JPH087104A - Pass / fail judgment threshold determination method - Google Patents

Abstract

(57) [Summary] [Purpose] In the inspection that judges the quality of the inspection object from the image data of the inspection object obtained from the visual sensor, the discrimination threshold with few misjudgments even for distributions that do not follow the normal distribution Enable the decision of. [Configuration] When two quality determination thresholds are not required, the calculation is performed by a frequency distribution calculation step 85 of good / defective products 1
Two frequency distributions, two if necessary, good / defective product frequency distribution calculation step 86, good / defective product distribution representative value calculation step 87, good / defective product distribution variance calculation step 88, feature amount range division A good / defective product frequency distribution calculation process 80 for obtaining two frequency distributions calculated in process 89, a good / defective product distribution representative value calculation process 81, and a good / defective product cumulative frequency distribution calculation process 82, A cumulative sum frequency distribution load sum calculation step 83 and a cumulative frequency distribution load sum minimum value detection step 84 are provided, and the feature value corresponding to the minimum value detected by the cumulative frequency distribution load sum minimum value detection step 84 is good or bad. Use as a judgment threshold.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a quality determination threshold value determining method. More specifically, the present invention relates to a threshold value deciding method which is useful when the quality of the inspection object is judged based on the characteristic amount of the inspection object obtained by using a visual sensor such as a television camera.

[0002]

2. Description of the Related Art Automation in production processes and the like has been promoted in various conventional industrial fields, and in these automation processes, an inspection process for judging quality of products, intermediate products, etc. plays an important role. Regarding such an inspection process, various ideas have been made so far, but among them, a method of making a pass / fail determination using a visual sensor such as a television camera is prevalent.

A conventional inspection method will be described with reference to the drawings. First, in FIG. 1, a mechanical component (10) to be inspected is combined with another mechanical component in a step subsequent to the inspection step. Grease (1
1) is applied and used for lubrication of sliding parts between machine parts. If the amount of the grease (11) is not appropriate or if the applied position is displaced, good lubrication cannot be obtained, and problems such as making abnormal noise during machine operation occur. Normally, the grease (11) is supplied from the tip of the nozzle of the dispenser (12). In this coating process, problems such as changes in the coating amount and displacement of the coating position occur, but the causes are changes in the viscosity of grease, nozzle clogging, bubbles entering the nozzle, and vibration of the nozzle tip position. There is a case where it is misaligned. Therefore inevitably applying position, it is necessary to perform the coating amount of the test. Figure 2
Shows an apparatus for the inspection, in which an inspection object (20), a dispenser (21), a television camera (22), and an image processing device (23) are arranged and input from the television camera (22). The applied state is determined for the image thus formed, and if the application is defective, the result is output to an external device through I / O, and the external device performs processing such as re-application. FIG. 3 shows the image processing device (2
3) is specifically shown and the whole is shown as a device (30). This is the configuration of the device (30). The signal from the camera is converted into a digital signal through the A / D converter (31) and stored in the image memory (32). The image digital signal is pixel-divided so that it becomes 512 pixels in the horizontal direction and 480 pixels in the vertical direction with respect to the scanning line of the camera, and its value becomes larger as it gets brighter. Key data. The coordinates in the horizontal and vertical directions to the scanning line of the camera are x and y, respectively, and the values of x and y increase as the position moves to the lower right from the coordinates (0,0) in the upper left corner of the image. To do. This digital image is a microcomputer (33)
And the application state is determined. Judgment result is I
It is output from / O (34). FIG. 4 shows an outline of processing by the microcomputer. The binarization (40) sets the value of a pixel whose pixel density is smaller (thus darker) than a predetermined threshold value to 1 and sets the others to 0.
The image obtained by this procedure is called a binary pixel, and thus the pixel value of the grease application portion is 1 and the others are 0.
The feature amount calculation (41) calculates the feature amount of the grease application portion separated by binarization. The feature amount is determined by the application area (number of pixels) and the center of gravity of the application area, and the application area calculation formula is as follows.

