination gauge head and communication system
Mode is bluetooth communication, and the communication of communication system and production lane controller uses RS232 communication modes, arm processor and photoelectricity
Limit switch electrical connection.
The CCD camera model MV-EM120M, CCD camera photosensitive area are 4.8mm × 3.6mm;The telecentric lens
Model HX2X-T110;The telecentric lens are zoom lens, focal length 12mm-36mm, minimum object distance 20cm;The LED
Annular light source model HDR-90-45.
Support V-groove 10 centimetres higher than fixed V-groove.
The above-mentioned part category method for quickly identifying of load in DSP.
Compared with prior art, the beneficial effects of the invention are that:
(1) this method is classified using support vector machines, is overcome since by single parameter, (there are two ginsengs by the application
Number, one is part gray level image characteristic parameter, and one is framework characteristic parameter, is merged two parameters by algorithm,
Complete classification.This method merges parameter attribute, only carries out a subseries, the classification time is short, and real-time is good) classify
Issuable error, improves the precision of Classification and Identification, and classification speed is fast.
(2) image is obtained, is handled by the identification device, classification and mechanical arm are integrated together, and visual examination gauge head is directly right
Data are handled, and with communication system by Bluetooth wireless transmission, are then believed by communication system and production lane controller transmitting
Breath, reduces the wiring of production line, meets the requirement quickly handled on production line, is suitble to the workpiece type on flexible production line fast
Speed identification.
(3) the application device connects each section component using truss, and truss adjustable height and spatial position can be suitble to not
The different size of crankshaft part detection of same type different length.
Detailed description of the invention
Fig. 1 is the present invention for the part category method for quickly identifying of crankshaft flexible production line and the whole knot of identification device
Structure schematic diagram;
Fig. 2 is the present invention for the part category method for quickly identifying of crankshaft flexible production line and the detection work of identification device
The schematic diagram of position;
Fig. 3 is the present invention for the part category method for quickly identifying of crankshaft flexible production line and the work shape of identification device
The main view of state;
Fig. 4 is the present invention for the part category method for quickly identifying of crankshaft flexible production line and the work shape of identification device
The side view of state;
Fig. 5 is that the present invention is surveyed for the part category method for quickly identifying of crankshaft flexible production line and the visual examination of identification device
The structural schematic diagram of head;
Fig. 6 is that the present invention is surveyed for the part category method for quickly identifying of crankshaft flexible production line and the visual examination of identification device
The hardware connection diagram of head image processing board;
Fig. 7 is the present invention for the part category method for quickly identifying of crankshaft flexible production line and the communication system of identification device
The hardware structural diagram of system;
Fig. 8 is that part category method for quickly identifying and identification device of the present invention for crankshaft flexible production line are based on depth
The flow chart that the crankshaft part type of study quickly identifies;(in figure: 1, truss;2, mechanical arm;3, visual examination gauge head;4, crankshaft
Part;5, detection station;6, sliding rail;7, photoelectric limit switch;8, bluetooth;201, support slot;301, LED annular light source;302,
Telecentric lens;303, Digital CCD Camera;304, image processing board;305, connector;306, visual examination gauge head power supply;501, it detects
Station platform;502, V-groove is supported;503, fixed V-groove;601, the crossbeam of interorbital)
Specific embodiment
Specific embodiments of the present invention are given below.Specific embodiment is only used for that present invention be described in more detail, unlimited
The protection scope of the application right processed.
