CN104772880A - Injection molding mechanical arm mold anomaly detection method based on LMDO (Local Multilayered Difference Operator) - Google Patents

Abstract

The invention provides an injection molding mechanical arm mold anomaly detection method based on an LMDO (Local Multilayered Difference Operator). The anomaly detection method comprises the following steps: (1) acquiring a standard template image when an injection molding machine opens a mold in place, and pre-processing to obtain a later difference background image; (2) waiting for working state information of the injection molding machine; upon detection of a situation that the injection molding machine is operated until the mold is opened in place, continuously acquiring the image of a mold cavity by a camera, extracting an average image of the plurality of images, and pre-processing the average image to do preparation for subsequent image processing, thereby obtaining a later difference foreground image; and (3) carrying out an anomaly detection algorithm based on the LMDO on the difference foreground image and the difference background image to obtain an abnormal region without a light illumination interference part. The injection molding mechanical arm mold anomaly detection method based on the LMDO, provided by the invention, has the characteristics of good instantaneity, strong robustness on illumination variation and the like; and whether the mold has an abnormal state or not can be monitored through mold opening information of the injection molding machine.

Description

A kind of injection molding mechanical arm mould method for detecting abnormality based on LMDO

Technical field

The present invention relates to injection machine industry in industrial automation, is a kind of intelligent surveillance device be furnished with to mould in the injection machine production process of manipulator.

Background technology

In recent years along with the continuous expansion of plastic products application, the demand of global manufacturing to injection machine presents lasting unprecedented soaring trend.Modern injection machine is all furnished with manipulator, and manipulator is the partial function that can imitate human arm, automatically can take out injecting products, make its according to pre-provisioning request to product carry out stacking, arrange and put.Manipulator is the automated production equipment of Injection Molding Machine Design exploitation specially, it can alleviate the heavy manual labor of workman, improve working conditions and ensure safety in production, enhance productivity, in the market competitiveness etc. strengthening enterprise, play extremely important effect.Nowadays manipulator of injection machine uses general, the safeguard protection in manipulator automatic clamp reclaiming process and module protection important all the more.In order to carry out monitoring in real time to realize the automatic protection to mould to abnormal conditions such as residual, the slide block dislocation existed with mould in the injection machine production process of manipulator, the demoulding are bad, mould protector is by the injection machine industry introduced in a large number with manipulator.

This monitoring system based on image processing techniques greatly improves security in injection machine mould production process and operating efficiency, reduces the labour intensity of repair a die cost and staff to a great extent.Mould protector function is in the market all more single; the image processing algorithm of the software that mold protecting device uses is relatively simple; intelligence degree is relatively low, and the system that result in often needs complicated manual calibration and experience to ensure the correctness of its abnormality detection result.But the working environment generally with the injection machine place of manipulator can be very complicated; often illumination variation is largely had; and adopt the mould protecting of traditional algorithm normally can not to work in this changeable environment, easily there is the problems such as abnormal area false alarm frequently.At this time user needs the relevant parameter of frequent adjustment System, or directly relearns, and causes complex operation and the low problem of operating efficiency.

Summary of the invention

In order to the real-time overcoming existing injection molding mechanical arm mould supervision method is not enough, to deficiencies such as illumination variation robustness are poor, the invention provides and a kind ofly have that real-time is good, the method for detecting abnormality based on Local Multilevel time difference operator (LMDO) to features such as illumination variation strong robustnesses, whether it can have abnormality by injection moulding machine mould open image information supervision molding.

The technical solution adopted for the present invention to solve the technical problems is:

Based on an injection molding mechanical arm mould method for detecting abnormality of LMDO, described method for detecting abnormality comprises following process:

1) gather standard form image when injection moulding machine mould open puts in place, and carry out pretreatment, as substracting background image afterwards;

2) injection machine work state information is waited for, when detect injection machine move to die sinking put in place time, by video camera to the continuous capture of mould cavity surface, extract the average image of a few width image, and pretreatment is done to the average image,, for successive image process is prepared, as difference foreground image afterwards;

3) Outlier Detection Algorithm based on LMDO is performed to difference foreground image and substracting background image, obtain the abnormal area getting rid of illumination interference sections; Process is as follows:

3.1) Local Multilevel time difference operator

Be R at radius, the image local area that neighbours' point is P, each neighbours put the contrast level t corresponding to pixel value _pgained is calculated by following formula:

g _s＝g _p-g _c(1)

