JPH0235342A - Visual inspection method - Google Patents

Abstract

PURPOSE:To implement a high processing speed by obtaining the average values of the gradation levels of picture elements constituting respective windows in comparison between the picture elements of a standard product and the picture elements of a body under inspection, and comparing the average values. CONSTITUTION:The entire image of an engine as a standard product is sensed in multiple gradations. A window A including a part under inspection (a) is set. The multiple gradation data which are outputted from picture elements G1, G2...Gn constituting the window A are processed to obtain multiple values. The data are summed up. The average value of the gradation levels is obtained from the added data. Said value is stored as the reference value of the window. At the same time, an allowable error value is set and stored. Then, the object to be sensed is changed into a body under inspection. Engines which are sequentially supplied by an appropriate conveying means are sensed, and the average value of the gradation levels of the picture elements constituting the inspecting window is computed. The average value is compared with the reference value. When the difference is within the allowable error range, it is judged that the external appearance of the part to be checked of the body under inspection is normal.

Description

[Detailed Description of the Invention] [Industrial Field of Application] The present invention relates to a visual inspection system for various types of equipment that can quickly detect foreign or missing parts, as well as detect misalignment, etc. Regarding the method.

[Conventional technology]

Conventionally, visual inspections such as inspections for missing parts, foreign parts, and positional deviations for various types of equipment have been performed visually by workers.

However, such work is monotonous and requires patience, resulting in severe fatigue and poor work efficiency, as well as frequent inspection errors. In order to solve these problems, attempts have been made in recent years to automate visual inspection using image processing technology. For example, in such a visual inspection method, image information of a normal product such as a standard product is digitized and stored in memory in advance, and then the subject is imaged to obtain image information, and both image information are combined. It is known to detect abnormalities in the appearance of a subject through comparison. Methods for comparing image information include binarizing the captured image, extracting feature points in the image and comparing them, or binarizing information obtained from corresponding pixels in the overall images of both. There is a method of comparing point by point.

[Problem to be solved by the invention]

However, with the method of extracting and comparing feature points, it is necessary for the operator to specify the feature points to be compared in advance. Visual inspection cannot be fully automated.

Furthermore, since the inspection site is specified in advance, there is a problem in that if an abnormality is found in an unexpected site, it cannot be detected. on the other hand,
The method of comparing all the pixel information constituting the entire image one by one has the problem that the processing time is long and high-speed processing cannot be performed. The most important thing is that whether comparing feature points or all pixel data, the pixel data to be compared is treated as binarized, so depending on how intermediate values are processed, The test results were subject to errors.

The present invention has been made in view of the current situation, and allows for complete automation of testing without any manual intervention, high-speed processing, excellent versatility, and the ability to prevent errors from entering test results. It also relates to an appearance inspection method, which is extremely rare.

[Means for Solving the Problem] In order to achieve the above-mentioned problem, the present inventor created an image (
The gradation data of all pixels constituting the inspection image (hereinafter referred to as the inspection image) is captured as multi-tone data, these multi-tone data are aggregated, the average value of the gradation data is calculated, and the average value is used to determine the characteristics of the inspection image. Regarded as reflected unique data,
We came up with a method to detect abnormalities in the inspection target by comparing this unique data with the IPI average of standard products.

Then, for comparison, calculate the gradation average of the standard product in advance and set the tolerance, and then use appropriate means to image the subjects sequentially supplied in front of the imaging device to obtain unique data for each subject. .

Then, comparing the specific data with the tone average of the standard product,
If the difference between the two was within a preset tolerance, the test subject was judged to be a good product.

Furthermore, if the imaging means is used continuously for a long period of time, the imaged region may become misaligned due to the influence of heat. In order to prevent such a situation, a fixed reference point is set in the vicinity of the subject, and when the area to be examined is imaged, the reference point is also imaged at the same time to correct the deviation of the area to be examined. is preferred.

[For production]

In such a visual inspection method, first, a standard product is imaged in multiple gradations, and a window containing a region to be inspected is set on the captured image. Next, the luminance information of each pixel constituting the window is subjected to multi-value processing, and then aggregated, and the average value of the gradation level within the window is determined, and the average value is set as a reference value. At the same time, a tolerance for allowing deviation from the average value is also set in advance.

Next, a subject is placed in the same position as the standard product, and the subject is imaged in multiple tones like the standard product to obtain a captured image. Then, on the captured image, an inspection window is set that allows imaging of the same area as the window set when imaging the standard product. At this time, it is necessary that the test site of the subject captured within the test window and the test site of the standard product captured within the reference window set in the standard product image completely match.