The present invention provides a kind of quick identification device of the part category for crankshaft flexible production line (referring to Fig. 1-5), packet
Include visual examination gauge head 3, photoelectric limit switch 7, communication system, production lane controller, mechanical arm 2;Communication system is connect by RS232
Mouth is connected with production lane controller;Truss 1 is made of three crossbeams and a vertical beam, and three crossbeams are in I-shaped, in I-shaped
Vertical beam is connected on crossbeam among type, visual examination gauge head 3 is installed below vertical beam, intermediate crossbeam drives under the drive of the drive
Mechanical arm all around moves, and vertical beam can move up and down, and the structure of specific driving device can use the prior art, such as line
Property mould group;
The visual examination gauge head 3 includes image capture module, image processing board 304, connector 305 and visual examination gauge head power supply
306, visual examination gauge head power supply 306 is the power supply of entire visual examination gauge head, and 3 one end of visual examination gauge head is fixed on the perpendicular of truss by two screws
Liang Shang, the other end install image capture module;Described image acquisition module includes Digital CCD Camera 303,302 and of telecentric lens
LED annular light source 301;Digital CCD Camera 303 is connect with telecentric lens 302;At the interface connection image of Digital CCD Camera 303
Manage plate 304;Image processing board uses the TMS320C6748DSP of TI company as core processor, using Xilinx company
The logic control of Spartan-6FPGA realization circuit;FPGA carry bluetooth and kilomega network interface chip, the gigabit network interface of carry
Chip is to complete the communication with camera;FPGA by parallel port and DSP interconnection be used for image data transmission, while DSP with
FPGA is as shown in Figure 6 by relevant informations, hardware connection diagrams such as serial ports transmission instructions;Select gigabit ethernet interface
Digital CCD Camera 303;The Digital CCD Camera 303, which is located at right above crankshaft part 4, to be shot;
The 303 model MV-EM120M of Digital CCD Camera, resolution ratio 1280*960;The Digital CCD Camera 303
Photosensitive area is 4.8mm × 3.6mm, has continuous, three kinds of Image Acquisition modes of external trigger and soft triggering;
The 302 model HX2X-T110 of telecentric lens;The telecentric lens 302 are zoom lens, focal length 12mm-
36mm, minimum object distance 20cm;
The 301 model HDR-90-45 of LED annular light source, internal diameter 45mm, outer diameter 90mm, power 9W;
The crankshaft part 4 is placed in detection station 5;The detection station 5 is including detection station platform 501 and symmetrically
Two support V-grooves 502 on detection station platform and two fixed V-grooves 503 are set;Pacify the lower part of detection station platform
On sliding rail 6, detection station platform can move back and forth on that track under the drive of the motor;Two support V-grooves 502 are equal
Positioned at the outside of two fixed V-grooves 503, support V-groove 502 about 10 centimetres higher than fixed V-groove 503, two of the inside are solid
Determine V-groove 503 be it is transportable, every 10 centimetres there are two circular hole on detection station platform, V-groove passes through circular hole and screw
It is fixed on detection station platform;The spacing of intermediate fixation V-groove 503 is most short for placing length in a collection of crankshaft part
Crankshaft part;When crankshaft is long, the two sides of the central axis of crankshaft can be placed on support V-groove 502, intermediate
Throw portion can be supported with fixed V-groove 503, reduce the stress among crankshaft, crankshaft can also be made to place more
Stablize;Fixed V-groove 503 can adjust different spacing according to different crankshafts, so that detection station 5 is placed more
The crankshaft part of different length.
On the slide rail, detection station 5 can move back on the sliding rail under the drive of the motor for the lower part installation of detection station 5
Dynamic, the crossbeam 601 of interorbital is used to keep track more firm, installs on the crossbeam 601 of the interorbital immediately below CCD camera
Photoelectric limit switch, the photoelectric limit switch 7 for judge detection station 5 whether in-position;
The communication system includes STM32L0 low-power consumption series A RM processor, communication conversion chip MAX3232, bluetooth;
The communication modes of visual examination gauge head (DSP) and communication system are bluetooth communication, and the communication of communication system and production lane controller uses
RS232 communication modes, arm processor preferably select STM32L051C8T6, and arm processor is electrically connected with photoelectric limit switch,
Hardware annexation is as shown in Figure 7;
The production lane controller controls entire crankshaft flexible production line;Production lane controller is existing structure.