In formula, g _pfor the gray value of P neighbours' point, g _cfor the gray value of regional area central pixel point, maxC, minC represent respectively neighbours point with central point between contrast maximum and minimum of a value, T represents contrast layering quantity; For each contrast level i, observe around central point and have the contrast value of how many neighbours' points to fall into this layer, to obtain the LMDO of every level:

${\mathrm{LMDO}}_{P.R}^i=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p,C_p=\left\{\begin{array}{c}1,t_p=i\\ 0,t_p\≠i\end{array}\right.,i=1...T---\left(3\right)$

By each level level is linked togather and constitutes LMDO _p.R:

${\mathrm{LMDO}}_{P.R}={\mathrm{LMDO}}_{P.R}^1/{\mathrm{LMDO}}_{P.R}^2/.../{\mathrm{LMDO}}_{P.R}^T$

3.2) LMDO value is utilized to carry out difference to background and foreground image

According to T level value judges the phase knowledge and magnanimity between pixel, any the i-th level of note background image 8bit binary value be Blmdo (j), j=0...7, i-th layer of this point of foreground image 8bit binary value be Flmdo (j), both difference, obtain the similarity of the i-th level:

$S_i=\underset{j=0}{\overset{P-1}{\mathrm{\Σ}}}(\mathrm{Blmdo}\left(j\right)\⊗\mathrm{Flmdo}\left(j\right))---\left(4\right)$

Wherein, for with or;

Ask the similarity between pixel must obtain the weight of each level, then by the similarity S of each level _iwith respective multiplied by weight, obtain overall similarity:

$S=\underset{i=1}{\overset{T}{\mathrm{\Σ}}}(a_i\×S_i)---\left(5\right)$

Wherein, a _irepresent the weight of i-th level, show through great number tested data, work as a ₁=a ₂=... a _ttime, compare the similarity best results of two width images; If when similarity S is less than a setting threshold value T, then namely this pixel is identified as prospect, then value is gray level 0, otherwise is background, and its value is gray level 255, thus realizes the segmentation of abnormal area;

4) continuous print open and close computing is carried out to abnormal area image;

5) by region-growing method, abnormal area is marked and area measurement;

6) obtain abnormal area profile finally by morphological Edge extraction and report to the police, injection machine stops pressing mold entering interlock protection measure.

Beneficial effect of the present invention is mainly manifested in: utilize the method for detecting abnormality based on Local Multilevel time difference operator, solve the problem of the system flase drop that illumination variation causes efficiently.Meanwhile, LMDO has less computation complexity and good texture features, and major part is comparison operation, realizes simple, improves image abnormity detection efficiency.These are all the key factors of the influential system that prior art is not considered, this Outlier Detection Algorithm greatly strengthen real-time and the robustness of system.

Accompanying drawing explanation

Fig. 1 is the structured flowchart of the die sinking method for detecting abnormality based on LMDO.

Fig. 2 is the acquisition process figure of foreground image LMDO value.

Fig. 3 is the acquisition process figure of background image LMDO value.

Fig. 4 is the learning process figure of background image.

Fig. 5 is the die sinking method for detecting abnormality flow chart based on LMDO.

Fig. 6 is that background LMDO figure is with prospect LMDO figure difference and image after Threshold segmentation.

Detailed description of the invention

Below in conjunction with accompanying drawing, the invention will be further described.

With reference to Fig. 1 ~ Fig. 6, a kind of injection molding mechanical arm mould method for detecting abnormality based on LMDO, described method for detecting abnormality comprises following process:

1) gather standard form image when injection moulding machine mould open puts in place, and carry out pretreatment, as substracting background image afterwards;

2) injection machine work state information is waited for, when detect injection machine move to die sinking put in place time, by video camera to the continuous capture of mould cavity surface, extract the average image of a few width image, and pretreatment is done to the average image,, for successive image process is prepared, as difference foreground image afterwards;

3) Outlier Detection Algorithm based on LMDO is performed to difference foreground image and substracting background image, obtain the abnormal area getting rid of illumination interference sections; Process is as follows:

3.1) Local Multilevel time difference operator

LMDO is the more a certain pixel of one and its neighborhood territory pixel value, and then the contrast between generation pixel, then contrast value is mapped to certain level to get on, obtain the LMDO value of many levels, using LMDO value as characteristics of image, being used for the image difference be identified under the environment of illumination variation has extraordinary effect.

