Laser additive manufacturing technologies have to be integrated into sophisticated process chains including the computer aided manufacturing (CAM) process planning to achieve best overall performance. But currently, additive manufacturing processes are developed to their limits without regards to the CAM process planning. Based on integrated CAM software, which handles specific laser process needs, successful implementations focus on individual geometries and offer solutions for the entire process of complex manufacturing and repairing tasks in laser deposition welding. Due to the change of process conditions and requirements, adequate process strategies need to be developed. Therefore, the capability for controlling these process parameters is considered within the CAM system. Hence, the shortcoming of laser deposition welding, the machining strategies can be fitted to the part's geometry.

Laser deposition welding capacitates efficient production of small batches and high value parts1,2 due to the tool's flexibility. In industrial applications, it is currently regarded as an isolated step in the process chain of additive manufacturing aiming for the best process efficiency. Therefore, adapting the process by computer aided manufacturing (CAM) planning, process control, and adaption of the machining system is an adequate method. However, for achieving best overall performance and thus a solution for robust and economic production and repair, a consistent process chain is mandatory due to the demands of higher effectiveness, automation, and precision. Material defects are the main issue for limiting part's lifetimes in almost every industry application. They can be caused by either the manufacturing process or from wear due to usage. Laser deposition welding is an efficient method to repair worn-out areas. It has been widely applied in industry for small batches and high value parts such as turbine-machinery and tool production to enhance the lifetime and the product quality.

This paper focuses on a method to carry out laser deposition welding in context of repairing parts considering the CAM process planning in order to deliver an optimal result. A process chain for laser additive manufacturing has been developed in which milling processes construct a geometric basis for laser deposition welding, and laser scanning is applied to acquire the part's exact shape and position. With regards to a continuous process chain being assisted by a CAx-chain, including metrology and data acquisition, clamping and referencing, and automated adaptive machining, the advantages of laser processes, geometrical flexibility, and material efficiency are being achieved. The CAM planning of laser deposition welding carries out laser tool path on the scanned geometry to guarantee the process accuracy. Besides this, the required precision is achieved by accurate process control and reproducibility of adaptive chipping processes. By combining these strengths, best overall performance can be reached.

Therefore, a CAM system was developed to satisfy the specific requirements of a precise laser deposition welding process and furthermore to enable the automation of the process and to guarantee the result's quality by minimizing the influences of noises and errors due to other processes. First, the flow of this process chain is introduced:

A laser-based 3D scanning process is utilized for data acquisition of the accurate shape and position of the part. As the outcome of the scanning process, a triangulated mesh contains the polyhedral form of a free-form geometry model with faces, edges, and vertices. Compared to a computer aided design (CAD) model, the mesh model cannot provide continuous and analytic information of the part's geometry. The measuring data is analyzed and prepared leading to a digitized model.

Based on this, by adaptive milling using special strategies and parameters, excavating is carried out to reach defined and adequate preconditions for refurbishing by laser deposition welding. This is followed by recontouring by adaptive milling for according primarily shape. Figure 1 visualizes a process chain for additive manufacturing technologies.

FIG. 1.

Process chain for additive manufacturing.

FIG. 1.

Process chain for additive manufacturing.

Close modal

The five-axis treatment of complex geometric parts is performed by an adequate machining system. Therefore, a gantry system including a numerical controlled (NC)-Rotary/Tilting unit is used for the wire based laser deposition welding. Its milling spindle was replaced by a laser processing head which enables powder and wire feeding and includes process control and surveillance. The machining's accuracy is about 8 μm which is similar to precise milling machines. Figure 2 shows the machining system which is used. Due to the high precision and high dynamic machining system for complex surface treatment, an exact path guidance and high reproducibility accordingly to path position, speed and acceleration3—a requirement for highly sensitive processes—the need for sophisticated process chains can be achieved. For enabling high productivity, the working space enables handling parts of 800 × 800 × 600 mm3 when being lighter than approximately 200 kg for processing five-axis laser deposition welding.

FIG. 2.

Laser machining center with a processing head with wire.

FIG. 2.

Laser machining center with a processing head with wire.

Close modal

Specialized CAM software is vital to the process planning and control of the complex process chain. The objective of the CAM software is to enhance the automation and effectiveness of the manufacturing processes. The main aspects of the CAM software include the following points:

To enable a precise and automated CAM planning of laser deposition welding within the process chain, a laser scanning process is applied to acquire the actual geometry and position information of each part automatically. The mesh data acquired from the laser scanning process after each process step is located in a coordinate system calibrated by measuring the fixture referencing components. This causes obstacles to the CAM planning of the laser process: First, the triangulated mesh represents the discrete surface information of the real part but contains no analytic data of the geometry. Thus, it is very difficult to obtain parameterized geometry features, like curves, edges, planes, etc., for laser tool path calculation. Second, due to various noise sources during the scanning process (surface shininess, ambient light influence, and digital data error in processing), the obtained data usually contains errors in the triangulated mesh (missing of triangles, gaps, blurs, and misplaced face segments) which cannot be eliminated before the CAM planning of laser process.