[0004]

[Equation 1]

However, a in the formula is a coating area, and b (x,
y) is the pixel value of the binary image at the coordinates (x, y). The formula for calculating the center of gravity of the application area is as follows.

[0006]

[Equation 2]

In the equation, (gx, gy) is the barycentric coordinate.
The quality determination means (42) is realized by the procedure shown in FIG. That is, if the feature amount is within the predetermined upper and lower limit threshold values for each feature amount, the product is a good product, and if there is even one feature amount outside the upper and lower limit range, it is a defective product. The inspection standard determining means (43) shown in FIG. 4 may be a method of determining by using a statistical method from the trial and error or the distribution of the characteristic amount data. In the trial-and-error method, a person changes the upper and lower threshold values based on the discrimination result to make a discrimination that seems to be correct. In the statistical method, the difference between the average value of each distribution and the threshold value is normalized by the standard deviation of each distribution based on the assumption that the distribution of feature quantities of good products and defective products is normal distribution. Many methods of obtaining the threshold value are adopted. Let m0 and m1 be the averages of the distributions of non-defective products and defective products of the inspection object, σ0 be the standard deviation of the distribution of non-defective products, σ1 be the standard deviation of the distribution of defective products, and the threshold value be t. It holds. This equation gives t.

[0008]

(Equation 3)

[0009]

However, the inspection standard determining means typified by such a conventional method by trial and error, or a method of determining the distribution of the characteristic amount data by using a statistical method is called the accuracy. The situation is not always satisfactory. In other words, the trial-and-error method cannot avoid the time and effort of human intervention and the effects of individual differences. Furthermore, the upper and lower limit threshold values may be affected by the temporal change of the feature amount. That is, as shown in FIG. 6, when the frequency distribution of the good product (60) and the defective product (61) with respect to the feature amount is shown, the feature amount value is x.
In most cases, the product is a defective product, but if the latest data in terms of time is a non-defective product, the feature value may result in a non-defective product.

On the other hand, in the statistical method, the distribution of the feature values often does not follow the normal distribution. For example, as illustrated in FIG. 7, when the good product frequency distribution (70) and the bad product distribution (71) do not follow the normal distribution, it is not possible to perform a good / bad determination of the inspection object with high accuracy. In FIG. 7, m1 and m2 are average values of distributions of good products and defective products, σ1 and σ2 are standard deviation magnitudes, and t is the above-mentioned equation 3.
It shows the discrimination threshold that satisfies the condition of.

The present invention has been made in view of the above circumstances, solves the drawbacks of the conventional quality determination method, is not affected by individual differences, and the feature amount does not follow the normal distribution. An object of the present invention is to provide a quality determination threshold value determination method that enables accurate quality determination of an inspection target.

[0012]

In order to solve the above-mentioned problems, the present invention uses image data obtained by digitizing the video signal obtained by a visual sensor picking up an image of an object to be inspected and outputting the video signal. In a test for calculating a feature amount of an object to be inspected and determining pass / fail by comparing the feature amount with a pass / fail determination threshold value, a frequency distribution with respect to the value of the feature amount is calculated for each of a good product and a defective product. Process,
The second step of calculating the representative value of the characteristic amount for each distribution of the good product and the defective product from the frequency distribution, and the distribution in which the representative value becomes large according to the magnitude relation of the representative value The third step of calculating the cumulative frequency in the direction in which the characteristic amount becomes larger or smaller in the direction in which the representative value becomes smaller and the distribution value in which the representative value becomes smaller, and the load factor for each cumulative frequency of the non-defective product and the defective product described above. And a fifth step of detecting the minimum value of the sum, and a value of the feature amount that minimizes the sum obtained in the fifth step is good or bad. There is provided a method for determining a pass / fail judgment threshold value (claim 1), wherein the judgment threshold value is used.

Further, according to the present invention, the pass / fail judgment threshold value is 2
In the inspection described above, which is necessary, the step of calculating the frequency distribution for the value of the characteristic amount for each of the good product and the defective product, and the calculation of the representative value of the characteristic amount for each distribution of the good product and the defective product from the frequency distribution. And the step of calculating the variance of the distribution of each good product and the defective product from the frequency distribution, and separating into a range where the feature amount is large and a range where the feature amount is small with the representative value of the distribution having the small variance as a boundary. The pass / fail judgment threshold value determining method (claim 2) is characterized by comprising a first step constituted by the step of obtaining the two frequency distributions.