Be R at radius, the image local area that neighbours' point is P, each neighbours put the contrast level t corresponding to pixel value _pgained is calculated by following formula:

g _s＝g _p-g _c(1)

In formula, g _pfor the gray value of P neighbours' point, g _cfor the gray value of regional area central pixel point, maxC, minC represent respectively neighbours point with central point between contrast maximum and minimum of a value, T represents contrast layering quantity.For each contrast level i, can observe around central point and have the contrast value of how many neighbours' points to fall into this layer, to obtain the LMDO of every level:

${\mathrm{LMDO}}_{P.R}^i=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p,C_p=\left\{\begin{array}{c}1,t_p=i\\ 0,t_p\≠i\end{array}\right.,i=1...T---\left(3\right)$

By each level level is linked togather and constitutes LMDO _p.R:

${\mathrm{LMDO}}_{P.R}={\mathrm{LMDO}}_{P.R}^1/{\mathrm{LMDO}}_{P.R}^2/.../{\mathrm{LMDO}}_{P.R}^T$

3.2) LMDO value is utilized to carry out difference to background and foreground image

According to T level value judges the phase knowledge and magnanimity between pixel, any the i-th level of note background image 8bit binary value be Blmdo (j) (j=0...7), i-th layer of this point of foreground image 8bit binary value be Flmdo (j), both difference, obtain the similarity of the i-th level:

$S_i=\underset{j=0}{\overset{P-1}{\mathrm{\Σ}}}(\mathrm{Blmdo}\left(j\right)\⊗\mathrm{Flmdo}\left(j\right))---\left(4\right)$

Wherein, for with or.

Because the pixel value of each level occupies certain contrast in regional area, ask the similarity between pixel must obtain the weight of each level, then by the similarity S of each level _iwith respective multiplied by weight, obtain overall similarity:

$S=\underset{i=1}{\overset{T}{\mathrm{\Σ}}}(a_i\×S_i)---\left(5\right)$

Wherein, a _irepresent the weight of i-th level, show through great number tested data, work as a ₁=a ₂=... a _ttime, compare the similarity best results of two width images.If when similarity S is less than a setting threshold value T, then namely this pixel is identified as prospect, then value is gray level 0, otherwise is background, and its value is gray level 255, thus realizes the segmentation of abnormal area;

4) carry out continuous print open and close computing to abnormal area image to affect with stress release treatment;

5) by region-growing method, abnormal area is marked and area measurement;

6) obtain abnormal area profile finally by morphological Edge extraction and report to the police, injection machine stops pressing mold entering interlock protection measure; If there is no abnormal area, then continue the injection moulding machine mould open status information waiting for next cycle.

Whole system device is primarily of analog video camera, Video Decoder, core dsp processor, touch-screen, liquid crystal display, the formations such as keyboard.Whole exception handling procedure: first, when injection moulding machine mould open state, learn command is performed by keyboard button, make camera collection standard form image, and carry out pretreatment, with information irrelevant in removal of images, as the background image afterwards for difference, and information is stored in the memory module of DSP core control panel (TMS320DM6437), as shown in Figure 4; Secondly, as shown in Figure 5, by the time when die sinking puts in place, system is in the monitoring state, by analog video camera to the continuous capture of mould cavity surface, by DSP core control panel to camera acquisition to a few width image zooming-out the average images, and pretreatment is done to the average image, for successive image process is prepared, as the foreground image for difference; Again background image and foreground image are carried out difference, and adopt Threshold segmentation to realize binaryzation the image after difference; And continuous print open and close computing is carried out on image affect with stress release treatment; By region-growing method, abnormal area is marked and area measurement; Check whether there is exception finally by air cavity detection, if there is exception, in touch-screen display alarm information, injection machine stops pressing mold entering interlock protection measure; Otherwise injection machine work state information when continuing to wait for that die sinking puts in place.

The described method for detecting abnormality based on LMDO, concrete steps:

As shown in Figure 2, be 1 at radius, neighbours' point is the image local area of 8, pixel value g _cpoint centered by the point of=130, each neighbours put pixel value and are respectively 134,27,60,4,127,221,82,187 from the upper left corner by being rotated counterclockwise corresponding pixel value.Get contrast layering quantity T=4, ask for the LMDO value of background image, first obtain the contrast level t corresponding to each neighbours point _pcalculated by following formula:

$g_{s_2}=g_{p_2}-g_c=27-130=-103$

$g_{s_3}=g_{p_3}-g_c=60-130=-70$

$g_{s_4}=g_{p_4}-g_c=4-130=-126$

$g_{s_5}=g_{p_5}-g_c=127-130=-3$

$g_{s_6}=g_{p_6}-g_c=221-130=91$

$g_{s_7}=g_{p_7}-g_c=80-130=-48$

$g_{s_8}=g_{p_8}-g_c=187-130=57$

For each contrast level i (i=1 ... 4), can observe around central point has the contrast value of how many neighbours' points to fall into this layer, to obtain the LMDO value of every level:

