Vision 3D integrated in machine-tool for the alignment of pieces in the rough

The mechanised of pieces in the rough of big dimensions requires high time of previous alignment in machine. The process of alignment requires of two stages: first, characterisation of the geometry of the piece in the rough and marked of references of alignment, and second, his alignment in machine. This process requires personal experienced, and, in addition to the high cost of personnel and availability of machine, exists the risk of not guaranteeing a suitable distribution of sobrematerial available in all the surfaces to mechanise, incurring in the final rejection of piece. The present work presents an approximation of vision 3D integrated in machine-tool based in technology fotogramétrica that allows to increase the level of automation and reliability in the processes of alignment, able to verify and calculate of autonomous form the optimum alignment of a piece in the rough. The potential of the new solution to reduce the total time of alignment is showed in two pilot cases of milling.

1.1. Alignment of the piece in the rough: importance and problematic

An important problem to be resolved by an operario of machine-tool is the initial placing of the piece in the rough before proceeding to his mechanised. The pieces in the rough are produced by processes of precision relatively low, such as the smelting and the welding, and many times do not are used to to have reliable surfaces that facilitate the initial process of positioning.

The initial alignment of the piece in the rough with regard to the axles of machine (fresadora, lathe, etc.) is a critical process, since it has to establish a sufficiently precise placing that guarantee the existence of sobrematerial in all the surfaces to mechanise. An imprecise positioning can cause the existence of an on-excessive thickness in a surface, and to his time, in the opposite side of the piece, the fault of material to be started, causing the rejection of the final piece mechanised. The cost of the piece refused can be very high, no only by the material of the piece, but also by the cost of the process of manufacture realizar until this moment.

At present, the initial alignment of pieces in the rough realizar by means of processes that require long, even in the order of the own time of mechanised for the case of components of big dimensions and value added. They exist different methods of alignment, but all coincide in his essentially manual character and no-repetitive (different according to each geometry of piece), increasing the probability of errors because of human factor.

Like characteristic example, the measurement of the geometry of the piece in the rough can realizar by means of a Machine of Measurement by Coordinates (MMC), but in the case of pieces of big dimensions is more frequent to realizar it planting the piece in a flat surface (gramil) and using manual tools for the measurement of heights (figure 1). The characterisation of the piece realizar comparing patterns of different heights with the dimensions of the final piece (Figure 1, left). Comparing the dimensions measured with the geometry of piece designed to mechanise (defined in planes, files CAD, etc.), mark point of reference on the piece (Figure 1, right) and calculates his optimum position in coordinates machine so that they serve of fiduciales of reference in the alignment of the piece.

After the geometrical control of the piece in the rough and the marked of references on piece, the positioning of the piece in machine is the second and last step before beginning the process of mechanised. In this step measure the coordinates machine in which they situate the marked points of reference in the piece, already was with a palpador or with the own tool of the machine. Of agreement to the deviation between his real position and his optimum position wished, determine the corrections of necessary alignment. The process is iterative until checking a precise positioning in machine of the scoreboards of reference, resulting in a process highly ineficiente and with high cost of availability of machine.

In sight of the cost and the difficulty that supposes the process of alignment of a piece in the rough, identified the opportunity to develop a new method. The aim was to develop a system to characterise pieces of big size by means of tools of portable ‘measurement' (no MMCs), and at the same time, a system that facilitated the final alignment of piece in the machine tool. Another of the clear-cut aims was that the process of alignment could be executed by any operario without the need to have knowledges about the process, advanced to a process of alignment of maximum level of automation and minimum dependency to subject subjective interpretation to human error. Regarding the precision required in the process of alignment, is necessary to begin for considering the sobrematerial half of the piece to be mechanised. For pieces mecanosoldadas of big size can consider an excess of material of 5 mm in the design. In this case, it would owe to accept a total precision of alienation of 1 mm. In the applications considered in this work, with pieces from smelting until 5 m of length, the excess of material considered in the design can easily reach the 20 mm. In this case, the acceptable precision for the system of alignment rondaría between 3 and 5 mm.

