CN102680102A - Automatic detection method of solar silicon chip colors based on machine vision - Google Patents

Abstract

The invention relates to the machine vision detection field, and discloses a method and a device for detecting surface color defects of solar silicon chips through the machine vision technology. The detection object is distribution information of the surface colors of the silicon chips after a process of plasma enhanced chemical vapor deposition (PECVD). The classifying scheme includes using an image filtering technique, a color image segmentation technique, a color image hue, saturation, and lightness (HIS) spatial analysis technique and the like for processing and analyzing collected images; extracting feature information of the images; classifying the silicon chips into good silicon chips and defect silicon chips according to surface quality after training and testing of the information through a mode identifying method which combines a support vector machine with an improved gravitational search algorithm; sending a classifying result to an executive mechanism and controlling a uniaxial robot to take out the defect silicon chips. According to the method and the device, mechanism is simple, maintenance and operation are facilitated, and detecting requirements can be satisfied.

Description

Solar energy silicon crystal chip color automatic testing method based on machine vision

Technical field

The present invention relates to a kind of solar energy silicon crystal chip color Automatic Measurement Technique based on machine vision; Be used for the defective silicon wafer that online automatic detection goes out surface generation colour deficient after process plasma enhanced chemical vapor deposition (Plasma Enhanced Chemical Vapor Deposition is called for short PECVD) operation.

Background technology

Silicon wafer is the core material of solar cell, and therefore the good and bad directly decision Solar cell performance of its quality must be rejected when detecting for the silicon wafer that has colour deficient.Characteristics such as the diversity that (comprises jaundice sheet, rubescent, the sheet that turns white, small particles, roller seal, fingerprint color spot, boat pollution color spot, chemical residual article color spot, metaphosphoric acid color spot, aluminium film and the jaundice of technology round dot etc.) because the silicon wafer surface colour deficient, complicacy cause that the research of its detection method is not had substantial progress always; Present many manufacture of solar cells producer is main with the artificial visual detection still mainly; This detection method is owing to there being problems such as instability, high fragment rate and low rate, so can't reach the requirement of modern industry production operation.

Summary of the invention

Deficiency to above technology; The object of the present invention is to provide a kind of stability height, noncontact, two-forty to reach device based on the method for the silicon wafer surface color detection of machine vision; It can real-time online, the stable silicon wafer surface colour deficient that detects accurately and efficiently; Automatically take out faulty goods, automatic statistical analysis silicon chip surface quality shows testing result in real time.

For realizing above-mentioned purpose, the present invention adopts following pick-up unit: the silicon wafer surface color detection system based on machine vision comprises material loading station, vision station and silicon chip sortation station.Said vision station device mainly comprises light source, high speed camera and industrial computer; Said industrial computer comprises the central processing unit and the display of built-in image processing software; Said light source adopts four bar shaped white fluorescent lamps, and it is evenly distributed on the top of lamp box, and the following use diffuse reflector of lamp box hides; Said high speed camera adopts area array CCD camera, is fixed on the cross bar of frame, and through interface and said industrial computer communication on the camera, adopts the 1394B communication mode; Said silicon chip sortation station device mainly is the single shaft robot.

The present invention provides the detection method of a kind of silicon wafer surface color detection system, may further comprise the steps:

(1) magazine that will fill silicon wafer places assigned address, starting outfit;

(2) put in place signal triggering ccd image harvester action of silicon chip, and the silicon wafer surface image of gathering is sent to graphics processing unit;

(3) after graphics processing unit adopts image filtering technology, color Image Segmentation, coloured image HSI spatial analysis technology etc. to handle the image of gathering in the step (2), and provide process result;

(4) with the feature extraction of the process of the result in the step (3), be converted into digital signal;

(5) with gained result in the step (4) through based on behind SVMs and the mode identification method training and testing that improved gravitation searching algorithm combines, the silicon wafer surface quality is divided into two types of good silicon wafer and defect silicon wafers.

(6) classification results is sent to action actuating mechanism, robot takes out the defective silicon chip by PLC control single shaft.

Improved gravitation searching algorithm described in the step (5) is meant the gravitation searching algorithm based on policy feedback; In algorithm, introduce the entropy of population distribution and the diversity that the averaged particles range index is described the particle population; Be used as measure function with this, improve the exploration and the development ability of algorithm during evolution.In step (5), utilize the nuclear parameter in above-mentioned the improved gravitation search algorithm optimizes SVMs, make up pattern recognition unit.