Further, in the present invention, there is provided a fourth step of calculating a sum of cumulative frequencies after multiplying by a load coefficient by which a coefficient by which the cumulative frequency of defective products is multiplied is larger than a coefficient by which the cumulative frequency of non-defective products is multiplied. There is also provided the above-mentioned pass / fail judgment threshold value determining method (claim 3).

[0015]

According to the invention of claim 1, a pass / fail judgment threshold value which is not affected by trial and error or individual differences in inspection standards, and which has less misjudgment even for a distribution that does not follow a normal distribution is determined. Is possible. Further, in the invention of claim 2, the pass / fail judgment threshold value can be determined even when two threshold values are required. Further, in the invention of claim 3, it is possible to determine a pass / fail judgment threshold value at which it is difficult to mistake a defective product for a non-defective product.

Hereinafter, the method of the present invention will be described in more detail with reference to examples.

[0017]

EXAMPLES Example 1 (Method for Determining Pass / Fail Discrimination Threshold when Pass / Fail Discrimination Threshold is 1) The inspection target, the inspection apparatus, and the image processing apparatus are the same as those in the conventional example, and FIG. , As shown in FIG. The outline of the processing by the microcomputer is similar to that of FIG. 4, but the inspection standard setting means is different from the conventional example.
Shows the flow chart. That is, the good / defective product frequency distribution calculation step (80), the good / defective product distribution representative value calculation step (81), the good / defective product cumulative frequency distribution calculation step (82), and the cumulative frequency distribution load sum calculation Process (8
3), which comprises a minimum value detection step (84) of the cumulative frequency distribution load sum, and in the frequency distribution calculation step (80) of a non-defective product and a defective product, the process branches depending on whether two quality determination thresholds are necessary. In other words, if two good / bad determination thresholds are not required, one frequency distribution is calculated by the frequency distribution calculation step (85) of good / defective products, but if two thresholds are required, the frequency of good / defective products is calculated. Distribution calculation process (86), representative value calculation process of good / defective product distribution (87), variance calculation process of good / defective product distribution (8)
8), two frequency distributions are calculated by the feature amount range dividing step (89).

The calculated feature amount can be, for example, the area and barycenter of the application region. Of these feature amounts, the procedure for determining the determination threshold value as the inspection reference for the area will be described below as an example. First, a plurality of objects are subjected to the same processing from image input to coating area calculation as in the conventional example, and area values are stored in a memory. Further, the pass / fail judgment result is similarly stored in the memory in association with each area value. However, the quality judgment result is the result of a judgment made by a person looking at the object. Next, processing is performed using the area value in the memory and the quality determination result. For example, if the frequency distribution of non-defective products and defective products of the inspection object is the distribution shown in FIG. 9 when the vertical axis is taken as the area, one threshold value is required, and the flow of FIG. Then, conditional branching that does not require two threshold values is performed.

That is, first, in the step (85) of calculating the frequency distribution of good / defective products, the frequency distribution (90) of good products and the frequency distribution (91) of defective products are calculated as shown in FIG. If the determination threshold value is t, the frequency at which a good product is erroneously determined to be a defective product or a defective product is determined to be a good product is indicated by the shaded area in FIG. Next, in the representative value calculation step (81) of the distribution of good and defective products, the representative value of each distribution is calculated. Here, the average value is used as the representative value. In FIG. 9, M0 indicates the average area of good products, and M1 indicates the average area of defective products.

Further, in the cumulative frequency distribution calculating step (82), the cumulative frequency of each distribution is calculated. In this step, the magnitude relationship of the representative values of the distributions calculated in the representative value calculation step (81) of the good / defective product distribution is M0 <M1, and in this case, the area of the good product distribution is small. For the direction and distribution of defective products, the cumulative frequency is taken in the direction of larger area.
This cumulative frequency distribution is, for example, as shown in FIG. 10. In this figure, the cumulative frequency distribution of good products is shown as (100) and the cumulative frequency distribution of defective products is shown as (101). The hatched portion shown in FIG. 9 is represented by the sum of distribution heights at t in FIG. 10, that is, m0 + m1.