To overcome the two difficulties mentioned above, a set of statistical methods to calculate laser path on the mesh have been researched and applied in the CAM system. Combined with the empirical knowledge of the geometry characteristics of the process areas, intersection between virtual geometric elements and the mesh can be used to determine the topology of the scanned data in order to acquire the basic geometry feature of the process area.

For example, in the repair process of a blade tip area, as designed in the whole process chain, a milling process should be applied before the laser deposition to ensure the surface quality: The material on the tip with defects will be milled away and a planar surface accomplished. The CAM planning of the laser deposition shall use the planar surface and the border contour of the tip surface as the geometry basis to generate a smooth and accurate laser path on the actual part. The statistic method in the CAM module applies the following steps to acquire the plane:

  1. Find the bounding box of the tip's mesh;

  2. In the range of the bounding box shoot an array of virtual straight lines onto the mesh and record all the intersection points between the lines and the mesh;

  3. Delete the bad points that are the intersection points on other areas and not on the tip;

  4. Use minimum-square method to calculate the formula of the equation of the plane.

After acquisition of the planar tip surface, the contour of the blade tip will be identified by the combination of edge-detection algorithm and smoothing method.4–7 The edge-detection can be realized by the following steps:

  1. Analyze the local curvature on the edge of each triangle facet in the mesh;

  2. Record all the triangle edge segments with a change of normal direction bigger than the threshold;

  3. Eliminate bad edge segments that do not belong to the contour;

  4. Use a numerical method to smooth the edge segments and construct the contour as NURBS (nonuniform rational B-spline) curves.

In the first step, as acquiring the curvature not in an analytic model but a discrete data set, it is necessary to adopt an efficient and accurate-enough numerical method to calculate the value. There have been several researches already investigating this task,4,5 which introduce an equality of the analytic curvature tensor's estimation as

M p = 1 2 π π + π κ p ( T θ ) T θ T θ t d θ ,

where κ p is the directional curvature at the direction T (T is a unit direction vector), to a numerical approximation as

M ̃ v i = v j V i w i j κ i j T i j T i j t ,

where T i j = ( 1 N v i N v i t ) ( v i v j ) ( 1 N v i N v i t ) ( v i v j ) , κ i j = 2 N v i t ( v i v j ) ( N v i t ( v i v j ) . This equation approximates the curvature tensor on each triangular facet in the mesh, by the means of the normal directions of its neighbor triangles. v i   is the current triangle, V i is the set of triangles in its adjacent neighborhood, T i j is the unit vector pointing from v i to v j , and N v i is the normal vector of v i . The w i j is a weight value defining the influence factor of each neighboring triangle to the current one. There are different ways to determine this value and it should also be adapted based on the geometrical characteristic of the mesh. For example, a typical way to calculate the weight is to count on the size of area of each neighboring triangle. In this case, since the triangles obtained from the laser scanning process can be generated with uniformed sizes, “1” can be used as the weight for all the triangles.

With this numerical method, the facets can be labeled in the mesh with the local curvature values. By predefining a threshold of curvature value, the triangles can be extracted that have big potential on the edge out of the whole mesh. Combined with empirical knowledge of the geometrical characteristics of the process area, the accurate boundary geometry from the scanned mesh is derived. For example, when trying to get the outer contour of the blade tip surface, by the curvature method introduced above, not only the triangle facets along the edge have been extracted from the mesh, but also the ones near the cooling holes or other sharp areas in the middle of the surface, which are not desired in this process. Then a numerical method has been applied that by obtaining the external hull of the data cloud, it is very easy to eliminate the undesired facets in the result.

Other boundary geometry elements can also be obtained by similar methods with statistic algorithms so that the errors and noises in the mesh can be appropriately eliminated to guarantee NC program's accuracy.

To ensure high deposition quality, the laser should move on the part surface with uniform interval between each two neighboring tracks on the free-form surface which is represented by a triangulated mesh. This special requirement cannot be achieved by the normal tool path generation algorithms that are designed for milling strategies (constant-cusp, parallel cutting, etc.) and also only for analytic geometry CAD model.

Basing on the acquired geometry information, a rapid algorithm generates the laser tool path in an equal-distance pattern and establishes them on the mesh vertices with geodesic distances from vertices to the initial geometry.8,9 An optimized algorithm based on fast marching method (FMM)10,11 spreads the laser tool path on the mesh in a similar pattern as wave-front propagation.