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^1=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×0+2^6\×1+2^5\×0+2^4\×1+2^3\×0+\\ 2^2\×0+2^1\×0+2^0\×0=80\end{array}$

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^2=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×0+2^6\×0+2^5\×1+2^4\×0+2^3\×0+\\ 2^2\×0+2^1\×1+2^0\×0=34\end{array}$

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^3=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×1+2^6\×0+2^5\×0+2^4\×0+2^3\×1+\\ 2^2\×0+2^1\×0+2^0\×0=136\end{array}$

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^4=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×0+2^6\×0+2^5\×0+2^4\×0+2^3\×0+\\ 2^2\×1+2^1\×0+2^0\×1=5\end{array}$

By each level level is linked togather and constitutes LMDO _8.1:

${\mathrm{LMDO}}_{8.1}={\mathrm{LMDO}}_{8.1}^1/{\mathrm{LMDO}}_{8.1}^2/{\mathrm{LMDO}}_{8.1}^3/{\mathrm{LMDO}}_{8.4}^4=80/34/136/5$

As shown in Figure 3, be 1 at radius, neighbours' point is the image local area of 8, pixel value g _cpoint centered by the point of=120, each neighbours put pixel value and are respectively 134,47,58,6,140,231,75,160 from the upper left corner by being rotated counterclockwise corresponding pixel value.Get contrast layering quantity T=4 equally, ask for the LMDO value of foreground image, first obtain the contrast level t corresponding to each neighbours point _pcalculated by following formula:

$g_{s_1}=g_{p_1}-g_c=134-120=14$

$g_{s_2}=g_{p_2}-g_c=47-120=-73$

$g_{s_3}=g_{p_3}-g_c=58-120=-62$

$g_{s_4}=g_{p_4}-g_c=6-120=-114$

$g_{s_5}=g_{p_5}-g_c=140-120=20$

$g_{s_6}=g_{p_6}-g_c=231-120=111$

$g_{s_7}=g_{p_7}-g_c=75-120=-45$

$g_{s_8}=g_{p_8}-g_c=160-120=40$

For each contrast level i (i=1 ... 4), can observe around central point has the contrast value of how many neighbours' points to fall into this layer, to obtain the LMDO value of every level:

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^1=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×0+2^6\×1+2^5\×1+2^4\×1+2^3\×0+\\ 2^2\×0+2^1\×0+2^0\×0=112\end{array}$

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^2=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×0+2^6\×0+2^5\×0+2^4\×0+2^3\×0+\\ 2^2\×0+2^1\×1+2^0\×0=2\end{array}$

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^3=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×1+2^6\×0+2^5\×0+2^4\×0+2^3\×1+\\ 2^2\×0+2^1\×0+2^0\×1=137\end{array}$

$\begin{array}{c}{\mathrm{LMDO}}_{8.1}^4=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p=2^7\×0+2^6\×0+2^5\×0+2^4\×0+2^3\×0+\\ 2^2\×1+2^1\×0+2^0\×0=4\end{array}$

By each level level is linked togather and constitutes LMDO _8.1:

${\mathrm{LMDO}}_{8.1}={\mathrm{LMDO}}_{8.1}^1/{\mathrm{LMDO}}_{8.1}^2/{\mathrm{LMDO}}_{8.1}^3/{\mathrm{LMDO}}_{8.1}^4=112/2/137/4$

Below according to 4 levels value judges the phase knowledge and magnanimity between pixel, the i-th level of note background pixel value 130 8bit binary value be Blmdo (j) (j=0...7), i-th layer of foreground pixel value 120 8bit binary value be Flmdo (j), both difference, obtain the similarity of the i-th level:

$S_1=\underset{j=0}{\overset{P-1}{\mathrm{\Σ}}}(\mathrm{Blmdo}\left(j\right)\⊗\mathrm{Flmdo}\left(j\right))=0\⊗0+1\⊗1+0\⊗1+1\⊗1+0\⊗0+0\⊗0+0\⊗0+0\⊗0=7$

$S_2=\underset{j=0}{\overset{P-1}{\mathrm{\Σ}}}(\mathrm{Blmdo}\left(j\right)\⊗\mathrm{Flmdo}\left(j\right))=0\⊗0+0\⊗0+1\⊗0+0\⊗0+0\⊗0+0\⊗0+1\⊗1+0\⊗0=7$