In the load sum calculation step (83) of the cumulative frequency distribution, the sum of the cumulative frequencies of each distribution multiplied by the load coefficient is calculated. When the weighting factor is set to 1 for both non-defective products and defective products, the sum of the cumulative frequency loads becomes the frequency of erroneous determination. The distribution when the sum of the cumulative frequencies of the respective distributions obtained in the cumulative frequency distribution calculation step (82) when the weighting factors are both set to 1 is as illustrated in FIG. 11, for example. Similarly uses the threshold value t shown in FIGS. 9 and 10. M shown in FIG.
0 + m1 becomes n in FIG. Since n is the frequency of misjudgment when the threshold is t, the threshold with the smallest n is the threshold with the smallest misjudgment. That is, as illustrated in FIG. 11, it is possible to perform the determination with the smallest erroneous determination when t ′ is the threshold value so that the cumulative frequency sum is minimized. In other words, in the minimum cumulative frequency distribution sum detection step (84), the optimal discriminating threshold value can be determined by determining the value of the feature amount that minimizes the total cumulative frequency. Embodiment 2 (Method of Determining Pass / Fail Discrimination Threshold of the Present Invention in Case of Two Pass / Fail Discrimination Thresholds) Upper and lower limits are necessary for pass / fail determination of feature quantity values depending on the object of inspection and the meaning of the feature quantity, and thus the quality is Consider a case where the frequency distribution of area values of defective products is as shown in FIG. 12, for example.

In FIG. 12, the frequency distribution of good products is (120) and the frequency distribution of defective products is (121) (122).
Is shown. In this case, two thresholds are required.
The procedure for setting the inspection standard is as shown in FIG. 8, and a conditional branch that requires two threshold values is performed. That is, the frequency distribution calculation process (86) for good / defective products and the representative value calculation process (87) for distribution of good / defective products are respectively the frequency distribution calculation process (85) for good / defective products and the good / defective products. The same processing as the distribution representative value calculation (81) is performed. Further, the variance / non-defective product distribution calculation step (88) is a process of calculating the variance from each frequency distribution of the good product and the defective product. The variance is calculated by the following equation.

[0023]

[Equation 4]

In the equation, d is the variance, n is the number of feature quantity data,
f (i) is the value of the i-th feature value, and i is an integer value from 1 to n. m is the average value of the feature amount, which is calculated by the following equation.

[0025]

(Equation 5)

