Method for classifying casting defects within the framework of an X-ray analysis

Abstract

The invention relates to a method for classifying casting defects in a casting within the framework of an X-ray analysis, wherein the casting defects are each automatically allocated to a known casting defect type by comparing the X-ray image of the casting with features from training images of known casting defect types and the casting defects present in the examined casting are thus established.

Description

CROSS-REFERENCE TO RELATED APPLICATION

This is a utility application that claims foreign priority benefits under 35 USC §119 (a) to German Patent Application No. DE 10-2005-019800.7, filed 28 Apr. 2005, which application is incorporated herein by reference.

BACKGROUND OF INVENTION

The invention relates to a method for classifying casting defects in a casting within the framework of an X-ray analysis.

Methods are known in which the classification of a casting defect is limited to a set value/actual value comparison with a specification. The specifications used are based in general on surface-based features. With these known classification methods the result is a simple good/bad decision. This means that in the automated method it is merely decided using an automatic defect recognition system (ADR system) whether the casting currently under examination by X-ray analysis has a casting defect or not. The disadvantage is that the casting defect actually present can only be established afterwards with the help of manual or visual assessment.

Moreover, it is known that the founder carries out the classification of a casting defect by viewing the X-ray image of the casting. A great deal of experience is necessary for this, and the casting defect is usually also detected only afterwards, i.e. if the examined casting has an unacceptable casting defect above the specification limit. This is extremely fatiguing for the founder who, with this method, must continuously monitor the live image at the casting machine, which under certain circumstances can result in false assessments as to the casting defect involved. The result is that although the parameters of the casting process are modified, a wrongly accepted casting defect cannot be remedied. The result is an avoidable, additional rejection of castings with casting defects compared with the situation had the casting defect actually present been correctly recorded.

SUMMARY OF INVENTION

The object of the invention is therefore to provide a method with which a casting defect actually present in the casting can be established automatically—i.e. without the fatiguing monitoring process for the founder.

This is achieved according to the invention using a method according to the features of claim 1. Because the X-ray image of the casting taken during the X-ray analysis is compared using features that have been calculated from training images of known casting defect types, the casting defect actually present in the examined casting can be established. This results in the automatic allocation according to the invention of the detected casting defect to a known casting defect type. This means that a rapid reaction by modifying the parameters of the casting process can then be initiated. Additional waste material is thereby avoided. Unlike the known method in which the founder must continuously carry out a fatiguing observation of the live image of a casting machine, with the method according to the invention the specific casting defect is established automatically. In addition, the casting defect can also be detected below the specification limit, with the result that the question of rejects simply does not arise if the trend towards a casting defect is counteracted at an early stage by appropriate adaptation of the parameters of the casting process.

An advantageous development of the invention provides that the known casting defect types be grouped into classes that have comparable features in the X-ray image. It is advantageous in particular if the classes with comparable features in the X-ray image are grouped into superclasses. It is equally possible, instead of the two given levels, to carry out even further groupings at levels above these. A structured implementation of the method is thereby possible in which the whole database of training images does not have to be constantly run through in order to arrive at a match with all known casting defect types. Time can be saved by the rough classification by type, and thus a real-time classification of the casting defect is possible. By saving additional time, an even earlier reaction with regard to the modification of the parameters of the casting process is possible, with the result that the quality of the castings can be still further improved.

In particular, the method according to the invention is particularly efficient if the classification is based on a decision tree which is used to proceed from the abstract superclasses or classes with simple features to the specific casting defect with complex features. A very reliable yet very rapid classification of the specific casting defect is thereby possible, as the few first-stage superclasses can be examined very rapidly and then only the branch of the decision tree under the simple features of which the detected casting defect falls need be pursued further. This then applies analogously to each further level of the decision tree until the specific casting defect with the complex features has been detected.

A further advantageous development of the invention provides that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect. Thus the pressure on the founder is even further eased and he must merely make a correction if the system makes a rough false classification. Otherwise, the progressive development of a casting defect is already counteracted at the earliest possible point, with the result that the trend towards this casting defect is automatically reversed by the adaptation of the parameters of the casting process. This results in a marked reduction in rejects, as the specification limit of the casting defects is not exceeded. Accordingly, defect-free castings are produced for the most part so that the required number can be produced in less time.

BRIEF DESCRIPTION OF THE DRAWING

FIG. 1: Further advantages and details of the invention are shown with the help of the embodiment of a decision tree. The single FIGURE shows an embodiment of a decision tree according to the invention with which a method according to the invention method can be carried out.