Next, the luminance information of all pixels within the inspection window is multi-valued and aggregated, and the average value of the gradation levels within the window is calculated to obtain unique data of the inspection window.

Finally, the unique data of the subject is compared with the gradation average value of the standard product, that is, the standard value, and if the difference between the two is within a preset tolerance, the appearance of the subject's test area is considered normal. This concludes the following.

Then, for the subjects supplied sequentially, the unique average value of each subject is determined through the same procedure as above, and each unique average value is compared with the standard average value to determine whether the test site of each subject is normal. , to determine abnormality. If the reference value is calculated and set first, there is no need to change it unless the type of inspection object changes, and the initially set value can be stored in an appropriate means and used continuously.

The method of the present invention compares the test site of the standard product and the test site of the subject by comparing the gradation levels of each subdivided part of each site point by point, as in the conventional method, and then Instead of determining the similarity of the appearance and shape of the entire inspection area by determining the similarity of the gradations and finally integrating each judgment result, we first calculate the average value of the gradation level of the entire inspection area and calculate the gradation level. Since the similarity of the appearance shapes of the test parts of the subject and the standard product is determined by comparing the average values with each other, a complicated calculation process is not required, and the speed of the determination process can be increased.

Moreover, since the gradation level of each pixel is multi-toned, accurate judgment can be made even when the inspection area has shading or the surface shape within the imaging area corresponding to one pixel is complex. .

Additionally, by setting a reference point outside the subject, it is possible to monitor the positional relationship of the examination area, and any deviation in the imaging area can be corrected, so accurate examinations can always be performed even if the imaging means is affected by heat. .

By the way, in the present invention, the appearance of the standard product and the test object is compared by comparing the average values of the gradation levels within the window, but with this method, there is a concern that the following situation may occur. In other words, if there are two parts within the test window that differ in external shape from the standard product, and the deviations from the average value of the gradation data of these two parts are opposite in sign and negative, and the sum of the two parts is When the deviations from the average value cancel each other out, there is a concern that an abnormality in appearance may not be detected. However, the present invention avoids this concern by treating the utterance data as multi-vocabulary data. In other words, when gradation data is treated as conventional monochrome data, deviations cancel each other out and there is a risk of overlooking abnormalities in appearance.However, in the method of the present invention, gradation data is multi-tone data, so even if deviations cancel each other out, abnormalities in appearance may be overlooked. Even if they do overlap, it is almost impossible for the deviation to be completely zero.

As a result of conducting various experiments with different inspection targets, the inventor found that the above phenomenon had almost no effect on the detection of external appearance abnormalities. It has also been found that even if the deviations completely cancel each other out by chance, the measurement results will not be affected as long as the effect of this phenomenon on the average gradation value is within a preset tolerance.

After the above process, the external appearance of the test object is inspected, and installation automatically detects foreign or missing parts, as well as misalignment. In particular, since the method of the present invention performs multi-tone processing on the gradation level of each pixel, high-precision inspection is possible without increasing the number of pixels, whereas conventional monotone methods do not require increasing the number of pixels. It has the advantage of being able to quickly inspect test objects with complex shapes, such as engine parts, which were previously impossible to handle.

〔Example] Next, details of the present invention will be explained based on illustrated embodiments.

Figures 1 and 2 show how an automobile engine was visually inspected using the wood method.

That is, FIG. 1 shows a captured image of the engine taken from above, and the image is captured in multiple tones using a high-sensitivity CCD (charge-coupled device) or the like. For example, when inspecting the external shape of the part indicated by a in the figure in this engine, the following procedure is performed.

First, the entire image of a standard engine is imaged in multiple tones, and a window A including the inspection area a is set. Next, pixels Gl, G2 .
. . . The multi-level data output from Gn is subjected to multi-value processing, and the multi-value processed data is aggregated. In this embodiment, the tone level is 8 hits and 256 words. The average value of the gradation level is determined from the aggregated data, and this value is stored as the reference value for that window. Further, a tolerance for allowing a certain deviation from the reference value of the window is also set and stored at the same time. The allowable error can be set manually, but it may also be set automatically, taking into consideration the size of the window, the distribution of each multi-tone data, and the like.