The characteristic amount range dividing step (89) is a step of dividing the characteristic amount range into two according to the size relation of the branches. The division of the feature amount range may be performed with the representative value with the smaller variance as the boundary, which means that the distribution with the smaller variance is considered to be distributed between the two threshold values. Also, there is a discrimination threshold in each of the large and small range of the feature amount. The procedure for dividing the feature amount range will be described with reference to FIG. 12, for example. As illustrated in FIG. 12, when the average value of the frequency distribution of non-defective products and the average of the frequency distribution of defective products are m0 and m1, the variance d0 of the frequency distribution of non-defective products is the variance d1 of the frequency distribution of defective products. It is clear that it is smaller than. In this case, the feature amount range is divided at the average m0 of the frequency distribution of non-defective products. The frequency distribution in this divided feature amount range is as shown in, for example, FIG. 13, and in this figure, (a) shows m0.
The boundary is the lower characteristic amount range, and (b) is the frequency distribution in the upper characteristic amount range. In FIG. 13, the frequency distributions (130) (131) of defective products and the frequency distributions (132) (133) of non-defective products are shown. The two threshold values can be obtained by applying the respective steps shown in the first embodiment to the frequency distribution in each of the ranges thus divided and determining the quality determination threshold value. Third Embodiment In the actual production of products, the meaning is different between the case where a good product is erroneously determined as a defective product and the case where a defective product is erroneously determined as a good product. That is, if a defective product is erroneously determined as a non-defective product, a product that is originally a defective product will be produced and shipped as a non-defective product, so this erroneous determination must be avoided. On the other hand, when a good product is erroneously determined to be a defective product, there is no such risk and it is allowed to some extent. In this Example 3,
For example, a determination threshold value is set so that it is difficult to erroneously determine a defective product as a good product as described above. In this embodiment, as illustrated in FIG. 8, the procedure similar to that of the first embodiment is performed up to the cumulative frequency distribution calculation step (82). The present invention is characterized by multiplying the cumulative frequency of defective products by a load coefficient larger than the cumulative frequency of non-defective products in the load sum calculation step (83) of the cumulative frequency distribution. That is,
In the cumulative frequency distribution shown in FIG. 10, the load cumulative frequency obtained by multiplying the cumulative frequency of good products by 1 and the cumulative frequency of defective products by 2 is shown in FIG. In this figure, 140 is a non-defective product and 141 is a defective product distribution. The subsequent steps are the same as in the first embodiment, and the judgment threshold value is determined by taking the sum of the cumulative load frequencies and finding the minimum value. The sum of the cumulative load frequencies can be shown, for example, in FIG. 15, and the determined judgment threshold value is shown by tw. In contrast to the discrimination threshold t ′ that minimizes m0 + m1 shown in the first embodiment, the discrimination threshold tw that minimizes m0−m1 × 2 reduces the number m1 of the number of mistaken defective products to be good. . In this way, by multiplying the frequency distribution of defective products by a weighting factor larger than that of good products, it is possible to determine a judgment threshold value that makes it difficult to erroneously judge defective products as good products.

In this example, the threshold value is determined according to the first embodiment, but it is not limited to this. It can also be determined according to the second embodiment.

[0028]

As described in detail above, according to the present invention, it is possible to accurately determine whether the inspection target is good or bad without being influenced by individual differences and even if the feature amount does not follow the normal distribution. . Further, in the present invention, it is also possible to determine a judgment threshold value that is unlikely to be erroneously judged as a non-defective product by multiplying the frequency distribution of the defective product by a load coefficient larger than that of the non-defective product. It is possible to significantly reduce the erroneous determination that the product is a good product.

[Brief description of drawings]

FIG. 1 is an explanatory diagram of an inspection object in a conventional method and an embodiment of the present invention.

FIG. 2 is an explanatory diagram of a conventional method and an inspection apparatus according to an embodiment of the present invention.

FIG. 3 is an explanatory diagram of a configuration of the image processing apparatus of FIG.

FIG. 4 is a schematic explanatory diagram of processing by a microcomputer according to a conventional method and an embodiment of the present invention.

FIG. 5 is an explanatory diagram of a quality determination unit in a conventional method.

FIG. 6 is a first explanatory diagram showing a problem in the conventional method.

FIG. 7 is a second explanatory diagram showing a problem in the conventional method.

FIG. 8 is an explanatory diagram of a quality determination threshold value determining method according to the embodiment of the present invention.

FIG. 9 is an explanatory diagram of frequency distributions of non-defective products and defective products according to the first embodiment of the present invention.

FIG. 10 is an explanatory diagram of cumulative frequency distribution of non-defective products and defective products according to the first embodiment of the present invention.

FIG. 11 is an explanatory diagram of a cumulative frequency load sum according to the first embodiment of the present invention.

FIG. 12 is an explanatory diagram of frequency distributions of non-defective products and defective products according to the second embodiment of the present invention.

FIG. 13A is an explanatory diagram of a frequency distribution in a divided lower feature amount range according to the second embodiment of the present invention. (B) It is explanatory drawing of the frequency distribution in the divided high-order feature-value range in Example 2 of this invention.

FIG. 14 is an explanatory diagram of a load cumulative frequency according to the third embodiment of the present invention.

FIG. 15 is an explanatory diagram showing a sum of load cumulative frequencies according to the third embodiment of the present invention.

