WO2010112894A1

WO2010112894A1 - Automated 3d article inspection

Info

Publication number: WO2010112894A1
Application number: PCT/GB2010/050464
Authority: WO
Inventors: Edmund Peter Sparks; Richard Winterbottom
Original assignee: Roke Manor Research Limited
Priority date: 2009-04-02
Filing date: 2010-03-19
Publication date: 2010-10-07
Also published as: GB0905718D0

Abstract

The present invention provides a method for inspecting a surface of an article. The method comprises the steps of providing a software model of the surface; positioning the article before a camera to provide an image of an observed portion of the surface of the article; aligning the software model of the surface and the image; determining a portion of the image to be examined for defects, in accordance with the observed portion of the surface of the article; and applying a defect detection method to the determined portion of the image to identify defects upon a corresponding portion of the surface of the article.

Description

AUTOMATED 3D ARTICLE INSPECTION

The present invention relates to the inspection of surfaces of articles. It is particularly of use in a production environment for automated inspection of manufactured articles for conformity with a given specification. Unfinished articles may be inspected at points within the manufacturing process, and completed articles may be inspected once finished.

Unfinished articles which fail to meet the given specification may be processed to correct the defect, if possible; may be recycled as raw material; or, if neither of these is a practical option, disposed of to avoid wasted processing into a finished article.

The present invention will be discussed with particular reference to the inspection of turbine rotor blades, but may be applied to any manufactured article which is required to meet manufacturing tolerances. Examples include machine parts, and automotive components.

Known solutions to this problem include employment of skilled labour to perform manual visual inspection. This is relatively costly, time consuming and may be inaccurate due to the subjective nature of the test and operator fatigue resulting from the repetitive nature of the task.

Complex and accurate parts are required, for example in Aerospace and automotive industries. Safety critical systems require very effective inspection for defects, and this represents a significant cost. The present invention provides faster inspection, which allows earlier yet reliable detection of defects, reducing waste and increasing productivity. In the field of industrial inspection, a system of the present invention may be expected to inspect thousands of articles per day.

The present invention provides article inspection methods as defined in the appended claims.

The above, and further, objects, characteristics and advantages of the present invention will become more apparent from the following description of certain embodiments thereof, in conjunction with the accompanying drawings, wherein:

Figs 1-5 illustrate some images used in methods according to the present invention.

The present invention employs a comparison between an image of an article and a CAD, or other, software model of an expected surface of the article. The software model and a corresponding portion of an observed image are compared, and any discrepancy may be interpreted as a defect. The type of discrepancy may be used to determine the type of defect, and this information may be used to decide whether to correct the defect, or to return the article for recycling, or to dispose of it. Results from the inspection method of the present invention may also be used to improve the respective manufacturing procedure to reduce the incidence of such defects in the future.

According to a feature of the present invention, the software model defines which parts of an image are inspected for which type of defect. The software model may also determine which orientation the article should be in to detect that type of defect. The light angle and article position in any particular image restricts which defects it is reasonable to search for in that image. The detection of defects is carried out, in each case, by comparing an observed image with a corresponding portion of a software model. According to the present invention, only a portion of each image need be examined for evidence of defects, representing a significant saving of time and processing power as compared to an arrangement in which the entirety of each image, or even the entirety of the representation of the article within each image, is examined.

The present invention requires a software model of an expected surface of the article. This may be derived from a CAD design of the article, if available, or may be created from observation of an example of an acceptable article. For example, known touch-probe arrangements may be used to map a surface of an example article, and a software model may be generated from those observations.

In a method according to an embodiment of the present invention, an observed piece is held in a suitable retainer, such as a lathe chuck. The software model is aligned with an image of the article captured by a suitably positioned camera, typically by using a least-squares fitting method on certain points of the software model. The software model, or the article, or both, is/are re-positioned, typically by rotation, until the least-squares fitting method determines that the software model is correctly aligned with the image of the article. During this procedure, some points may show a large error in least-squares fit and that error may not reduce, despite the majority of points indicating a close alignment. According to certain embodiments of the present invention, those points may be ignored, and the remaining points used to align the software model with the image of the article. The ignored points may be taken to represent defect sites.

Preferably, alignment of the software model and the image article converges, and the fit of all points during convergence is assessed. Preferably, points are progressively ignored in relation to the error in their fit.

Once the software model and the image are aligned, any regions of significant deviation of the image from the software model may be highlighted for remedial action or manual confirmation of a suspected defect.

The article and the corresponding part of the software model are compared in this manner, with the article in a first orientation. The article, the camera, or both, is/are then re-positioned, to present a different part of the article to the camera. The software model is also rotated to align with the new position of the article. This may be arranged in alternative ways.

