JP6411757B2 - Inspection capability determination apparatus and inspection capability determination method for surface defect inspector - Google Patents

Description

  TECHNICAL FIELD The present invention relates to a surface defect inspector inspection capability determination device and an inspection capability determination method, and more particularly, to determine an inspection capability of an inspector when inspecting a defect existing on the surface of an inspection object having specular reflection. The present invention relates to an apparatus and a method.

  Inspection for the presence or absence of surface defects is in the direction to be automated, but surface defects that occur on the surface of products with strong specular reflection are not being automated due to difficulties in image processing. Currently. Therefore, the inspector visually inspects for the presence of surface defects.

  However, the inspection ability of an inspector who visually inspects depends on personal ability based on years of experience, etc., but it has been difficult to make a determination while quantifying the inspection ability. This was thought to be caused by the absence of an index for quantitatively evaluating the ability of the inspector. In particular, in the inspection of an inspection object having specular reflection, visual inspection is more difficult than an inspection object having no specular reflection, and individual differences are conspicuous.

  Non-Patent Document 1 describes a psychometric curve, and Non-Patent Documents 2 and 3 describe a technique that applies signal detection theory, but inspect defects on surfaces with strong specular reflection. It was not about an evaluation that was suitable for doing.

JP 2007-333709 A

Yuki Kimiyama, Basics of psychological measurement and application in survey, Data Analysis Laboratory, (2009), p.228 Hidetoshi Nakayasu, rationalization in appearance inspection / defect inspection and evaluation of inspector ability, Information Organization (2012), p.26 Hidetoshi Nakayasu, Examination on skill evaluation of defect inspectors applying signal detection theory, Proceedings of Academic Lecture 61, (2012-05), pp.221-222

  Patent Document 1 is an apparatus for automating appearance inspection based on the skill evaluation methods disclosed in Non-Patent Documents 1 to 3. However, Non-Patent Documents 1 to 3 evaluate the inspection skill of the inspector in one aspect, and the technique of Patent Document 1 to which the theory is applied samples the inspection result by the inspector and inspects based on the sampling result. Although the standard is obtained, since only one aspect of the inspection skill of the inspector is evaluated, a comprehensive evaluation cannot be obtained. Moreover, the technique of Patent Document 1 does not evaluate the inspection skill of the inspector but sets the inspection standard with reference to the inspection skill of the inspector and automates it.

  The present invention has been made in view of the above-mentioned various points, and an object thereof is an apparatus and method capable of comprehensively and quantitatively evaluating the inspection skill of an inspector who inspects a surface defect by visual observation. Is to provide.

  Therefore, the present invention relating to the inspection capability determination device for the surface defect inspector is a device for determining the inspection capability of the inspector who visually inspects the surface defect of the product, and is provided on a plurality of products. A storage means for storing a result of inspecting the presence or absence of a surface defect of the product by a plurality of inspectors based on an inspection reference value based on a predetermined size of the surface defect, and the inspection result for each inspector A psychological measurement curve generating means for generating a psychological measurement curve indicating a probability of determining the presence or absence of the surface defect with respect to the size of the surface defect, and a judgment reference value and variation in the inspection of each inspector from the psychological measurement curve A means for deriving a value; a means for obtaining a deviation degree of the judgment reference value based on a difference between the inspection reference value and the judgment reference value; A means for generating a normal distribution relating to the probability density of the size of the defect to be inspected for both decisions, and determining the distance between the two distributions from the two normal distributions for both decisions, while distinguishing from decisions having no defects Means for deriving parameters by normalizing each numerical value for the degree of deviation and variation values of the determination reference value and the distance between the two distributions, while comparing the results with a plurality of inspectors, and the parameters on the radar chart And an evaluation unit that evaluates the inspection ability based on the area formed when disassembled.

  According to the above configuration, after storing the inspection results by a plurality of inspectors, for the inspection results by the individual inspectors, deriving the degree of deviation of the criterion value, the variation value, and the distance between the two normal distributions. Therefore, it is possible to determine how much a specific inspector has by comparing with the whole. Then, accuracy can be quantified by the degree of deviation of the criterion value, reliability can be quantified by the variation value, and discrimination force can be quantified by the distance between the distributions of the two normal distributions, and these parameters can be quantitatively evaluated. be able to. Further, since the area formed when the parameters are decomposed into radar charts can be obtained, the accuracy, reliability, and discrimination power can be comprehensively evaluated. Regarding the presence or absence of surface defects, when the size of the surface defects is larger than the inspection reference value, it is evaluated that the surface defects are present, and when the size of the surface defects is smaller than the inspection reference value, the surface defects are determined. It is evaluated that it does not have (no surface defects). That is, the determination that there is no surface defect is not limited to the case where there is no surface defect, but includes the case where the surface defect has a predetermined size or less, and it is determined whether the product is defective or not. .