[Explanation of symbols]

 10 mechanical parts 11 grease 12 dispenser 20 inspection object 21 dispenser 22 television camera 23 image processing device 30 image processing device 31 A / D converter 32 image memory 33 microcomputer 34 I / O 40 binarization 41 feature quantity calculation 42 pass / fail judgment Means 43 Inspection standard determination procedure 60 Frequency distribution of non-defective product 61 Frequency distribution of defective product 70 Frequency distribution of non-defective product 71 Frequency distribution of non-defective product 80 Calculation process of frequency distribution of non-defective / defective product 81 Calculation process of representative value of non-defective product 82 Cumulative frequency distribution calculation process for good / defective products 83 Cumulative frequency distribution load sum calculation process 84 Cumulative frequency distribution load sum minimum value detection process 85 Good / defective product frequency distribution calculation process 86 Good / defective product frequency distribution calculation process 87 Representative value calculation process of good / defective product distribution 88 Distributed calculation process of good / defective product distribution 89 Feature amount range division 90 Frequency distribution of non-defective products 91 Frequency distribution of defective products 100 Cumulative frequency distribution of non-defective products 101 Cumulative frequency distribution of non-defective products 120 Frequencies distribution of non-defective products 122 Frequencies distribution of non-defective products 130 Frequencies distribution of non-defective products 131 Non-defective products Frequency distribution 132 Good frequency distribution 133 Bad product frequency distribution 140 Good product load cumulative frequency distribution 141 Bad product load cumulative frequency distribution

Claims (6)

[Claims] 1. A visual sensor picks up an image of an object to be inspected, outputs a video signal, calculates a characteristic amount of the object to be inspected from image data obtained by digitizing the image signal, and judges whether the characteristic amount is good or bad. In the inspection to judge pass / fail by comparing with the threshold value, the first step of calculating the frequency distribution with respect to the value of the characteristic amount for each of the good product and the defective product, and the characteristic of each distribution of the good product and the defective product from the frequency distribution For the second step of calculating the representative value of the quantity and the distribution in which the representative value becomes large according to the magnitude relation of the representative value, the direction in which the feature quantity becomes larger than small, and the distribution in which the representative value becomes small A third step of calculating the cumulative frequency in the direction in which the feature amount becomes smaller than a large value, and a fourth step of multiplying the cumulative frequencies of the good product and the defective product by a load coefficient and then calculating the sum thereof. And the minimum of the sum A pass / fail determination threshold value determining method, comprising: a fifth step of detecting a value, wherein the value of the feature amount having the minimum sum obtained in the fifth step is used as a pass / fail determination threshold value. 2. The inspection according to claim 1, wherein two pass / fail judgment thresholds are required, and a step of calculating a frequency distribution with respect to the value of the feature amount for each of the good product and the defective product, and the good product based on the frequency distribution. A step of calculating a representative value of the characteristic amount for each distribution of defective products, a step of calculating the variance of each distribution of the good product and the defective product from the frequency distribution, and a boundary of the representative value of the distribution in which the variance is small 2. The pass / fail judgment threshold value determination method according to claim 1, further comprising a first step including a step of obtaining two frequency distributions separated into a range where the feature amount is large and a range where the feature amount is small. 3. A fourth step of calculating a sum of cumulative frequencies after multiplying by a load coefficient by which a coefficient by which a cumulative frequency of defective products is multiplied is larger than a coefficient by which a cumulative frequency of non-defective products is multiplied. 3. The method for determining a pass / fail threshold value according to claim 1 or 2. 4. The pass / fail judgment threshold value according to claim 1, further comprising a second step in which the representative value of each distribution of the non-defective product and the defective product is an average value of the characteristic amount of each distribution. How to decide. 5. The method according to claim 1, further comprising a second step in which the representative value of each distribution of the non-defective product and the defective product is set to a value of the characteristic amount that maximizes the frequency of each distribution.
A method for determining a threshold for determining whether the quality is good or bad. 6. The method according to claim 1, further comprising a second step in which a representative value of each distribution of the non-defective product and the defective product is set as a value of a feature amount that is the center of the frequency of each distribution.
A method for determining a threshold for determining whether the quality is good or bad.