At present already exist methods portables available in the market for the characterisation of pieces in the rough. Between the possible solutions are the scan estereométrico and the laser radar. By means of both methods can obtain a representation of the piece in 3D with an acceptable precision. Some of the limitations that present these solutions are been due to the big quantity of information generated for the characterisation of components of big dimensions. The processing of the information of all the piece, and the need to define manually which surface measured has to be compared with the dimensions of the piece designed can suppose a big effort, out of the scope for a no qualified worker. Taking into account that for the positioning of the piece the surfaces to take into account are only the surfaces that go to mechanise, the solutions mentioned previously suppose an on-effort and therefore refused for the system developed.

For the measurement of specific points of piece the most notable methods are the laser tracker and the photogrammetry. In this case it opted by the photogrammetry, because of his low cost and simplicity that presented. The precision that obtains by means of the measurements fotogramétricas is in the order of 1/5.000 of 1/10.000 of the size of the piece, sufficient for the specifications required in the present work.

1.2. State of the art

Exist several automatic methods to position the piece in the rough. Cuypers Et al. They present alternative methods to the use of MMC-s ‘virtual' for the measurement of pieces of big size [1]. One of the practical cases that presents is the measurement of a box of previous changes to his alignment in machine. In this case, it used a method of triangulation, which compared with a system of projection of patterns (scanned). The second method resulted to be much costlier regarding time.

For the alignment of the points obtained in the measurement with the system of coordinates of the piece to mechanise, generally propose approximations where select strategic points like base for the adjust. The result of the alignment offers in representations of the type CAD specifying the excess or fault of material.

The alignment of surfaces of complex geometry has to be realizar before the mechanised, to the equal that the analysis metrológico of the piece mechanised. It has devoted a big effort to resolve this problematic that it is mathematically and computationally complex. Chatelain And Fortín tackled the problem to distribute the sobrematerial of the piece, for like this avoid the problem of fault of material. The article [2] describes the mathematical method developed, that consists in employing a cloud of points in the surfaces to mechanise and the solid model of the piece mechanised final. It developed an algorithm of alignment nolineal that owed to apply of iterative way, using for this the already known algorithm Simplex, which gives place to an optimum solution without deriving the objective function. Later, Chatelain improved the algorithm by means of the utilisation of a logarithmic function of penalty [3], giving place to a faster convergence. Goch Proposes an algorithm for the alignment of complex surfaces [4] by means of the minimisation of Gauss (square minima) and of Tschebischeff (point with greater error). Goch Developed this method further, to be able to use a cloud of points each one with his normal [5], in place to do use of the complete mathematical definition of the surface. Benko Et al. They concentrated in a problem seemed to fit data of several curves and surfaces using the method of reverse engineering [6]. They optimised the algorithm in such a way that it diminished the big computational work that required the treat big quantities of points. Galantucci Et al. They centred in the problem designated ‘registration' (the adjust between surfaces of pieces measured and designed) realizar it in two steps: gross and fine [7], where the adjust gross is realizar by means of the utilisation of neural networks and the adjust fine by means of genetic algorithms.

Nowadays, the adjust of a surface scanned with a surface designed is quite common in the commercial systems of reverse engineering. Some available solutions in the market are the developed by Aicon, GOM, Konika Minolta, Delcam and Creaform. Regarding the precision offered by systems fotogramétricos, Rieke-Zapp et al. [8] they showed that, an accurate calibration of cameras of high quality, can give place to an uncertainty in the rank of 1/50.000.

Like conclusion, there is relatively lower previous work realizar around the measurement of the piece in the rough, and however, quite a lot of approximations proposed on algorithms of adjust of complex surfaces, with availability of commercial solutions. Nevertheless, they have not found notable contributions that treat on the automatic alignment of the piece in the rough.

The aim of the present development is the obtaining of a new method automated for the alignment of pieces in the rough before proceeding to his mechanised. A method automated has to to answer to the following problems:

1. Characterisation (measurement) of the surfaces of the piece to mechanise.
2. Obtaining of the information required of the geometry of the piece of way automated, and independent to the system used for his definition (CAD, etc.).
3. Automatic algorithm of virtual ‘alignment', calculating the optimum lace of piece in the rough and geometry of piece to obtain.
4. Alignment in machine.