The article may be held within an accurately-controlled retainer, such as a computer-numerically-controlled (CNC) lathe or a milling machine, and the software model and the article may be initially aligned. Thereafter, the retainer and the software model may each be rotated by an identical angle, to a new orientation. The software model and the image of the article may be expected to remain at least approximately in alignment. Small adjustments may be carried out in a re-alignment process at each orientation of the article. The article may be installed within its mount with arbitrary alignment between the article and the software model, and the described least squares fit, or equivalent, used to minimise total errors and align the image and the software model together.

If the article is not held in an accurately controlled retainer, the least-squares algorithm may be used again, as described above, for the new orientation of the article, to realign the software model with the image of the article. The software model may attempt to track movement of the image to provide approximate alignment of the model and the article before the least-squares fit algorithm is employed. Alternatively, no attempt may be made to track the changing relationship between the image and the software model, with a complete least-squares fit being performed, as described above, for each orientation of the article.

With the article in its new orientation, the aligned software model is compared with the observed image of the article, and any significant discrepancies may be categorised as described above.

Feedback from the re-alignment process to a manipulation device controlling positioning of the article may allow calibration of the manipulation device.

This procedure is typically repeated until the whole of a surface of interest of the article has been examined. The surface of interest may not represent the whole surface of the article, as some regions may be identified as being of no interest for the purpose of this comparison. For example, cast or other moulded articles may include "runners": these are protrusions which are solidified moulded material which hardened in the access channels to the mould cavity in which the article is formed. Similarly, features of the article may be provided to allow it to be gripped in the retainer, such as a chuck of a lathe. The manufacturing process in question will include a step of removing these runners and other features, so there is no need for these to be considered in the inspection method of the present invention. There may be parts of the article for which defects such as may be detected by the present invention may be tolerated. The surfaces of such parts may be excluded from the inspection of the present invention. The software model may include definitions of which parts of the surface of the article are to be included or excluded from the inspection.

By providing a single stationary camera and a rotary mount, such as a lathe chuck, the article may be rotated about an axis to provide the camera with radial views of an entire outer surface. Such camera may, however, not form images of axial end surfaces of the article. This may be cured by providing multiple cameras, viewing the article from different angles in order to obtain images of all sides of the article or moving the camera to at least one further position to obtain further images of the article. Alternatively, once images have been derived of the article throughout a complete revolution, the article may be repositioned within the retainer (e.g. a chuck of a lathe) and rotated about a different axis, for example an axis perpendicular to the previously-used axis, to allow images of the remaining surfaces to be obtained, while using only a single stationery camera. A single camera, with the article rotated to thirty different positions before the camera is essentially equivalent to thirty static cameras and corresponding lighting sources, but may be less expensive and onerous to provide. The angle of lighting may be important in obtaining clear images of the article, and light sources may be arranged as required to provide images of an acceptable contrast. It may be necessary to change lighting arrangements to emphasise types of defect. For this purpose, multiple light sources may be arranged around the article, and selectively activated to generate multiple images of a same part of the surface of the article, to allow different types of defect to be identified more clearly in respective images. For example, surface protrusions and depressions are more clearly apparent if lighting is used with a direction of incidence almost parallel to the surface in question; while edges of an article may be most clearly identified in silhouette. The software model may define which parts of the surface of the article should be inspected for which defect, under a range of lighting conditions, for a particular position of the article.

In an improved embodiment, the software model may be used by a controller to orient the article in preferred positions, for example using a controlled robot, to provide preferred orientations for defect detection.

While the above arrangements particularly describe a rotating article, with one or more fixed cameras and one or more fixed lighting sources, other arrangements are possible within the scope of the present invention. For example, the article and light sources may be in fixed positions, while one or more camera(s) move(s) about the article. Alternatively, the position of the light sources may be fixed, or limited, with respect to the position of the camera(s), with the camera(s) and light source(s) moving together about the article. In another arrangement, the article, the light source and multiple cameras may all remain stationary, provided that a sufficiently large number of cameras is provided to observe the entire surface of interest of the article. In such methods, the lighting angle and intensity is important in observing defects. For complex three-dimensional articles such as turbine blades, a lighting arrangement which is suitable for imaging one part of the surface is likely to be unsuitable for imaging another part of the surface of interest, even if visible within the image. Preferably, for each orientation, a surface portion of interest is defined, for example within a data structure of an overall controller circuit or software. Rather than analysing all points in the image for conformity with the aligned software model, only points within the surface portion of image for that orientation are analysed. The article may then be moved to another orientation, and a surface portion of interest for that orientation may be observed. The surface portions of interest are preferably defined such that each point of the surface is analysed only in one orientation, and unnecessary re- evaluation of a single point in multiple orientations is avoided.