  In the above invention, the inspection results are a plurality of results obtained by performing a similar inspection at different times, and the parameters normalized with respect to the degree of deviation and variation values of the determination reference value and the distance between the two distributions are a plurality of inspections. The result is a plurality of parameters that are derived individually, and for the plurality of parameters, the minimum value is subtracted from the maximum value, and the fluctuation rate obtained by dividing this by the minimum value is derived as a parameter for each fluctuation rate. The evaluation means can evaluate the inspection capability based on an area formed when each parameter including the variation rate is decomposed into a radar chart.

  According to the above configuration, the stability of the inspection result of the inspector can be obtained by obtaining the variation rate of the degree of deviation, variation value, and inter-distribution distance of the judgment reference value that changes for each inspection based on the inspection results of a plurality of times. Can also be quantitatively evaluated.

  In each of the above inventions, the product has specular reflection, the surface defects are scratches, dents and foreign matters, and the inspection for the presence or absence of defects is performed by checking the width of the scratches, the diameter of the dents and the foreign matters. It can depend on the size of each diameter.

  According to the said structure, evaluation of the inspection capability with respect to a different kind of surface defect is enabled about the inspection target object which has specular reflection. This means that even the same inspector can grasp that the evaluation of the inspection capability may differ depending on the type of surface defect.

  Furthermore, in each of the above inventions, a configuration can be provided that includes display means for displaying a product image and input means for inputting an inspection result for the product image displayed on the display means.

  According to the above configuration, it is possible to inspect the image data stored in advance without using an actual product, and to store the specific image and the inspection result in association with each other. It is possible to quickly and easily perform generation of a curve, derivation of a criterion value and a variation value from a psychological measurement curve, generation of a normal distribution, and derivation of a distance between distributions.

  On the other hand, the present invention according to the method for determining the inspection capability of a surface defect inspector is a method for determining the inspection capability of an inspector who visually inspects a surface defect of a product, and is provided on a plurality of products. A step of inspecting the presence or absence of surface defects of the product by a plurality of inspectors based on an inspection reference value based on a predetermined size of the surface defect, a step of inputting the result of the inspection into the storage means, From the inspection results stored in the storage means, for each inspector, generating a psychometric curve indicating the probability of determining the presence or absence of the surface defect relative to the size of the surface defect, A step of deriving a judgment reference value and a variation value in the inspection by the inspector, a step of obtaining a degree of bias of the judgment reference value based on a difference between the inspection reference value and the judgment reference value, Generating a normal distribution for the probability density of the size of the defect to be inspected for both decisions while distinguishing between the decision having defects and the decision having no defects, and two normal distributions for both decisions A step of obtaining a distance between both distributions from each other, and a step of deriving a parameter by normalizing each numerical value while comparing with a result of a plurality of inspectors regarding a degree of deviation and a variation value of the determination reference value and a distance between the both distributions, and And an evaluation step of evaluating the inspection capability based on the area formed when the parameter is decomposed into a radar chart.

  According to the above configuration, by inputting the inspection result by each inspector for each product, the deviation degree of the criterion value, the variation value, and the inter-distribution distance are sequentially derived, and these are normalized parameters. The inspection capability can be converted into a value that can be evaluated. Here, the evaluated numerical value can evaluate the superiority or inferiority of the inspection skill of the inspector by relative comparison with other inspectors (the entire inspector). Although the handling of superiority and inferiority of this inspection skill is artificial, when constructing an inspection line that inspects the same product by multiple inspectors, material to consider personnel allocation so that the inspection results are not biased for each line In addition, if the superiority or inferiority of the inspection skill varies depending on the type of surface defect, it is used to arrange an inspector to compensate for the specific surface defect for which the inspection skill of a specific inspector is inferior. be able to.

  Further, in the above invention, the step of inspecting for the presence or absence of surface defects is performed a plurality of times at different times, and the step of deriving the parameter is a criterion value for each of the inspection results performed a plurality of times. A parameter obtained by normalizing the degree of deviation and variation of the distribution and the distance between the two distributions, and subtracting the minimum value from the maximum value for the plurality of parameters and dividing the result by the minimum value. And a step of deriving the obtained variation rate as a parameter of each variation rate, and the evaluation step evaluates the inspection capability by an area formed when each parameter including the variation rate is decomposed into a radar chart. be able to.

  According to the above configuration, when the inspection result of each inspection by a single inspector can be converted into a numerical value that can be used to evaluate the degree of deviation of the judgment reference value, the variation value, and the distance between distributions, The fluctuation rate of each parameter can be added to the evaluation of the inspection skill. Therefore, an inspector with a small variation rate can be regarded as having a stable inspection result, and enables comprehensive evaluation of skills in each time zone.