Part for being transported on corresponding processing stations or removing underproof crankshaft part by the mechanical arm 2
Production line, mechanical arm one end are fixed on the vertical beam of truss, and the other end grabs part to be identified.
Part category method for quickly identifying and identification device working principle and work of the present invention for crankshaft flexible production line
It is as process:
Principle: after reaching designated position by the detection station of crankshaft part, photoelectric limit switch 7 is given by communication system
Produce one signal of lane controller;After production lane controller receives signal, is sent and instructed to visual examination gauge head by communication system;
Visual examination gauge head receives acquisition instructions, acquires a frame image, while opening the parallel port in DSP and receiving program, and FPGA enters at this time
Acquisition state;It is converted into simply being pre-processed (filter after parallel data through FPGA by kilomega network by the digital picture of CCD acquisition
Except noise, balance illumination etc.);The image data after simply pre-processing is carried out through FPGA, and DSP, DSP are passed to by parallel interface
Image is handled and (loads identification method in DSP), and feature extraction;By the characteristic value input design instruction of extraction
The support vector machine classifier perfected carries out the classification of correlated characteristic;And by classification results (crankshaft be which kind of crankshaft, according to
Different crankshafts needs to put it on different processing stations and call corresponding processing program, the processing of every kind of crankshaft
Mode is different, and cutting parameter and technological parameter are also different.) logical value FPGA is passed to by serial ports, FPGA passes through communication system
It is sent to production lane controller, production lane controller can control lower one of manufacturing procedure according to the information and call corresponding processing
Program or to the part in one removing command detection station of mechanical arm.
The course of work of identification device of the present invention the following steps are included:
Step 1 is placed in detection station by mechanical arm crawl blank, there is support 502 He of V-groove in detection station
(spacing of intermediate fixation V-groove 503 is for placing the shortest crankshaft zero of length in a collection of crankshaft part to fixed V-groove 503
Part) carry out fixed crankshaft part, detection station 5 is moved to specified position, the i.e. lower section of visual examination gauge head, light under the drive of the motor
Electric limit switch 7 detects that detection station reaches designated position, and being sent to production lane controller one by communication system (can be with
For high and low level) signal;
After step 2, production lane controller receive the signal of photoelectric limit switch 7, by communication system to visual examination gauge head 3
Instruction is sent, visual examination gauge head 3 receives acquisition instructions, and DSP sends the instruction of one frame image of acquisition by serial ports to FPGA, simultaneously
The parallel port for opening DSP receives program, and FPGA enters acquisition state at this time, and FPGA passes through RTL8211 (Ethernet special purpose interface core
Piece) series of instructions are sent for opening camera, it focuses, the parameters such as setting frame per second, delay acquire image, close camera, release
The operation such as camera, carries out Image Acquisition;
Step 3, digital picture input are converted into after parallel data by Ethernet Special Interface Chip RTL8211 through FPGA
Simply pre-processed (filtering, balance illumination etc.);The image data after simply pre-processing is carried out by connecing parallel through FPGA
Oral instructions enter DSP;There is the crankshaft part type method for quickly identifying downloaded in DSP;
The process of crankshaft part type method for quickly identifying is: being removed background to image, workpiece portion is extracted
Come, using image subtraction algorithm, the image with crankshaft part is done into subtraction with the image without crankshaft part taken in advance;
Feature extraction is carried out to image respectively using skeleton method and AlexNet network;Two features of extraction are merged;It will fusion
Feature afterwards, which is input in trained support vector machines, carries out Classification and Identification;It comprises the concrete steps that:
(1) image subtraction algorithm is used, by the image for having crankshaft part and the figure without crankshaft part sequenced in advance
As doing subtraction, workpiece portion is extracted;
(2) burn into expansion, opening operation and closed operation are carried out to the workpiece image extracted, obtains gray level image;