The new procedure developed in the present work has been patenting (EP 11380068.4) and provides an integral solution with the following approach:

• Characterisation of the piece in the rough, by means of a solution fotogramétrica fast and of low cost, basing in the measurement of the optical scoreboards planted only in the surfaces to mechanise (generally, flat surfaces and cylinders),
• automatic Determination of the ideal geometry of the piece, by means of the decodificación of the paths of axles of available machine in programs of mechanised standard (ISO in program CNC or CAM),
• Module of virtual automatic ‘alignment', by means of which the coordinates of the scoreboards measured and the ideal geometry of the piece are automatically associated and ranged, calculating the optimum position in coordinates machine of optical scoreboards of reference (fiduciales).
• System of vision 3D integrated in the machine to measure the real position in coordinates machine of the scoreboards used like fiduciales of reference, calculating automatically the correction required in the alignment of piece.

Each module describes in detail in the following sections. The results are presented and argued for two pilot cases of milling (structural components of a machine tool, to which does reference like type To and B in the following sections), where the pieces in the rough (from smelting) have to be ranged before his mechanised with a precision between 3 and 5 mm.

2.1. Characterisation of the piece in the rough

The solution adopted consists in employing the photogrammetry with scoreboards encoded and no-encoded on the piece in the rough. The photogrammetry is a technician used to calculate the coordinates 3D of a number of points of an object by means of a group of images taken from different positions and free orientations. The figure 2 represents the process of photogrammetry. They take several images from different positions by means of a camera. By means of the identification of the points encoded in different images is possible to calculate the position and orientation of the camera in each one of the images, to the equal that the coordinates of the points encoded. To continuation obtain the correspondences of all the scoreboards no-encoded in all the photos, with which realizar a more precise calculation of the position of the cameras and the coordinates 3D of the scoreboards, so much encoded as no-encoded.

The scoreboards encoded identify by means of technicians of artificial vision applied to each one of the photos. One of the most important problems that gives in the practice when realizar measures employing this technical is the identification of the scoreboards encoded when his orientation with regard to the main axis of the camera is upper to 45º. In this case, the scoreboards encoded can not detect of robust form. This problem solved using a polyhedral piece resembled a semiesfera, where planted 14 scoreboards (figure 3, up, left). The angle between adjacent scoreboards is lower that 25º with what attains that at least three scoreboards can be always identified from any point of view of the camera.

The figure 3 (down) shows the process of measurement of one of the pilot cases (type B).

The scoreboards encoded use mainly to calculate the position of the camera, whereas the no-encoded are employed to characterise the surfaces of the piece to mechanise. One of the important advantages to use the photogrammetry is that when employing scoreboards no-encoded only in the surfaces to mechanise, the identification of the surfaces that owe to range can realizar of automatic form. However one of the disadvantages that presents this method in front of the scan complete is the need to plant the scoreboards in the characteristic points (deeper, etc.) of the surfaces to mechanise, to guarantee the control of sobrematerial in all they. Nevertheless, to practical effects, did not consider that it presented a greater problem in the cases tackled. For the characterisation of cylindrical surfaces also employed scoreboards no encoded with magnetic fixation.

The figure 4 sample the result of the characterisation of an in the rough realizar piece by means of the method fotogramétrico (pilot case of type B), facilitated by the coordinates 3D of the scoreboards no-encoded (>100 scoreboards).

2.2. Decodificación Automatic of the geometry

In the state of the current art, the design or final geometry of piece defines by means of archives CAD of widespread use. To be able to compare the piece measured with the final geometry, would be necessary to extract the surfaces of interest of the design from CAD, with a strong dependency with the different existent standards in the structure and format of files. Incidentally, the user would need to define the surfaces that will have to be mechanised by means of a graphic interface that would owe to develop to this end, hampering the automation of this step of the process.