  According to the present invention relating to the inspection capability determination apparatus for surface defect inspectors, it is possible to quantitatively evaluate the accuracy based on the deviation degree of the determination reference value, the reliability based on the variation value, the discrimination force based on the distance between distributions, and the like. Since the area formed when the parameter is decomposed into the radar chart can be obtained, the accuracy, reliability, and discrimination power can be comprehensively evaluated. Further, in the configuration provided with means for deriving the variation rate of each parameter, it is also possible to evaluate the inspection ability of the inspector that varies with time. Although the present invention is performed separately from the normal work, it is possible to grasp the inspection ability of the inspector by a short time inspection. Can be confirmed.

  On the other hand, according to the present invention relating to the inspection capability determination apparatus for the surface defect inspector, the level of the inspection technique of the inspector can be quantitatively and comprehensively evaluated by numerical values, so that the appropriateness as an inspector is determined. be able to. And by detecting the surface defects of products that have surface reflections that are difficult for the inspector to visually check, the inspector's inspection ability is evaluated, which contributes to the stability of the quality of the product being shipped and the placement of the inspector. Considering this, it will be possible to reduce personnel.

It is explanatory drawing which shows the outline of the inspection capability determination apparatus of a surface defect inspector. It is a graph which illustrates psychological measurement polarity. It is a graph which shows the relationship between the normalization random variable and the magnitude | size of a surface defect. It is a graph which shows the normal distribution of a signal and noise. It is a figure which illustrates a ternary radar chart. It is a graph which illustrates the grade of the change of the parameter which shows accuracy for every time. It is a figure which illustrates a 6-value radar chart. 6 is a graph showing the results of Experimental Example 1. 10 is a graph showing the results of Experimental Example 2.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

<First Embodiment of Inspection Capability Determination Device>
FIG. 1 is a diagram illustrating an outline of an embodiment according to an inspection capability determination apparatus. As shown in this figure, the device 1 of this embodiment is connected to a monitor 2 and an input means 3 which are external devices. The inspection capability determination device is constituted by a personal computer including a processing device, and the storage means 11 and data stored in the storage means 11 are processed by the means 12 to 18. The storage means 11 includes a memory in addition to the HHD. The storage unit 11 stores data appropriately input by the input unit 3 and stores processing results by the units 12 to 18. Each of the means 12 to 18 for processing data obtains data stored in the storage means 11 and processes the data according to a predetermined procedure. The processed data is stored temporarily or in a long term. 11 is stored.

  The inspection capability determination apparatus 1 includes means 12 to 18 and processes data as appropriate. These processes are calculated by a personal computer functioning as a processing apparatus. As each processing means, psychological measurement curve generation means 12, judgment reference value / variation value derivation means 13, judgment reference value bias degree calculation means 14, normal distribution generation means 15, inter-distribution distance calculation means 16, parameter derivation means 17 And the evaluation means 18 is comprised.

Therefore, the processing contents of the processing means 12 to 18 will be described together with the processing procedure.
The storage means 11 stores as various data when the inspection results by a plurality of inspectors are input from the input means 13, and more specifically, a plurality of product numbers and a surface provided for each product number. For each inspector, the type of defect, the inspection reference value based on the size of the surface defect determined in advance for each type of surface defect, and the number of the inspector and the result of inspection by the inspector are stored.

  As the types of surface defects, three types of scratches, dents (dents), and foreign matters (projections) can be exemplified. For these surface defects, a plurality of products having different sizes are prepared, and the results of inspection by the inspector are input. As for the size of the surface defect, in the case of an abrasion, the width dimension is changed every 0.005 mm in the range of 0.005 mm to 0.040 mmn, and the dent and foreign matter have a diameter of 0.20 mm. In the range of .about.0.50 mm, those changed every 0.05 mm can be used. Note that the inspection reference value can be set to 0.020 mm in the case of an abrasion and 0.30 mm in diameter for a dent and a foreign object, but this is appropriately changed according to the product to be inspected. It is.

Then, the plurality of samples are presented to the inspector at random multiple times to inspect for the presence of surface defects. Each sample is presented M times (for example, 20 times to maintain accuracy) at the time of inspection, and the probability P _j of the number m _j at which the j-th sample is determined to have surface defects in M times of presentation is P _j = m _j / M (Equation 1) can be calculated.

  Here, visual inspection of surface defects is entrusted to a psychological reference value that the individual inspector has as a sensation based on experience and skill. For this reason, the sense in the vicinity of the reference value changes continuously, and there is no possibility of discriminating with a probability of 100% using the reference value as a boundary. However, obviously large surface defects and clearly fine surface defects can clearly determine the presence or absence of surface defects, and thus the above-mentioned reference value is estimated to exist between the two. Therefore, the psychological measurement curve generation means 12 generates a psychological measurement curve.