(3) sample selects 1000 gray level images to be trained as gray level image library to network and support vector machines;
1000 training samples carry out the extraction of framework characteristic and the training of convolutional neural networks respectively;
Carry out skeletal extraction:
1. calculating the Euclidean distance transformation and projective transformation of bianry image (gray level image):
One width bianry image I=X ∪ S, X is prospect shape, and S is background, and the Euclidean distance of image I converts DsFor prospect shape
Each point x (x on shape X1,x2) arrive background S the shortest distance:
Wherein,
Y is the point on background S, i.e. y1Corresponding x1, y2Corresponding x2。
Projective transformation are as follows:
PS(x)=y ∈ S | d (x, y)=DS(x)}
2. calculating δx
δxThe radius of the maximum shape distortion for the skeleton branches starting point being connected with x can be regarded as.It enables:
Nx=μ ∈ I | d (x, μ)=1 }
NxFor the direct neighbor point of point x on image.The Euclidean distance of the point x and its consecutive points μ on image are calculated first
Convert Ds(x) and Ds(μ), finds subpointWithTie pointWithObtain line segment Zx,μ, z Zx,μOn point.For
It is further reduced calculation amount, by point z is approximately line segment Zx,μMidpoint.DzFor point z to the Euclidean distance of background S, it is with point z
Center is not r with the radius of the background S greatest circle intersectedx,μ, it is calculated by following formula:
rx,μ=sup { Dz|z∈Zx,μ}
Enabling m=(x+ μ)/2 is the midpoint of line segment [x, μ], so as to calculate δx:
ConditionIt can be to avoid the offline section of consecutive points μ ratio xOn perpendicular bisector it is close.
3. obtaining skeleton drawing
By meeting condition δxThe skeleton drawing MA of skeletal points composition prospect shape X of >=δ a series of, the formula of skeleton drawing MA is such as
Under:
MAδ(x)=x ∈ X | δx≥δ}
Wherein, δ is the parameter for being used to eliminate false branch.
Skeleton drawing is expressed as the matrix of an a × b with framework characteristic matrix, and a, b are the integer greater than zero.
Gray level image feature extraction:
The base that the embodiment of the present invention has selected the classical model AlexNet of deep learning to identify as convolutional neural networks
Plinth;AlexNet model joined Dropout layers after full articulamentum to prevent over-fitting, use MAXPooling, enable pond
Layer overlapping, can reduce over-fitting;Nonlinear activation function ReLU is used, with traditional activation primitive sigmoid, tanh phase
Than ReLU mathematical formulae is simple, only one threshold value 0, calculating speed is fast, does not have the phenomenon that gradient disperse, and make
When optimizing network with stochastic gradient descent method, convergence ratio is very fast;The mathematic(al) representation of ReLU function are as follows:
The characteristics of on the one hand the present embodiment is needed according to crankshaft image, to input picture size, the number of plies of convolutional layer and volume
Product core size does a little adjustment;It on the other hand is exactly that following optimization is carried out to the network itself: first is that by AlexNet network model
LRN (local acknowledgement's normalization) layer remove, reduce and calculate time and memory consumption;Second is that reduce by Dropout layers of number,
One Dropout layers are only used in network, because Dropout will increase the required the number of iterations of convergence, Dropout layers are added to
After the last one full articulamentum;The principle of Dropout are as follows: in each training process, Dropout algorithm can be with certain probability
Randomly make the output valve 0 of a part of neuron node, i.e. a part of neuron of random drop, it is multiple on training data to prevent
Miscellaneous synergistic effect, to improve network generalization;
On the basis of guaranteeing the detailed information of input picture, in order to reduce data volume, collected crankshaft image is turned
The image for turning to 150 × 150 sizes, the input as network;
Using 3 convolutional layers, 2 full articulamentums in the present embodiment method, the convolution kernel size of first convolutional layer is 3 ×
3, second convolution kernel size is 7 × 7, and third convolution kernel size is 13 × 13, and each convolution kernel corresponding one non-linear sharp
Function ReLU living, it is possible to reduce number of parameters;Traditional AlexNet network has 5 convolutional layers, 3 full articulamentums, and traditional
AlexNet network, by adjusting convolution kernel size, the number of convolutional layer and full articulamentum, is guaranteeing to know compared to the present embodiment method
In the case where other precision, reduce the time of calculating;