Because of the disadvantages that presents the procedure that bases in the archives CAD, opted by an alternative procedure. The method developed begins for extracting the geometrical information of the paths of finishing of the code of CNC or CAM, since these contain the just and necessary information to define of precise form the geometry of the surfaces to mechanise. Besides, one of the advantages that presents this method is that it does not require of an effort added to define the surfaces to mechanise, offering the potential of a complete automation of this step of the process. The figure 5 sample the result of the automatic method of decodificación for the obtaining of the geometry of the piece (flat surfaces and cylinders) in both pilot cases (type To and B).

2.3. Automatic algorithm of alignment of the piece in the rough

Once obtained the geometrical information of the piece, the following step consists in ranging virtually the piece in the rough with the piece designed. It has implemented the solution with the method of square minima with restrictions, where minimises the distance of each point measured in the characterisation fotogramétrica (figure 4) to his corresponding surface associated to the final geometry of piece (figure 5) subject to the restriction to keep a sobrematerial minimum positive in all the surfaces.

In this step is necessary to know the correspondences of each marker with the surface in which it finds . The method initially relates the marker with the nearest surface, with which, in the majority of the cases, the clear-cut relation will be erroneous. To continuation, displaces and ranges virtually the piece in the rough of agreement to the problem of square minima. Because of this trip obtains a better alignment, and in each iteration recalculan the correspondences between scoreboards and surfaces. To measure that advances with the process of alignment, the number of erroneous correspondences reduces to zero and, once have defined properly all the correspondences, the algorithm converges very quickly.

The figure 6 sample esquemáticamente the automatic process carried out to define the correspondences between scoreboards and surfaces of the piece. To the left of the figure represents the final geometry of piece in black, whereas the points measured and the piece in the rough have drawn in blue and red. The points and the arrows in blue represent the correspondences effected successfully, whereas the represented in red mean that the correspondences are not correct. The figure of the left shows the once effected result the alignment, where can observe that all the points except the indicated in red have associated properly.

Once observed that it keeps the number of associations, and that the errors are low (in the rank of some millimetres, depending on the application), in the following iterations proceeds to apply the restrictions of sobrematerial positive. These restrictions are very simple: the excess of material in each one of the scoreboards has to to be positive and elder that the value that have specified , for example, 1 mm. Once they fulfil these restrictions, the solution is acceptable. On the contrary, if there is not solution that fulfil with the restrictions, the piece in the rough will not be valid to obtain the final piece wished. The big advantage of this method is that it knows if the piece is valid before proceeding to his put in machine, avoiding the costs that would comport the mechanised of a piece that later would be refused.

In the figure 7 can observe results of adjust obtained in the two pilot cases of milling, like characteristic examples of prediction of fault of sobrematerial. The surfaces with sufficient sobrematerial indicate in green, and the zones indicated in red mean that there is not sufficient sobrematerial in the zone. For the pilot case of type To (figure 7, left), have calculated sobrespesores minimum and maxima of 23,77 and -1,66 mm respectively. Of the same way, for the pilot case of type B (figure 7, right), have calculated sobrespesores of 24,37 and -13,70 mm.

In case that the piece was not valid, can adopt different solutions to avoid the rejection and like this can use the piece in the rough, according to the functionality that have the surface where exists the shortage of material and the loss of geometry that generates this fault of material. By means of the welding can contribute material locally until some millimetres where exist shortage of material. For greater faults of material, the mechanised of cashiers and back atornillado of blocks pre-mechanised of basic material allows the recovery of faults of sobrematerial of until several centimetres. For the examples showed (figure 7), could apply the first method of repair to save the pilot case To, whereas the second could be used for the pilot case B. The surfaces that the system developed considers are flat surfaces and cylinders. It would be possible to consider more surfaces, such as the NURBS or other mathematical models used to generate and represent curves and surfaces, as for example, moulds, matrices or álabes of the turbines. For this case, can find several methods in the bibliography (see section 2.1). However, as it has mentioned previously, it considers more interesting do use of the information contained in the archives of program piece that in the archives CAD, with the end to have a universal method and automatizable that was independent of the system to use for the definition of the final geometry of piece.