The psychological measurement curve generation means 12 psychologically measures the probability (determination probability) that the inspector determines the presence or absence of the surface defect with respect to the size of the surface defect based on the inspection result stored for each inspector. It is organized as a curve. The psychometric curve, the magnitude of x _j of surface defects on the horizontal axis, in which the probability P _j is determined that there is a surface defect on the surface defects of the size and the vertical axis. FIG. 2 illustrates an inspector's psychological measurement curve. As is clear from this figure, the physical measurement curve is a curve with a low probability of determining that there is a surface defect when the surface defect is small, and with a high probability of determining that there is a surface defect when the surface defect is large. The psychological measurement curve generated in this way can also be displayed on the monitor 2.

The relationship between the surface defect size x _j and the probability P _j determined that there is a surface defect can be expressed by a normal distribution cumulative distribution function as shown in Equation 2 below. In Expression 2, μ _i is a judgment reference value of the individual (i-th inspector) of the inspector, and σ _i is a value indicating variation in the inspection result of the individual (i-th inspector). Then, by obtaining these μ _i and σ _i , the judgment reference value and the variation value of the individual inspector are derived.

Therefore, the determination reference value / variation value deriving means 13 derives the determination reference value μ _i and the variation value σ _i . The procedure of deriving μ _i and σ _i by the judgment reference value / variation value deriving means 13 is derived using the cumulative distribution function after normalizing the random variable x to the normalized random variable y. . The details are as follows. That is, the random variable x is normalized as shown in Equation 3 to obtain a normalized random variable y, which is substituted into Equation 2 to obtain Equation 4.

Equation 4 obtained as above becomes a cumulative distribution function of a standard normal distribution, and a normalized probability variable y _j can be calculated from the probability P _j . Further, assuming that 1 / σ _i = a and −μ _i / σ _i = b with respect to Equation 3, and between the normalized random variable y _j and the surface defect size x _j , the above Equation 5 Since a linear relationship such as (y = ax + b) is established, the parameters a and b can be obtained by the method of least squares using the relationship between the normalized random variable y _j and the surface defect size x _j as Equation 5. . Incidentally, as long required parameters a and b, sigma _i can calculate the sigma _i = 1 / a, further, mu _i can be calculated as μ _i = -bσ _i.

Here, with respect to the inspection result by the i-th inspector, with respect to the j-th sample, the size x _j of the surface defect is taken as the horizontal axis, and the normal obtained from the probability P _j that the sample is determined to have a surface defect. FIG. 3 shows an example of a graph in the case where the randomization probability variable y _j is the vertical axis. As shown in this figure, between the magnitude x _j and the random variable y _j of surface defects and certain variations exist. Since this variation is expected to affect the estimated values of the parameters a and b, the estimation may be performed again by the maximum likelihood estimation method in order to minimize the influence.

The re-estimation method in this case is as follows.
From the above equations 4 and 5, p _j = φ (ax _j + b) can be derived. From this, P _j is calculated from x _j and substituted into the following equation 6 indicating the likelihood function of the binomial distribution. , The likelihood value l (n _j ) for the j th sample is determined. When all samples are considered as n _j (j = 1, 2,..., N), the likelihood function is expressed by the following equation 7, and log is obtained by taking logarithms on both sides of equation 7. (1 (n ₁ , n ₂ ,... N _j ,... _{N N} )) = L (n ₁ , n ₂ ,... N _j ,... _{N N} ) I can do it.

Therefore, by solving Equation 8, P _j corresponding to n _j can be obtained with maximum likelihood for all samples. Based on the calculated P _j , similarly to the above, y _j is obtained from the above equation 4, and parameters a and b are calculated to obtain σ _i (variation) and μ _i (judgment reference value). be able to.

  The determination reference value derived as described above is the size of the surface defect when the determination probability indicates 50%, and serves as an index indicating the boundary when determining the presence or absence of the surface defect. In addition, the variation value is an indicator that the degree of variation is large in determining whether there is a surface defect when the value is large, and conversely, if the value is small, the variation value is an indicator that the degree of variation is small. In this sense, the degree of variation can be a material for determining whether the same inspection is performed on each sample (product) (reliability).