DETAILED DESCRIPTION OF DRAWING

As a rule, ADR systems can also detect defects below the specification limits. If these defects are reported to the founder in good time, he can introduce appropriate countermeasures, depending on the casting defect type, before the critical limit of this casting defect type—the specification limit allocated to it—is exceeded. In order to subsequently improve the casting process, it is also necessary to provide detailed statistics on the casting defects that have occurred. Also of particular interest in this connection is how many parts display which casting defect type in the presence of which parameters of the casting process. A prompt, detailed classification of the casting defects that occur is therefore required in order to predict a trend or compile statistics on the defect types.

The represented embodiment shows an adaptive decision tree which facilitates a detailed classification according to the requirements of the respective user. Thus each user can himself decide which specific casting defects he would like to detect at the lowest level and how he would like to group these casting defect types at a level lying above it or would like to further group them at levels lying above these so that the topmost level contains only single features about which a decision can very easily be reached.

Using a large number of sample images, the training images, of found casting defect types or also theoretically predetermined casting defect types, the individual levels can be trained from the roots to the leaves. For this, in each level specific features must be allocated with the help of which a decision is made for a specific type—at the lowest level the specific casting defect type. It therefore suggests itself to divide the features at the topmost level, where merely simple features are to be decided, down into the more complex features for the specific casting defect types.

In the specifically cited embodiment, only a simple threshold value comparison between light and dark takes place at the first level, the superclasses. It is thereby established whether the casting currently under examination has a higher density or a lower density than that which was predetermined.

If the density is higher, it is assumed in the present embodiment that an inclusion—at the second level, i.e. the level of the classes of casting defect types—is involved. In the present case, the “higher density” superclass is not split any further, but this is by no means mandatory. Nor is the “inclusion” class further divided, which likewise is by no means mandatory.

The “lower density” superclass covers a total of three classes at the second level in the embodiment: “shrink hole”, “blowhole” and “surface defect”. Here also, different, fewer or also more classes can always be defined—depending on the user.

In the embodiment, the “shrink hole” class is subdivided into the specific casting defect types “single shrink hole”, “cluster of shrink holes” and “sponge”. At this lowest level, the allocation is carried out to the highest level of detail.

In the represented embodiment, the “blowhole” class is subdivided into the specific casting defect types “single blowhole” and “porosity”.

In the represented embodiment, the “surface defect” class is divided into the specific casting defects “die mark” and “blacking defect”.

As already stated above, the whole decision tree can be redesigned, depending on the application and intention of the user, using suitable training images according to the respective requirement. An individually tailored, detailed classification with respect to different casting defect types is thereby made possible. Instead of the shown division of the decision tree, a different division—as already indicated above—at the lowest level of the specific casting defect types is also possible. The only important point is that the appropriate training images are made available to the system in each case so that all the casting defect types relevant for the respective casting process can be recognized and the specification limits are best never exceeded, so that efficiency is increased during the casting process.

Claims

1. Method for classifying casting defects in an examined casting having a casting defect within the framework of an X-ray analysis comprising:

acquiring an X-ray image of the examined casting;

automatically comparing the X-ray image of the examined casting with features from training images of known casting defect types; and automatically allocating the casting defects present in the examined casting to a known casting defect type.

2. Method according to claim 1, characterized in that the known casting defect types are grouped into classes that have comparable features in the X-ray image.

3. Method according to claim 2, characterized in that classes with comparable features in the X-ray image are grouped into superclasses.

4. Method according to claim 3, characterized in that still further groupings take place on one or several levels.

5. Method according to claim 2, characterized in that the classification is based on a decision tree which is used to proceed from the abstract superclasses or classes with simple features to the specific casting defect with complex features.

6. (canceled)

7. Method according to claim 3, characterized in that the classification is based on a decision tree which is used to proceed from the abstract superclasses or classes with simple features to the specific casting defect with complex features.

8. Method according to claim 4, characterized in that the classification is based on a decision tree which is used to proceed from the abstract superclasses or classes with simple features to the specific casting defect with complex features.

9. Method according to claim 1, characterized in that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect.

10. Method according to claim 2, characterized in that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect.

11. Method according to claim 3, characterized in that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect.

12. Method according to claim 4, characterized in that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect.

13. Method according to claim 5, characterized in that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect.

14. Method according to claim 7, characterized in that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect.

15. Method according to claim 8, characterized in that, once the specific casting defect has been automatically recognized, the casting process is automatically controlled and the parameters of the casting process are automatically modified according to the recognized casting defect.