2.4. Alignment in machine

After obtaining a virtual alignment that guarantee sufficient sobrematerial in each one of the surfaces of the piece to mechanise, through a system of measure by vision 3D integrated in the cabezal of machine verifies the position of the piece in machine measuring the coordinates machine of optical scoreboards of reference had on the piece in the rough. The system of measure consists of a digital camera (5 Mpíxel, JAI BM-500) integrated in the cabezal of the machine (figure 8, left). The taking of images of the scoreboards of reference, with a minimum of two images by each marker (figure 8, right), allows to determine the position 3D in coordinates machine of each marker. The measurement of the coordinates machine of at least three scoreboards of reference allows to determine the location and relative orientation of the piece in the rough. Like result, the difference between the location and orientation measured and the optimum calculated in the previous step (see section 2.3) allows to calculate of automatic form the necessary corrections for a precise alignment of the piece.

The errors obtained when measuring the position of the scoreboards were inferior to 0,1 mm, more than sufficient for this application. Between the main advantages of the integrated system stand out, first, the reduction of the time of measure, when substituting processes of measure by contact by the measure by vision of scoreboards of optical reference, and second, the independence with regard to the geometry of piece for the definition of references fiduciales, when allowing the use of scoreboards of optical reference had on any easily accessible zone selected by the user.

The aim of the present work has been the development of a new system of alignment of pieces in the rough of low cost, robust, reliable and fast, that had a total precision between 3 and 5 mm for pieces until 5 m of length.

Because of the degree of automation and universality of the solution, the use of the system does not require skilled personnel, since the main part essentially manual limits to the placing of the scoreboards fotogramétricos for the characterisation of the geometry of the piece in the rough and the back taking of images (usually of 50 to 200). From this point, proceeds to the calculation fotogramétrico and to the virtual alignment, being both totally automatic processes. Finally, the system estereofotogramétrico of vision 3D integrated in the CNC of machine helps to guide of fast form to the process of alignment of piece, substituting to slow processes of positioned based in approximations by contact.

The aptitudes that the user of the system will owe to have summary in the following: know plant the scoreboards no-encoded in points that characterise properly to the concrete surfaces to mechanise; plant properly the scoreboards encoded and take the necessary images so that the system fotogramétrico resolve of robust form; and carry out the process of alignment of piece with the help of the system of vision 3D integrated in machine. In conclusion, the skills that the user has to dominate are relatively lower that in the processes of conventional alignment.

The procedure described in the present work is independent of the system used for the definition of the final geometry of piece (design CAD, etc.) since this decodifica directly from the paths of mechanised available in universal files type CAM. The automatic and exclusive extraction equipment equipment of the surfaces to mechanise allows the calculation disregarded of the optimum alignment of piece through a module of virtual alignment.

Although in the present work have opted by the photogrammetry for the characterisation of the piece in the rough, basically by his low cost and simplicity, the laser tracker could be a very reliable alternative by his widespread use. By the expense of a higher price and probably greater time of measurement, could obtain solutions more precise.

[1] Cuypers, W., they Go Gestel, N., Voet, To., Kruth, J.P., Mingneau, J., Bleys, P. Optical measurement techniques for mobile and large-scale Dimensional metrology. Optics  and Laser in Engineering, 2008, Vol 47, No.3-4, pp 292-300.

[2] Chatelain, J. F., Fortin, C. To balancing technique for optimal blank part machining. Precision Engineering, 2001, Vol 25, No 1, pp 13-23.

[3] Chatelain, J.F. To level-based optimization algorithm for complex part localization. Precision Engineering, 2004, Vol 29, No 2, pp 197-207.

[4] Goch, G. Efficient Multi-Purpose Algorithm for Approximation and Alignment Problems in Coordinate Measurement Techniques. Annals of the CIRP, 1990, Vol 39, No 1, pp 556-556.

[5] Goch, G., Tschudi, Or. To universal algorithm for the alignment of any sculptured surface. Annals of the CIRP, 1992, Vol 41, No 1, pp 597-600.

[6] Benko, P., Kos, G., Várady, T., Andor, L., Martin, R. Constrained fitting in reverse engineering. Computer Aided Geometric Design, 2002, Vol 19, pp 173-205.