The judgment reference value μ _i derived as described above is calculated as the degree of bias μ _i ′ by the judgment reference value bias degree calculation means 14. This degree of deviation is the difference between the inspection standard value for surface defects (desired size for determining surface defects, predetermined size) x ₀ and the standard value derived from the results of actual inspection by the inspector. It is calculated, and is calculated by μ _i ′ = μ _i −x ₀ . If this degree of deviation is ± 0, it means that the presence or absence of surface defects was inspected according to the reference value to be inspected, but if this shifts to plus or minus, the inspector's criterion is the inspection reference value It means that it deviates from (is biased). In this sense, it can be an index indicating whether the inspector has made an accurate determination based on the inspection reference value (accuracy).

  The normal distribution generation means 15 classifies a plurality of products into a result determined to have surface defects and a result determined to have no surface defects, and generates a normal distribution for both results. This normal distribution is a normal distribution related to the probability density of the size of the defect to be inspected, and the horizontal axis represents the surface defect size, and the vertical axis represents the probability density.

In general, sensory stimuli perceived by humans (such as vision) are not purely physical stimuli and always have psychological effects (psychological noise). The ability to recognize a sensory stimulus that is a correct signal from this psychological noise can be an important factor in assessing the ability of an inspector. Therefore, the signal can be the size of the surface defect to be inspected, and the noise can be a psychological state such as the physical condition of the inspector or the inspection environment. Further, based on the signal detection theory, the probability density distribution of the signal and noise is represented by a normal distribution, and the distance between the two distributions is calculated to determine the discrimination ability of the surface defects. According to the signal detection theory, the probability density distribution of the signal and noise can be expressed by normal distributions N _S (σ = 1, μ _S ) and N _N (σ = 1, μ _N ) with σ = 1, respectively. it can. When the distance d = μ _S −μ _N between the two distributions is large, the influence of psychological noise is small, and when the distance d is small, the influence of psychological noise is large.

Therefore, first, based on the inspection result by the inspector, a case where it is determined that there is a surface defect and a case where it is determined that there is no surface defect are classified, and the probability that the determination is made is obtained. That is, when the j-th sample having a surface defect larger than the reference value (in the case of a signal) is presented M times, and it is correctly determined that there is a surface defect m times (m _{SN, j} ), the correct answer determination Can be expressed by the following equation (9). In addition, when the j-th sample (in the case of noise) having a small surface defect equal to or less than the reference value is presented M times, it is determined that the surface defect is correctly determined m times (m _{N, j} ). Can be expressed by the following equation (10).

Then, the probability density distribution of the signal and noise, respectively, as described above, sigma = 1 in the normal distribution _{N S (σ = 1, μ} S), N N (σ = 1, μ N) from to follow, i th If the standard deviation of the size of the surface defect is obtained for the inspection results of the inspectors, the cumulative distribution functions of both of the random variables v and u (v = x / σ _x , u = x / σ _x ) are expressed by the following equation 11: , 12 can be expressed.

The normal distribution is shown in FIG. In the figure, the upper normal distribution illustrates the signal distribution N _S (σ = 1, μ _S ), and the lower normal distribution illustrates the noise distribution N _N (σ = 1, μ _N ). It is a thing. As shown in this figure, the signal distribution is divided into a region A and a region C with the determination reference value x ₀ / σ _x as a boundary. Region A is a “hit” region that has been correctly determined as having surface defects for surface defects having a size greater than or equal to the reference value, and region C has been incorrectly determined as having surface defects that are smaller than the reference value. This is the “miss” area. Similarly, the noise distribution is also divided into a region B and a region D with the determination reference value x ₀ / σ _x as a boundary. The region D is a “correct-rejection” region that is correctly determined as having no surface defect for a small surface defect equal to or less than the reference value, and the region B is an erroneously determined “false alarm” region. The normal distribution generated in this way can also be displayed on the monitor 2.

  As is apparent from this figure, when the distance between the two normal distributions is small, the regions B and C are large, so that the psychological influence is acting greatly, and when the distance between the two distributions is large. Since the regions B and C are small, the psychological influence is small. Therefore, this inter-distribution distance can be used as an index of the inspector's discrimination ability.

Since the distance d between the two distributions is obtained by d = μ _S −μ _N , the distance between the distributions is finally obtained by calculating these μ _S and μ _N. This calculation is processed by the inter-distribution distance calculation means 16, and the procedure is as follows. That is, the cumulative distribution function described above is as shown in Equations 11 and 12, and as in the case of the psychological measurement curve, a normalized random variable y _j is obtained from the probability P _j and finally μ _S and μ _N Is calculated.

Based on the above calculation results, parameters for technical evaluation are derived by the parameter deriving means 17. The parameters derived here relate to the judgment reference value μ _i , the variation value σ _i, and the inter-distribution distance d _i . For determination reference value mu _i is to the deviation degree of the parameter, the difference μ _{i '=} μ _i -x ₀ of the reference value x ₀ as described above is used.