AlexNet network is trained by 1000 training samples, after network training is good, by network to gray scale
Image carries out feature extraction;The feature of extraction is the vector that a b for having same column with framework characteristic matrix is tieed up;
Two kinds of features are merged, the matrix of (a+1) × b dimension is obtained, fused feature is input to support
Support vector machines is trained in vector machine;After support vector machines and AlexNet network have all trained, so that it may download
Identification into image processing board for crankshaft on production line;
Step 4 inputs step 3 pretreated digital picture in crankshaft part type method for quickly identifying, to image
After carrying out background operation, framework characteristic extraction and above-mentioned AlexNet network are carried out using above-mentioned skeletal extraction mode
After carrying out gray level image feature extraction, then carry out two Fusion Features, be input to after Fusion Features trained support vector machines into
Row classification, output element classification results.Classification results are sent to production lane controller by communication system, and production lane controller can
To control the corresponding processing program of lower one of manufacturing procedure calling according to the information or give mechanical arm one removing command detection
Part on station.
Crankshaft part is identified by Fusion Features, discrimination can achieve 95% or more.
The present invention does not address place and is used in the prior art.
Claims (6)
1. a kind of the step of part category method for quickly identifying for crankshaft flexible production line, this method, is:
Step 1: obtaining the part image of identification to be sorted by industrial camera, image subtraction algorithm is used to the image of acquisition
Be removed background operation, workpiece portion extracted, i.e., by with crankshaft part image in advance set without
The image of crankshaft part does subtraction and obtains workpiece image;The existing part image for choosing crankshaft flexible production line simultaneously carry on the back
Scape operation obtains sample workpiece image library, and the sample size in sample workpiece image library is N, and N is the integer greater than 800;
Step 2: the workpiece image to the first step carries out burn into expansion, opening operation and closed operation, identification part to be sorted is obtained
Gray level image and sample workpiece gray level image library;
Step 3: carrying out the extraction of framework characteristic using the gray level image in sample workpiece gray level image library, skeleton drawing, bone are obtained
Frame figure is expressed as the matrix of an a × b with framework characteristic matrix, and a, b are the integer greater than zero;
Step 4: carrying out the training of AlexNet network model using the gray level image in sample workpiece gray level image library;
Establish the AlexNet network model of deep learning, including 3 convolutional layers and 2 full articulamentums, the volume of first convolutional layer
Product core size is 3 × 3, and the convolution kernel size of second convolutional layer is 7 × 7, the convolution kernel size of third convolutional layer for 13 ×
13, the corresponding nonlinear activation function ReLU of each convolution kernel;
LRN layer in AlexNet network model is removed, only uses one Dropout layers, Dropout layers are added to last
After a full articulamentum;Using maximum pond MAXPooling, pond layer is enabled to be overlapped;
The characteristics of according to crankshaft image, converts the gray level image in sample workpiece gray level image library in the figure of 150 × 150 sizes
Picture, the input as network;AlexNet network model is trained with the gray level image in sample workpiece gray level image library,
After network training is good, feature extraction is carried out to gray level image by trained network, obtains one and framework characteristic matrix
A b with same column ties up gray level image feature vector;
Step 5: Fusion Features Training Support Vector Machines
The gray level image feature of the framework characteristic of third step and the 4th step is subjected to Fusion Features, i.e., by gray level image feature vector
The last line being accordingly added in the skeleton drawing eigenmatrix of third step obtains the matrix of (a+1) × b dimension, after fusion
Feature be input in support vector machines support vector machines be trained, obtain trained support vector machines;
Step 6: quickly identification
The gray level image of the identification part to be sorted of second step is carried out to the AlexNet in framework characteristic extraction and the 4th step respectively
Network model carries out gray level image feature extraction, then after carrying out two Fusion Features, is input to trained support vector machines and carries out
Classification, output element classification results.