Because the pixel value of each level occupies certain contrast in regional area, ask the similarity between pixel must obtain the weight of each level, then by the similarity S of each level _iwith respective multiplied by weight, obtain overall similarity:

$S=\underset{i=1}{\overset{T}{\mathrm{\Σ}}}(a_i\×S_i)=7\×1+7\×1+7\×1+7\×1=28,$ Get a ₁=a ₂=... a _t=1

If when similarity S is less than a setting threshold value T, then namely this pixel is identified as prospect, then value is gray level 0, otherwise is background, and its value is gray level 255, thus realizes the segmentation of abnormal area;

Under mould die opening state, by the template image of camera collection, and carry out pretreatment, with the image of noise in removal of images and irrelevant information.When mould die sinking puts in place, under system is in the monitoring state, by video camera to the continuous capture of mould cavity surface, extract the average image of a few width image, and pretreated image is done to the average image.Under when mould die sinking puts in place, many moulds do not come off, can the efficiency of influential system.

As shown in Figure 6, for background image and foreground image are through the result figure of LMDO abnormality processing gained, in figure, black part is exactly the result of abnormal area segmentation.

This LMDO method for detecting abnormality effectively can remove the impact of illumination variation, and computing is simple, accelerates the speed of abnormality detection, effectively improves operating efficiency.

Claims (1)

1., based on an injection molding mechanical arm mould method for detecting abnormality of LMDO, it is characterized in that: described method for detecting abnormality comprises following process:

1) gather standard form image when injection moulding machine mould open puts in place, and carry out pretreatment, as substracting background image afterwards;

2) injection machine work state information is waited for, when detect injection machine move to die sinking put in place time, by video camera to the continuous capture of mould cavity surface, extract the average image of a few width image, and pretreatment is done to the average image,, for successive image process is prepared, as difference foreground image afterwards;

3) Outlier Detection Algorithm based on LMDO is performed to difference foreground image and substracting background image, obtain the abnormal area getting rid of illumination interference sections; Process is as follows:

3.1) Local Multilevel time difference operator

Be R at radius, the image local area that neighbours' point is P, each neighbours put the contrast level t corresponding to pixel value _pgained is calculated by following formula:

g _s＝g _p-g _c(1)

In formula, g _pfor the gray value of P neighbours' point, g _cfor the gray value of regional area central pixel point, maxC, minC represent respectively neighbours point with central point between contrast maximum and minimum of a value, T represents contrast layering quantity; For each contrast level i, observe around central point and have the contrast value of how many neighbours' points to fall into this layer, to obtain the LMDO of every level:

${\mathrm{LMOD}}_{P.R}^i=\underset{p=0}{\overset{P-1}{\mathrm{\Σ}}}{2}^p\·C_p,C_p=\left\{\begin{array}{c}1,t_p=i\\ 0,t_p\≠i\end{array}\right.,i=1...T---\left(3\right)$

By each level level is linked togather and constitutes LMDO _p.R:

${\mathrm{LMDO}}_{P.R}={\mathrm{LMDO}}_{P.R}^1/{\mathrm{LMDO}}_{P.R}^2/.../{\mathrm{LMDO}}_{P.R}^T$

3.2) LMDO value is utilized to carry out difference to background and foreground image

According to T level value judges the phase knowledge and magnanimity between pixel, any the i-th level of note background image 8bit binary value be Blmdo (j), j=0...7, i-th layer of this point of foreground image 8bit binary value be Flmdo (j), both difference, obtain the similarity of the i-th level:

$S_i=\underset{j=0}{\overset{P-1}{\mathrm{\Σ}}}(\mathrm{Blmdo}\left(j\right)\⊗\mathrm{Flmdo}\left(j\right))---\left(4\right)$

Wherein, for with or;

Ask the similarity between pixel must obtain the weight of each level, then by the similarity S of each level _iwith respective multiplied by weight, obtain overall similarity:

$S=\underset{i=1}{\overset{T}{\mathrm{\Σ}}}(a_i\×S_i)---\left(5\right)$

Wherein, a _irepresent the weight of i-th level, show through great number tested data, work as a ₁=a ₂=... a _ttime, compare the similarity best results of two width images; If when similarity S is less than a setting threshold value T, then namely this pixel is identified as prospect, then value is gray level 0, otherwise is background, and its value is gray level 255, thus realizes the segmentation of abnormal area;

4) continuous print open and close computing is carried out to abnormal area image;

5) by region-growing method, abnormal area is marked and area measurement;

6) obtain abnormal area profile finally by morphological Edge extraction and report to the police, injection machine stops pressing mold entering interlock protection measure.