  Further, for each of these parameters, an average value and a standard deviation are calculated from determination results of all the inspectors (assuming M persons), and are defined as follows based on the average value and the standard deviation. The parameter (accuracy) related to the criterion value is shown in Expression 13, the parameter (reliability) related to the variation value is shown in Expression 14, and the parameter (discrimination ability) related to the distance between the distribution of the signal and noise is expressed in Expression 15. .

Note that the parameters μ _{D, j} , σ _{D, j} , d _{D, j} calculated as described above follow a normal distribution with an average value of 4 and a standard deviation of 1, and the values of each parameter are in the range of 1-7. Collect with a probability of 99.7% (± 3σ). Also, the smaller the values of μ _i ′ and σ _i , the higher the accuracy and reliability. Therefore, μ _{D, j} and σ _{D, j} are negatively signed and have a large value so that the inspection capability increases. It was adjusted. Each parameter thus normalized is finally evaluated by the evaluation means 18.

The evaluation means 18 decomposes the parameter into a radar chart, and evaluates the inspection capability based on the size of the area (triangular area) formed on the ternary radar chart. Examples (two types) of radar charts at this time are shown in FIG. As shown in the figure, the scale of each axis of the radar chart is 1 and the reference value is 4. For comparison with the reference value 4, the parameters use values of (μ _D -4), (σ _D -4), and (d _D -4). Further, the difference of 1 scale corresponds to the difference of 1σ (68.3% probability), and the difference of 2 scales corresponds to the difference of 2σ (95.4% probability).

  In the case of the figure, the area of the radar chart (regular triangle) of the reference value (when all is 4) is 20.8, whereas the area of the inspector C is 18.5. The overall evaluation of the inspection capability is low, and on the other hand, the area of the inspector J is 36.6, so that the evaluation that the inspection capability is high is obtained.

  As a final evaluation display method, the numerical value of the area may be used, but the area ratio with respect to the reference value 20.8 may be expressed as + or −. Further, a method may be used in which an area of a predetermined range in the vicinity of the reference value is evaluated as “possible”, more than that is evaluated as “good” or “excellent”, and less than that is expressed as “impossible”. Further, instead of these or together with them, a radar chart can be displayed on the monitor 2.

  Since this embodiment is configured as described above, for a plurality of inspectors, by inspecting surface defects under the same conditions, and inputting the inspection results, by relative comparison with a plurality of inspectors, The inspection ability can be digitized and judged quantitatively and comprehensively. As a plurality of inspectors at this time, all inspectors in the same company may be used, or inspectors with different years of experience may be picked up and an average value by these inspectors may be referred to. Furthermore, for the ability determination, an image for inspection is prepared in advance, and inspection results from a plurality of inspectors are stored in advance, so that a specific inspector inspects under the same conditions using the inspection image. You may comprise so that a test | inspection capability can be evaluated as a relative comparison with the inspector memorize | stored beforehand by inputting a result.

  The inspection capability determination method is in accordance with the process described above, and can be summarized as follows. With respect to the surface defects provided in the plurality of products, the result of the inspection of the presence or absence of the surface defects of the products by the plurality of inspectors based on the inspection reference value based on the predetermined size of the surface defects is stored from the input unit 3 11 is input.

  When the inspection result is input, each process is started. From the inspection result stored in the storage unit by the psychological measurement curve generation unit 12, for each inspector, the surface relative to the size of the surface defect. A psychometric curve is generated that indicates the probability of determining the presence or absence of a defect. Subsequently, the judgment reference value / variation value deriving means 13 derives the judgment reference value and the variation value in the inspection of each inspector based on the psychological measurement curve, and the judgment reference value bias degree calculating means 14 further calculates. The deviation degree of the judgment reference value is calculated from the difference between the inspection reference value and the judgment reference value.

  Further, for each inspector, the normal distribution generation means 15 distinguishes between the determination of having a defect and the determination having no defect from the inspection result, and the normality related to the probability density of the size of the defect to be inspected for both determinations. A distribution is generated, and the distance between both distributions is calculated from two normal distributions by the distribution distance calculation means 16.

  Then, the parameter deriving means normalizes each numerical value while comparing with the results of a plurality of inspectors regarding the bias degree and variation value of the determination reference value calculated by the above and the distance between the two distributions, and the parameter for determination Is derived, and the evaluation means 18 comprehensively evaluates the inspection capability based on the area formed when the parameters are decomposed into radar charts.

<Second Embodiment of Inspection Capability Determination Device>
Next, a second embodiment will be described. In the present embodiment, parameters used in the first embodiment described above are added. That is, in the first embodiment, a parameter (μ _D ) indicating accuracy, a parameter (σ _D ) indicating reliability, and a parameter (d _D ) indicating discrimination ability are used. For these parameters, the fluctuation rate is further calculated and added to the parameters.