2. part category method for quickly identifying according to claim 1, which is characterized in that third step middle skeleton feature extraction
The step of be:
(1) the Euclidean distance transformation and projective transformation of bianry image (gray level image) are calculated:
One width bianry image I=X ∪ S, X is prospect shape, and S is background, and the Euclidean distance of image I converts DsFor on prospect shape X
Each point x (x1,x2) arrive background S shortest distance Ds(x):
Wherein,
Y is the point on background S, and coordinate is (y1, y2),
Projective transformation Ps are as follows:
PS(x)=y ∈ S | d (x, y)=DS(x)}
(2) the radius δ of the maximum shape distortion for the skeleton branches starting point being connected with x is calculatedx, enable
Nx=μ ∈ I | d (x, μ)=1 }
NxFor the direct neighbor point of point x on image, first the Euclidean Distance Transform D of the point x on calculating image and its consecutive points μs
(x) and Ds(μ), finds subpointWithTie pointWithObtain line segment Zx,μ, z Zx,μOn point;Point z is approximate
For line segment Zx,μMidpoint, DzFor point z to the Euclidean distance of background S, centered on point z, not with the background S greatest circle intersected
Radius is rx,μ, rx,μBy rx,μ=sup { Dz|z∈Zx,μBe calculated,
Enabling m=(x+ μ)/2 is the midpoint of line segment [x, μ], to calculate δx:
(3) skeleton drawing is obtained
By meeting condition δxThe skeleton drawing MA of skeletal points composition prospect shape X of >=δ a series of, the formula of skeleton drawing MA are as follows:
MAδ(x)=x ∈ X | δx≥δ}
Wherein δ is the parameter for being used to eliminate false branch.
3. a kind of quick identification device of part category for crankshaft flexible production line, which is characterized in that the device includes visual examination
Gauge head, photoelectric limit switch, communication system, mechanical arm;Communication system is connected by RS232 interface with production lane controller
It connects;Truss is made of three crossbeams and a vertical beam, and three crossbeams are connected perpendicular on the crossbeam of I-shaped centre in I-shaped
Visual examination gauge head is installed in beam, vertical beam lower section, and mechanical arm is mounted on vertical beam, and mechanical arm can be moved up and down with truss or so;
The visual examination gauge head includes image capture module, image processing board and visual examination gauge head power supply, and visual examination gauge head power supply is entire
The power supply of visual examination gauge head, visual examination gauge head one end are fixed on the vertical beam of truss by two screws, and the other end installs Image Acquisition mould
Block;Described image acquisition module includes Digital CCD Camera, telecentric lens and LED annular light source;Digital CCD Camera and telecentric mirror
Head connection;The interface of Digital CCD Camera connects image processing board;Image processing board is using DSP as core processor, with FPGA reality
The logic control of existing circuit;FPGA carry bluetooth and kilomega network interface chip, the kilomega network interface chip of carry to complete with
The communication of CCD camera;FPGA is interconnected by parallel port and DSP, while DSP and FPGA is sent by serial ports and instructed;The CCD camera
Right above crankshaft part;
The crankshaft part is placed in detection station;The detection station includes detection station platform and is symmetricly set on detection
Two support V-grooves and two fixed V-grooves on station platform;The lower part installation of detection station platform on the slide rail, detects
Station platform can move back and forth on that track under the drive of the motor;Two support V-grooves are respectively positioned on two fixed V-grooves
Outside;
Photoelectric limit switch is installed, the photoelectric limit switch is for judging on the crossbeam of the interorbital immediately below CCD camera
Detection station whether in-position;
The communication system includes arm processor, communication conversion chip, bluetooth;The communication modes of visual examination gauge head and communication system
For bluetooth communication, the communication of communication system and production lane controller uses RS232 communication modes, arm processor and photoelectric position limit
Switch electrical connection.