Beneficial effect of the present invention: utilize machine vision technique to gather the image information of silicon wafer surface fast; Real-time online is stablized the defect recognition that precise and high efficiency carries out silicon wafer; This system also establishes automatic warning; Automatic statistical analysis silicon chip surface quality shows testing result in real time, realizes the robotization that the identification of silicon wafer colour deficient is handled.

Description of drawings

Fig. 1 outfit of equipment layout

Fig. 2 system global structure figure

Fig. 3 image classification process flow diagram

Numbering 1 is an air knife among Fig. 1, the 2nd, and sucker, the 3rd, cylinder, the 4th, the translation module, numbering 5 is high speed cameras, the 6th, fluorescent light, numbering 7 is that the single shaft robot is the sub-material module, the 8th, the buffering pipeline.

Embodiment

Further specify below in conjunction with the accompanying drawing specific embodiments of the invention.

Automatic testing process: the magazine that manual work will be filled silicon chip is positioned over assigned address; Press the device start button; Cylinder is pulled to material level with magazine; Jacking silicon chip servo action is moved on the silicon chip; The induction of correlation optoelectronic switch; Jacking puts in place; Laterally move the module action to grasping the position; Sucker work; Blow the action of material air knife; Jacking silicon chip module descends; Cylinder moves down; Draw silicon chip; Move on the cylinder; Module moves to the blowing station; Cylinder moves down; Sucker stops suction; Silicon slice placed places on the driving belt line; Silicon chip moves to phase machine testing station; The camera detection of taking pictures; Detection finishes; Silicon chip moves to the check strap line; Silicon chip moves to sortation station; Classification die set, cylinder, sucker action are classified to silicon chip and specify in the charging tray; The device motion circulation.Magazine completely sends induced signal; Equipment downtime; Press ACK button; Pull out magazine and get material; Getting material finishes; Press reset button; Magazine is pushed into the branch material level; Induction of signal; The latch location; Close the door.

Is image file with the ccd image capture card with the two-dimensional matrix stored in form; Graphics processing unit adopts 5 * 5 median filtering algorithms that the image of gathering is carried out Filtering Processing then; Use the K means clustering algorithm that coloured image is cut apart through filtered image; At last the information of coloured image in the HSI space is carried out statistical study, respectively the characteristic of H and I component is extracted, as the input of sorter.Select for use SVMs to carry out image classification, adopt improved gravitation searching algorithm that nuclear parameter is carried out optimizing, the model of training classification based on the RBF kernel function.Described improved gravitation searching algorithm is based on the gravitation searching algorithm of policy feedback, in algorithm, introduces the entropy of population distribution and the diversity that the averaged particles range index is described the particle population, is used as measure function with this.Main improvement step comprise following some:

1) calculate particle i and with the Euclidean distance R of the immediate particle j of particle i _{I, j}

2) the Euclidean distance R of calculating particle i and whole space optimum position best _{I, best}

3) to R _{I, j}And R _{I, best}Ratio is set threshold value C, works as R _{I, j}/ R _{I, best}During>=C, still upgrade by former evolutionary equation,

Work as R _{I, j}/ R _{I, best}During＜C, the position of particle i will be adjusted by formula (1) again:

X _i(new)＝X _i(old)*D (1)

$D=\left\{\begin{array}{ccc}{R}_{i,j}*U(-\frac{1}{2},\frac{1}{2})& \mathrm{if}& R_{i,\mathrm{best}}\≥1\\ 1+\∂*U(-\frac{1}{2},\frac{1}{2})& & \mathrm{otherwise}\end{array}\right.$

Wherein Be to return one

Between equally distributed pseudo random number, parameter

Be constant, be taken as 10 ^-6

Judgement of the present invention and controlling index

(1) whether the time 1.5 exists with the interior defective that detects;

(2) exposure is multistage adjustable, scalable camera white balance;

(3) have storage, the query function of record, statistics, classification;

(4) have quick startup, function such as detect, stop, promptly stopping.

Claims (2)

1. solar energy silicon crystal chip surface color automatic testing method based on machine vision is characterized in that: its classification of defects method is based on the mode identification method that SVMs combines with improved gravitation search optimized Algorithm.

2. the improved gravitation search optimized Algorithm described in the claim 1 is based on the gravitation search optimized Algorithm of policy feedback.

Images