The fluctuation rate refers to the variation for each time zone from the inspection results performed under the same conditions in different time zones. Specifically, the same inspection is performed several times (for example, 5 times at 10 am, 11 am, 2 pm, 3 pm, 4 pm) at different times, and the accuracy of the results of each inspection is shown. A parameter (μ _D ), a parameter (σ _D ) indicating reliability, and a parameter (d _D ) indicating discrimination capability are obtained. FIG. 6 illustrates the state of change in the parameter (μ _D ) indicating the accuracy at that time. “Time 1” in the figure is 10:00 am, and shows the one performed five times according to the above example. In the case of the inspector illustrated in this figure, the evaluation of the inspection result at 3 pm (Time 4) is the lowest, and the inspection result at 4 pm (Time 5) is the highest. Therefore, if such a change is small, it is evaluated that stable inspection is possible regardless of the time zone (has stability), but if the above change is large, an evaluation that the stability is lacking is evaluated. Is possible.

Therefore, in order to quantify the degree of change, the variation rate deriving means 19 (described in FIG. 1) derives (variable maximum value−minimum value) / (minimum parameter value) as the variation rate. is there. Since the fluctuation rate derived here is the ratio of the change amount to the minimum value of the parameter, the average value and standard deviation are also calculated from the judgment results of all the inspectors. The parameters are normalized by the deviation. The variation rate is calculated not only for the parameter (μ _D ) indicating accuracy but also for the parameter (σ _D ) indicating reliability and the parameter (d _D ) indicating discrimination ability.

As with the other parameters, the fluctuation rate parameter follows a normal distribution with an average value of 4 and a standard deviation of 1. The value of each parameter ranges from 1 to 7 with a probability of 99.7% (± 3σ). get together. Three types of parameters relating to the rate of variation are derived: those relating to accuracy (μ _t ), those relating to reliability (σ _t ), and those relating to discrimination ability (d _t ).

Each parameter relating to these fluctuation rates is decomposed into a radar chart by the evaluation means 18 together with the already derived three parameters, and comprehensive evaluation is performed based on the area at that time. The scale of each axis of the radar chart is also set to 1 and the reference value is set to 4. For each parameter related to the fluctuation rate for comparison with the reference value 4, the values of (μ _t -4), (σ _t -4), and (d _t -4) are used, and the difference of one scale is 1σ. The difference (68.3% probability) and the difference of the two scales correspond to the difference of 2σ (95.4% probability).

  FIG. 7 shows an example of a state in which each parameter (6 in total) is decomposed into a radar chart. As shown in this figure, by calculating the area of the hexagonal radar chart, it is possible to comprehensively judge the inspection capability including the variation rate. Since the area of the radar chart (regular hexagon) in the case of the reference value 4 is 46.8, the area of the radar chart of the inspector W in the figure is 66.4, and the area of the radar chart of the inspector X is 66.4. Was 72.4, and thus both exceeded the reference value. When comparing the inspectors W and X, it can be determined that the inspector X is superior to the inspector W in inspection capability.

<Experimental example 1>
An experiment for calculating the area of a ternary radar chart was performed on 12 inspectors with different years of experience according to the first embodiment. A sample having three types of surface defects, scratches, dents and foreign matters, was prepared for a product having specular reflection as a whole. For the scratches, the width of the scratches was changed by 0.005 mm in a range of 0.005 mm to 0.040 mm (8 in total). The thickness was changed by 0.05 mm within a range of 50 mm (total of 7 pieces each). These were randomly inspected by an inspector, and each sample was subjected to inspection 20 times. All 12 test results were processed and individual inspectors were evaluated. In the process, an average value is calculated from all the inspectors, parameters for evaluation of each inspector are derived, and an area on the ternary radar chart is calculated. The result is shown in FIG. In the graph shown in the figure, the vertical axis represents the area of the radar chart, and the horizontal axis represents the inspector. The horizontal axis shows the results of inspectors who have years of experience from left to right.

  As can be seen from this figure, the inspector with a long experience as a whole (the right side of the figure) has a larger radar chart area than the inspector with a short experience, although there are differences among individuals. Thereby, it is estimated that the inspection capability can be evaluated quantitatively and comprehensively.

<Experimental example 2>
Similarly, an experiment for calculating the area of a six-value radar chart was performed for four inspectors. The experimental method was the same as in Experimental Example 1, and there were 12 inspectors, but the area of the radar chart was calculated and evaluated only for four people. The result at this time is shown in FIG. Also in this figure, the vertical axis represents the area of the radar chart, and the horizontal axis represents the inspector. The horizontal axis shows the results of inspectors who have years of experience from left to right.