4. the quick identification device of part category according to claim 3, which is characterized in that the CCD camera model MV-
EM120M, CCD camera photosensitive area are 4.8mm × 3.6mm;The telecentric lens model HX2X-T110;The telecentric lens
For zoom lens, focal length 12mm-36mm, minimum object distance 20cm;The LED annular light source model HDR-90-45.
5. the quick identification device of part category according to claim 3, which is characterized in that support V-groove is than fixed V-groove
It is 10 centimetres high.
6. the quick identification device of part category according to claim 3, which is characterized in that in DSP load claim 1 or
Part category method for quickly identifying described in 2.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910737919.1A CN110363254B (en) | 2019-08-12 | 2019-08-12 | Quick identification device for types of parts of flexible crankshaft production line |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910737919.1A CN110363254B (en) | 2019-08-12 | 2019-08-12 | Quick identification device for types of parts of flexible crankshaft production line |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110363254A true CN110363254A (en) | 2019-10-22 |
CN110363254B CN110363254B (en) | 2024-02-02 |
Family
ID=68224356
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910737919.1A Active CN110363254B (en) | 2019-08-12 | 2019-08-12 | Quick identification device for types of parts of flexible crankshaft production line |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110363254B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111331599A (en) * | 2020-03-02 | 2020-06-26 | 上海媒智科技有限公司 | Automatic directional article grabbing method and system based on mechanical arm |
CN112733694A (en) * | 2021-01-04 | 2021-04-30 | 中南大学 | ORB feature-based track laying roller identification method and system |
CN114022748A (en) * | 2022-01-06 | 2022-02-08 | 深圳市海清视讯科技有限公司 | Target identification method, device, equipment and storage medium |
CN114558820A (en) * | 2022-02-28 | 2022-05-31 | 重庆长安汽车股份有限公司 | Flexible tray for crankshaft workpiece suitable for robot cleaning machine |
CN114918659A (en) * | 2022-06-29 | 2022-08-19 | 中国电子科技集团公司第十四研究所 | Flexible high-precision control method for assembling joint assembly |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018000731A1 (en) * | 2016-06-28 | 2018-01-04 | 华南理工大学 | Method for automatically detecting curved surface defect and device thereof |
CN108133188A (en) * | 2017-12-22 | 2018-06-08 | 武汉理工大学 | A kind of Activity recognition method based on motion history image and convolutional neural networks |
CN108655826A (en) * | 2018-06-25 | 2018-10-16 | 河北工业大学 | A kind of tool wear visual examination device and visual examination method for numerical control workshop |
CN108918540A (en) * | 2018-09-20 | 2018-11-30 | 苏州华维诺智能科技有限公司 | Capture and detection device and method based on mechanical arm camera shooting vision to product parts |
CN109117826A (en) * | 2018-09-05 | 2019-01-01 | 湖南科技大学 | A kind of vehicle identification method of multiple features fusion |
WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
CN210006066U (en) * | 2019-08-12 | 2020-01-31 | 河北工业大学 | part type recognition device for crankshaft flexible production line |
-
2019
- 2019-08-12 CN CN201910737919.1A patent/CN110363254B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018000731A1 (en) * | 2016-06-28 | 2018-01-04 | 华南理工大学 | Method for automatically detecting curved surface defect and device thereof |
WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
CN108133188A (en) * | 2017-12-22 | 2018-06-08 | 武汉理工大学 | A kind of Activity recognition method based on motion history image and convolutional neural networks |
CN108655826A (en) * | 2018-06-25 | 2018-10-16 | 河北工业大学 | A kind of tool wear visual examination device and visual examination method for numerical control workshop |
CN109117826A (en) * | 2018-09-05 | 2019-01-01 | 湖南科技大学 | A kind of vehicle identification method of multiple features fusion |