The FMM method is a numerical algorithm, which is invented to simulate various kinds of thermal/mechanical/chemical processes which have the common pattern of front propagation, such as thermal dispersion, sound wave propagation in material, and flow of viscoelastic fluid. This method is aiming at solving the Eikonal equation

| T ( x ¯ ) | = F ( x ¯ ) , x ¯ Ω ,

where x ¯ is the vector in Rn, Ω is an open set in Rn, F ( x ¯ ) is a function with positive values. The T ( x ¯ ) should subject to the condition T | Ω = 0 . The physical meaning of this function is in the field Ω, the solution T ( x ¯ ) is the shortest time needed to travel from the boundary Ω to the position x ¯ , and F ( x ¯ ) is the unit time cost at x ¯ .

As the velocity is the reciprocal of time, this equation can be interpreted in another way that the propagation speed at each position in a field depends on the local property of the medium. Taking the heat propagation in the material as an example, the necessary conditions to apply the FMM method are

  1. Boundary condition: heat source's type (point, linear, or face) and border of the material;

  2. Propagation field: material's shape and local thermal conductivity which influences the propagation speed at the current position.

Based on these inputs, the FMM can calculate the local temperature at any position in the material at any moment.

The similarities between the process above and the issue in laser deposition are quite intuitive:

  1. Boundary condition: the first laser track on the surface defines a boundary (this is normally defined in the laser process for the certain area on the part);

  2. Propagation field: the surface curvature field which is represented from the mesh.

If taking the initial track of the laser path as a linear heat source, then all the tracks generated afterwards with equal distance with each other are just like the curves which represent the thermal distribution in the heat propagation simulation. Hence, an adaption of the FMM method for the generation of the laser path on the whole surface, just like simulating the process to propagate the heat from the source, is possible. The rapid algorithm to calculate the equal-interval tracks on the mesh can be simply described as steps below:

  1. Define the first track of the laser path on the mesh;

  2. Construct the field by calculating the geodesic distance of each vertex in the mesh to the first track;

  3. Find all the points on the vertices/edges of triangles with the same geodesic distance to the first track;

  4. Use the points to construct laser track by creating smooth NURBS curves;

  5. Iterate the steps 3 and 4 until the whole surface is covered with laser tracks.

The step 2 is a key factor which decides the speed of the calculation since it has to iterate all of the triangles (number n) to build up the geodesic-distance field. The FMM method applies a Dijkstra algorithm to update the field in a very efficient.

Specifically speaking, to calculate the geodesic values of each vertex on the mesh is similar to solve the shortest path problem. The difference is Dijkstra algorithm which is designed to search the shortest path between two nodes in a graph but the shortest path needs to be found between a vertex and the boundary of the mesh, and accordingly get the length of the shortest path as the geodesic distance. The process steps can be described as follows:

  1. Set T(x) = −1 for all the vertices; and T(x) = 0 for all of those on the given initial boundary;

  2. Calculate T(x) of all the vertices that are adjacent to the front, set the calculated T(x) to these vertices, and find the vertex P with smallest value T(x);

  3. Update all vertices' T(x) that are adjacent to P according to the new calculated values;

  4. Jump to step (2) until T(x) of all the vertices on the mesh are calculated.

Within this process, the geodesic distance of each vertex on the mesh to the initial boundary with the same computation complexity of n·log(n) can be obtained. Table I shows the efficiency of the algorithm tested on some mesh.

TABLE I.

Efficiency of the algorithm.

No. Number of triangles Elapsed time (ms)
3991  30 
3436  26 
No. Number of triangles Elapsed time (ms)
3991  30 
3436  26 

As a short conclusion of the method to calculate precise laser path on the scanned data of part, a set of numerical and statistical methods have been employed to identify the geometry features as the basis of CAM planning. Then an adapted FMM method is designed to rapidly calculate the positions of laser moves on the triangulated mesh. These methods have been already applied in several industrial use-cases and proved to be accurate and efficient.

The use of wire, which was researched, increases the complexity of the CAM planning. The wire has to be fed onto the part's surface against the moving direction of the laser tool path. Feeding wire material outside of a certain range causes inconsistent process quality and results. Since the wire conveyor is fixed on one side of the processing head, the feeding direction of the wire is also fixed and cannot be controlled dynamically. Hence, the NC program of the process has to consider that the movement relative to the part's surface fits the wire feeding requirements. Figure 3 shows a picture of the processing head within the CAM planning software.

FIG. 3.

Laser processing head with wire.

FIG. 3.

Laser processing head with wire.

Close modal

This objective is achieved by a special postcalculation of the laser tool path generation. First, a 5-axis laser tool path is generated, and second, the moving direction of each motion is calculated to give the relative angle between the moving direction and the wire feeding direction. If the calculated angle exceeds the range's limit, an extra command for rotating or tilting the table in another direction to readjust the wire feeding direction is added. Accordingly, a path smooth function has to be applied to ensure the machine moving in a fluent way. If the laser path has a complex shape and hence this method can still not achieve the desired direction of wire feeding, this operation must be divided into different segments so that each can be treated separately with desired wire feeding direction.