Once the reference value, which is the average value of the standard product, and the allowable error value are set, the imaging target is switched to the test object, and the engines that are sequentially supplied to the inspection position and stopped by appropriate transport means are imaged, and the above-mentioned Similarly, an inspection window is set within the captured image, and the average value of the gradation levels of each pixel constituting the inspection window is calculated. The inspection window must completely match the window set when imaging the standard product in both position and size. Therefore, it is preferable to use the same imaging means for imaging the standard product and the subject.

The unique data, which is an average value unique to the subject, is compared with a reference value stored in memory, and if the difference is within a tolerance, the appearance of the subject's test region is determined to be normal.

By the way, when an imaging means such as a CCD is used continuously for a long period of time, the imaging area changes due to the influence of heat etc. even though the fixed position of the camera does not change. Misalignment may occur.

In order to prevent this deviation from the window setting position, it is considered to set a reference point as shown as P in FIG. 1 and set the inspection window based on that position. For example, if the engine is transferred to a belt conveyor in one direction (as shown in the figure), it is preferable that these reference points 1 to P are fixedly arranged at the sides of the belt conveyor. By monitoring the positional relationship, it is possible to correct the misalignment between the window and the inspection site.

The reason why the reference point is provided outside the subject is to eliminate the need to mark each subject as a reference void. However, it is also possible to attach a mark of a specific shape to a specific part on the subject, or to focus on a specific part on the subject and use this as a reference point. In this case, in addition to being able to detect a misalignment between the inspection area and the window due to thermal effects, it is also possible to detect a misalignment in the installation position of the subject.

Incidentally, in the above-mentioned embodiment, only one area was inspected, but as shown in FIG. 3, the windows were set to AI, A2°A3.
It is also possible to set a plurality of windows, such as A4, and perform visual inspections in these windows at the same time. In this case, the tolerance for each window may be set to a few percent, but for windows with particular precision or windows that do not require precision, adjustments may be made by adding individual weights to the tolerance. It's okay.

〔Effect of the invention〕

In the appearance inspection method of the present invention, when comparing the image of the standard product and the image of the subject, the average value of the gradation level of the pixels constituting each window is determined and the average values are compared. , the processing speed can be increased compared to the conventional method in which corresponding pixels within a window are compared point by point.

Furthermore, since the gradation level of each pixel is processed as multi-gradation data, it is possible to prevent appearance abnormalities caused by the cancellation phenomenon of errors that tend to occur in conventional monotone processing from being overlooked.

In addition, with conventional monotone processing methods, the only way to accurately inspect objects with complex external shapes such as engines is to increase the number of pixels, but with the method of the present invention, the gradation level of each pixel can be multileveled. Because it has a 100% resolution, high-precision inspection is possible without increasing the number of pixels. Furthermore, the method of the present invention enables color comparison, which is difficult with the monochrome method.

As stated in the second claim, when the reference point is set near the subject, it becomes possible to detect and correct window deviations caused by temperature rises, vibrations, etc. No measurement errors occur during inspection, and stable appearance inspection is possible.

[Brief explanation of the drawing]

FIGS. 1 and 2 are explanatory diagrams showing a state in which an automobile engine is inspected by the appearance inspection method of the present invention, and FIG. 3 is another embodiment. A: Window, Gl, G2. ...: pixel, K:
Hertconhair, P: Reference point.

Claims (1)

[Scope of Claims] 1) A standard product is imaged in multiple tones, a window is set in the captured image that includes the part to be visually inspected, and multiple tones data of each pixel constituting the window is recorded. A process of performing aggregation processing to calculate the average value of the gradation levels within the window and using it as a reference value, and setting a tolerance that allows a certain deviation from the reference value, and a process of making the arrangement state the same as the standard product. The subject is imaged in multiple tones, an examination window whose imaging area is the same as the window is set in the captured image, and the multitone data of each pixel constituting the examination window is aggregated and processed. and calculating the average value of the gradation level within the inspection window, and comparing the average gradation value of the subject with the average gradation value of the standard product, which is a reference value, and determining that the difference between the two is within the range of the above-mentioned tolerance. A visual inspection method comprising: a step of determining whether a test site of a subject is normal or abnormal by testing whether or not the test site is normal or abnormal; 2) Set a fixed reference point near the subject at a position where simultaneous imaging can be performed with the part to be inspected, set an inspection window at that relative position, and monitor the positional relationship of the reference points. The external appearance inspection method according to claim 1, wherein the deviation between the inspection window and the inspection site is corrected by correcting the deviation between the inspection window and the inspection site.