CN108918540A (en) * | 2018-09-20 | 2018-11-30 | 苏州华维诺智能科技有限公司 | Capture and detection device and method based on mechanical arm camera shooting vision to product parts |
CN210006066U (en) * | 2019-08-12 | 2020-01-31 | 河北工业大学 | part type recognition device for crankshaft flexible production line |
Non-Patent Citations (1)
Title |
---|
王晓倩等: "基于骨架特征的曲轴快速辨识方法研究", 组合机床与自动化加工技术, vol. 530, no. 4, pages 117 - 118 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111331599A (en) * | 2020-03-02 | 2020-06-26 | 上海媒智科技有限公司 | Automatic directional article grabbing method and system based on mechanical arm |
CN112733694A (en) * | 2021-01-04 | 2021-04-30 | 中南大学 | ORB feature-based track laying roller identification method and system |
CN114022748A (en) * | 2022-01-06 | 2022-02-08 | 深圳市海清视讯科技有限公司 | Target identification method, device, equipment and storage medium |
CN114022748B (en) * | 2022-01-06 | 2022-04-08 | 深圳市海清视讯科技有限公司 | Target identification method, device, equipment and storage medium |
CN114558820A (en) * | 2022-02-28 | 2022-05-31 | 重庆长安汽车股份有限公司 | Flexible tray for crankshaft workpiece suitable for robot cleaning machine |
CN114918659A (en) * | 2022-06-29 | 2022-08-19 | 中国电子科技集团公司第十四研究所 | Flexible high-precision control method for assembling joint assembly |
CN114918659B (en) * | 2022-06-29 | 2024-01-16 | 中国电子科技集团公司第十四研究所 | Flexible high-precision control method for assembling joint assembly |
Also Published As
Publication number | Publication date |
---|---|
CN110363254B (en) | 2024-02-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110363254A (en) | Part category method for quickly identifying and identification device for crankshaft flexible production line | |
CN109483573A (en) | Machine learning device, robot system and machine learning method | |
CN111046948B (en) | Point cloud simulation and deep learning workpiece pose identification and robot feeding method | |
CN109013405A (en) | It is a kind of independently detected with cast(ing) surface and substandard products sorting function robot system | |
CN108109174A (en) | A kind of robot monocular bootstrap technique sorted at random for part at random and system | |
CN111007073A (en) | Method and system for online detection of part defects in additive manufacturing process | |
CN106853639A (en) | A kind of battery of mobile phone automatic assembly system and its control method | |
CN111266304A (en) | Coal briquette identification, detection and sorting system | |
CN109927033B (en) | Target object dynamic adaptation method applied to conveyor belt sorting | |
CN110428464B (en) | Multi-class out-of-order workpiece robot grabbing pose estimation method based on deep learning | |
CN110633738B (en) | Rapid classification method for industrial part images | |
CN113267452A (en) | Engine cylinder surface defect detection method and system based on machine vision | |
CN107167072A (en) | A kind of vision detection system corrected for LED filament spot welding and detection method | |
CN112718552A (en) | Automatic quality defect inspection device for LED circuit board and working method thereof | |
CN112643207A (en) | Laser automatic derusting system and method based on computer vision | |
CN107344171A (en) | A kind of low-voltage air switch the System of Sorting Components and method based on Robot Visual Servoing | |
CN108156359A (en) | Intelligent industrial camera | |
CN113744247A (en) | PCB welding spot defect identification method and system | |
CN109454501A (en) | A kind of lathe on-line monitoring system | |
CN207366467U (en) | A kind of surface defect real-time detection apparatus for increasing material manufacturing | |
CN210006066U (en) | part type recognition device for crankshaft flexible production line | |
CN112975970A (en) | Vision grabbing mechanical arm system | |
WO2020241540A1 (en) | System control method and system | |
CN112699903A (en) | Full-process visual tracking control method and system for motorcycle engine assembly | |
CN209408173U (en) | A kind of robot polishing system |
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 |