It may be decided to observe the entire surface of interest two, or three, times, to be sure that all defects have been detected. However, each point on the surface should preferably be observed only once during each observation.

In a preferred arrangement, each image is compared with the correspondingly aligned software model separately, and there is no need to provide alignment between images. In each comparison, sites identified as possible defect sites may be recorded, and presented to an operator after automated inspection for corrective action, or disposal. If the automated inspection identifies a defect which is not significant, the operator may be permitted to override this classification and pass the article on for further processing. In some embodiments, identified defects may be labelled, for example being picked out by visible laser, or marked with a visible paint or dye, for operator inspection. The paint or dye may be chosen to be visible to the operator only under certain lighting conditions, for example under UV light.

In some embodiments, it may be possible to arrange automated repair of detected defects, for example using a robot-operated grinder.

In certain embodiments, an overall controller circuit or software may be arranged to determine whether a detected defect is significant or not. For example, small recesses or protrusions in a surface may be deemed insignificant, while larger ones are significant enough to cause the article to be repaired or reprocessed. Different tolerance levels may be applied to different parts of the surface of interest. For example, some more critical regions of an article may be defined as having a maximum recess or protrusion height of 0.1mm, while a height of 0.3mm may be tolerated in other parts of the surface of interest.

Multiple tolerance layers may be provided: for example, in a certain part of the surface of interest, a maximum recess or protrusion of 0.05mm may be deemed to pass the test; recesses or protrusions of up to 0.3mm may be deemed to cause remedial reworking of the article, while recesses and protrusions of more than 0.3mm may cause the article to be re-cycled as raw material.

Acceptable tolerance specifications may be defined in the software model: they may already be present if the software model was derived from the CAD design of the article; alternatively, they could be added to a software model derived from an example article. Defect/fault detection is performed by methods specific to each fault type. If a point fails the least-squares fit, or similar, that point generates an error. The types of defect associated with such errors may include geometric defects, regions deformed out of tolerance limits, missing sections, small surface defects, protrusions, recesses, holes and porosity. Each is detected in a manner specific to the type of defect sought.

Small surface defects such as protrusions, recesses and holes are identified by viewing the surface when illuminated by light directed at a small angle to the surface. Porosity can be located using penetrating dyes; from comparison of the surface texture to that expected from the computer model; in comparison to another part of the same piece; or by comparison with other similar articles inspected and deemed to be acceptable.

The automated system of the present invention is reliable, in that it provides known coverage of all parts of the surface of the article, and can generate statistical information on defects. This statistical information may be used to improve the manufacturing process to reduce the incidence of the various types of defect detected. It can track the development in the occurrence of each type of defect.

Certain examples of embodiments of the present invention will now be discussed, with reference to the accompanying drawings.

Before an attempt is made to align an image with the software model, the camera used should be calibrated. Camera calibration is the process of describing the geometry of the imaging system, for example including camera, lens, and digitiser, in terms of a parameterised model, for example specifying a focal length and aspect ratio. Typically, calibration of the camera will involve determination of intrinsic parameters of the camera, such as focal length, aspect ratio, optical centre and distortions. Such calibration is necessary to allow images captured by the camera to be interpreted and related to the software model, enabling correlation between a captured 2D image and a 3D software model. External parameters affecting interpretation of obtained images must also be determined. These may include offsets in position and angle of view between multiple cameras, if used; and determination of the axis (or axes) of rotation of the article within the image. These parameters must also be measured to allow images captured by the camera to be interpreted and related to the software model, enabling correlation between a captured 2D image and a 3D software model.

The angle between camera-article-illumination is critical to the effectiveness of defect detection and frequently multiple cameras/illuminators will be required.

In the example of a moulded component, various faults which may be expected include: partly filled mould; air bubbles retained in the mould; porosity; residual material from the mould remains on the part; incorrectly formed mould; damage to the pattern used to create the mould.

In some applications, it may be deemed sufficient to detect only certain types of possible faults. In other applications, every fault type needs to be checked for, particularly where there is no manual back-up. It has been found that the inspection method of the present invention is effective at detecting defects in highly polished surfaces, which have been found difficult to detect by human operator inspection.

Fig. 1 illustrates an article 150, in this case a turbine blade, as captured in an image according to a step in a method of the present invention. A software model of the article has been aligned to the image of the article, and is represented by wire frame lines 152, 154, 156, superimposed on the image and representing features such as edges and transitions in the software model. According to a step in the method of the present invention, only the portion bounded by line 152 is examined for defects in this image. Those parts of the software image relating to other portions of the article, such as those bounded by lines 154, 156 are used only to align the software model to the image.