  Also in this figure, although there are individual differences, the evaluation of the inspector with a long experience is high, and the evaluation of the inspector with a short experience is low. Thereby, it is estimated that the inspection capability can be evaluated quantitatively and comprehensively.

DESCRIPTION OF SYMBOLS 1 Test capability determination apparatus 2 Monitor 3 Input means 11 Storage means 12 Psychometric measurement curve generation means 13 Judgment reference value / variation value derivation means 14 Judgment reference value bias degree calculation means 15 Normal distribution generation means 16 Inter-distribution distance calculation means 17 Parameters Deriving means 18 Evaluation means 19 Fluctuation rate deriving means

Claims (6)

A device for judging the inspection ability of an inspector who visually inspects a surface defect of a product,
With respect to surface defects provided in a plurality of products, a plurality of inspectors memorize the inspection results obtained by inspecting the presence or absence of surface defects of the products for each product based on a predetermined inspection standard value based on the size of the surface defects. Storage means;
For each inspector, psychological measurement curve creating means for generating a psychological measurement curve indicating the probability of determining the presence or absence of the surface defect relative to the size of the surface defect from the inspection result;
Means for deriving a judgment reference value and a variation value in the examination of each inspector from the psychological measurement curve;
Means for determining a degree of bias of the determination reference value based on a difference between the inspection reference value and the determination reference value;
Means for generating a normal distribution of the probability density of the size of the defect to be inspected for each decision, distinguishing between a decision having defects and a decision having no defects for each inspector;
Means for determining the distance between the two distributions from the two normal distributions for both decisions;
Means for deriving parameters by normalizing each numerical value while comparing with the results of a plurality of inspectors for the degree of deviation and the variation value of the judgment reference value and the distance between both distributions;
An inspection capability determination apparatus for a surface defect inspector, comprising: evaluation means for evaluating an inspection capability based on an area formed when the parameter is decomposed into a radar chart. The test results are a plurality of results obtained by performing a similar test at different times,
The parameters normalized for the degree of deviation and variation value of the criterion value and the distance between both distributions are a plurality of parameters derived individually for a plurality of inspection results,
A means for deriving a variation rate obtained by subtracting a minimum value from a maximum value and dividing this by the minimum value as a parameter for each of the plurality of parameters,
2. The surface defect inspector inspection capability determination device according to claim 1, wherein the evaluation unit evaluates the inspection capability based on an area formed when each parameter including the variation rate is decomposed into a radar chart.   The product has specular reflection, and the surface defects are scratches, dents and foreign matters, and the inspection for the presence or absence of defects depends on the respective sizes of the width of the scratches, the diameter of the dents and the diameter of the foreign matters. The surface defect inspector's inspection capability determination device according to claim 1 or 2.   4. The inspection capability determination of the surface defect inspector according to claim 1, further comprising display means for displaying a product image, and input means for inputting an inspection result for the product image displayed on the display means. apparatus. A method for determining the inspection ability of an inspector who visually inspects a surface defect of a product,
A step of inspecting the presence or absence of a surface defect of the product based on an inspection reference value based on a predetermined size of the surface defect for the surface defect provided in the plurality of products,
Inputting the result of the inspection for each product into a storage means;
Generating a psychological measurement curve indicating the probability of determining the presence or absence of the surface defect with respect to the size of the surface defect for each inspector from the inspection result stored in the storage means;
Deriving a judgment reference value and a variation value in the examination of each inspector from the psychological measurement curve,
Determining the degree of bias of the judgment reference value by the difference between the inspection reference value and the judgment reference value;
For the inspector, generating a normal distribution related to the probability density of the size of the defect to be inspected for both decisions while distinguishing between the determination having the defect and the determination having no defect from the inspection result;
Obtaining a distance between the two distributions from the two normal distributions for both determinations;
The step of deriving parameters by normalizing each numerical value while comparing with the results of a plurality of inspectors for the degree of deviation and the variation value of the judgment reference value and the distance between both distributions;
And an evaluation step of evaluating the inspection capability according to an area formed when the parameter is decomposed into a radar chart. The step of inspecting for the presence or absence of surface defects is performed a plurality of times at different times,
The step of deriving the parameter is a step of deriving a parameter obtained by normalizing a degree of deviation and a variation value of each determination reference value and the distance between the two distributions for the inspection result performed a plurality of times.
Further, for a plurality of the parameters, a step of subtracting a minimum value from a maximum value, and dividing a variation rate obtained by dividing the minimum value by the minimum value, is derived as a parameter of each variation rate,
6. The method according to claim 5, wherein the evaluation step evaluates the inspection capability based on an area formed when each parameter including the variation rate is decomposed into a radar chart.

Images