According to a step in the method of the invention, the captured image is compared to the software model within the selected portion 152, and any differences examined to determine whether they might indicate defects.

In Fig. 2, possible defect sites 162, 164, identified by the method of the present invention within the determined portion, are illustrated. By comparing this drawing with Fig. 1, it can bee seen that these sites of possible defect showed up as bright spots in the image of Fig. 1. The defects may be protrusions or recesses, for example. The relative positioning of the illumination and the camera, and the angle of the imaged surface, will determine to some extent which defects show as bright spots, and which show as dark spots.

Fig. 3 shows another article 170, in this case a fan rotor, as imaged in a step of a method according to the present invention. A software model of the article has been aligned to the image of the article, and is represented by wire frame lines 172, 178, superimposed on the image and representing features such as edges and transitions in the software model.

According to a step in the method of the present invention, only portion 174, representing a single fan blade is examined for defects in this image. Those parts of the software image relating to other portions of the article are used only to align the software model to the image.

According to a step in the method of the invention, the captured image is compared to the software model within the selected portion 174, and any differences examined to determine whether they might indicate defects.

In this case, a defect is identified at 176. Where the corresponding part 178 of the software model indicates that a part of the fan blade should be present, the image shows no part of the fan blade. The portion 176 of the image is therefore identified as a possible defect site. This appears to represent a missing portion of the article, and may be due to any of several causes, for example: under-filled mould; damaged mould; damaged pattern used to create mould or mechanical damage to the blade after moulding. In the illustrated example, the blade 174 is imaged by reflected light. In an alternative, back-lighting may be used, in which case the blade may show up as a dark region of the image, while a missing portion may show as a bright region in the image.

Fig. 5 shows a part 180 of an article in which porous regions 182 are shown. These regions may be highlighted by careful illumination, or application of a dye or stain which will assist detection of the porous region. The periphery of each porous region 182 may be detected by the method of the present invention as an unexpected feature, not present in the software model. This potential defect may then be identified and signalled for manual verification.

Claims

1. A method for inspecting a surface of an article, comprising the steps of: (a) providing a software model of the surface;

(b) positioning the article before a camera to provide an image of an observed portion of the surface of the article;

(c) aligning the software model of the surface and the image;

(d) determining a portion of the image to be examined for defects, in accordance with the observed portion of the surface of the article; and

(e) applying a defect detection method to the determined portion of the image to identify defects upon a corresponding portion of the surface of the article.

2. A method according to claim 1, wherein the step of aligning comprises rotating the software model to align with the image of the article.

3. A method according to claim 1, wherein the step of aligning comprises rotating the article, to align the image with the model

4. A method according to claim 1, wherein the step of aligning comprises rotating both the article and the model into mutual alignment.

5. A method according to any preceding claim, wherein the method further comprises:

(f) re-positioning the article such that the image shows a different observed portion of the surface of the article; and

(g) repeating steps (c) to (e).

6. A method according to any preceding claim, wherein the method further comprises the step of repeating steps (f) and (g) a plurality of times.

5

7. A method according to any preceding claim wherein the portion of the image to be examined for defects is determined by identifying the portions of the surface of the article visible in the image, and selecting parts of the image which correspond to parts of the surface which are

10 required to be examined, and which are presented in the image at an appropriate angle for the applied defect detection method to be effective.

8. A method according to any preceding claim wherein two or more defect detection methods are applied, and wherein different portions of

15 the image are examined for defects by different defect detection methods.

9. A method according to any preceding claim wherein the step of aligning comprises identifying corresponding positions within the image and the software model and identifying points corresponding to such

20 positions; calculating a least-squares fit on the positions of the points; and aligning the software model and the image to minimise the misalignment of the points as calculated by converging of the least-squares fitting method.

25 10. A method according to any preceding claim wherein points which do not converge to a fit in the least-squares fitting method, despite a majority of points so converging, are ignored for the purpose of the least-squares fit, and are identified as possible defect sites.

11. A method according to claim 10 wherein the fit of all points during convergence are assessed and points are progressively ignored in relation to the error in their fit.

12. A method according to any preceding claim, comprising the step of determining whether a detected defect is significant.

13. A method according to any preceding claim wherein the defect detection step comprises applying differing tolerance levels to different parts of the surface of interest.

14. A method according to any preceding claim wherein the defect detection step comprises applying differing tolerance layers to detected defects; thereby determining which of a plurality of categories a detected defect falls in to, selected from among the following: causes the article to be reworked; re-processed; disposed of; or wherein the defect is deemed not significant.

15. A method according to any preceding claim, further comprising the step of collating statistical data on detected defects.