A specialized postprocessor is implemented to translate the calculated laser tool path to the NC program which is customized for the laser machining system. Therefore, an analysis has been proceeded to comprehend the time sequence of the laser process and the interaction between handling system components including laser controller, inert gas pump, wire conveyor, and machine kinematics. An xml-based postprocessing software, called NCProfiler, has been developed by Fraunhofer Institute for Production Technology IPT for flexible and versatile NC program generation. Hence, the specific commands for laser and material control can be easily integrated into the output NC program. Although the typical scenarios of system control, events have been analyzed and accordingly the commands have been integrated within the postprocessor.

A simulation is indispensable to visually verify the NC program on a certain machining system to avoid process failure and damage due to crashes. A machine simulation component is developed based on ModuleWorks simulation libraries12 and has been integrated into the CAM system. Based on the generated laser tool path, a collision check can be run to check any pair of geometry components from the part or the machining system. It can also check whether the NC program will exceed the limits of machining axes during the process. The machine simulation is shown in Fig. 4. The kinematics is defined within an XML file which enables flexible definitions for other types of machines.

FIG. 4.

Machine simulation.

FIG. 4.

Machine simulation.

Close modal

Effectiveness in laser additive manufacturing technologies requires integration of laser processes into sophisticated process chains. The adaption of the process chain enables combining specific strengths of laser deposition welding. By creating specialized CAM systems, special laser process needs, like the direction of wire feeding and distance between the deposited runs, can be handled and thus offer global solutions for the entire process chain of complex manufacturing and repairing tasks of small batches and high value parts. For validating the research on the CAM supported process planning, it was carried out using a wire out of X38CrMoV5-1 having a diameter of 0.8 mm and providing a deposited run's overall width of approximately 0.7 mm.

This work was supported by the Deutsche Forschungsgemeinschaft within the Cluster of Excellence “Integrative Production Technology for High-Wage Countries.”

1.
G.
Bi
and
A.
Gasser
, “
Restoration of nickel-base turbine blade knife-edges with controlled laser aided additive manufacturing
,”
Phys. Procedia
12
,
402
409
(
2011
).
2.
G.
Bi
,
A.
Gasser
,
K.
Wissenbach
,
A.
Drenker
, and
R.
Poprawe
, “
Identification and qualification of temperature signal for monitoring and control in laser cladding
,”
Opt. Lasers Eng.
44
(
12
),
1348
1359
(
2006
).
3.
F.
Klocke
,
D.
Heinen
,
F.
Schongen
,
K.
Arntz
,
V.
Bäcker
, and
B.
Feldhaus
, “
Wear protection of tools by local geometry and material optimization
,” in
9th Tooling Conference/4th International Conference on Heat Treatment of Tools and Dies
(Gutenberghaus, Knittelfeld, Austria,
2012
), pp.
613
620
.
4.
G.
Taubin
, “
Estimating the tensor of curvature of a surface from a polyhedral approximation
,” in
Proceedings of the International Conference on Computer Vision
(IEEE Computer Society, Washington D.C.,
1995
), Vol. 5, pp.
902
907
.
5.
D. L.
Page
,
A.
Koschan
,
Y.
Sun
,
J.
Paik
, and
M. A.
Abidi
, “
Robust crease detection and curvature estimation of piecewise smooth surfaces from triangle mesh approximations using normal voting
,” in
Proceedings of the Conference on Computer Vision and Pattern Recognition
(IEEE Computer Society, Washington D.C.,
2001
), Vol. 1, p.
162
.
6.
Y.
Sun
,
D. L.
Page
,
J. K.
Paik
,
A.
Koschan
, and
M. A.
Abidi
, “
Triangle mesh-based edge detection and its application to surface segmentation and adaptive surface smoothing
,” in
International Conference on Image Processing
(IEEE Computer Society, Washington D.C.,
2002
), Vol. 3, pp.
825
828
.
7.
J.
Xu
,
Y.
Feng
,
Z.
Zheng
, and
X.
Qing
, “
A Concave Hull algorithm for scattered data and its applications
,” in
Proceedings of the Conference on Image and Signal Processing
(IEEE Computer Society, Washington D.C.,
2010
), Vol. 5, pp.
2430
2433
.
8.
D.
Martínez
,
L.
Velho
, and
P.
Carvalho
, “
Computing geodesics on triangular meshes
,”
Comput. Graphics
29
(
5
),
667
675
(
2005
).
9.
V.
Surazhsky
,
T.
Surazhsky
, and
D.
Kirsanov
, “
Fast exact and approximate geodesics on meshes
,”
ACM Trans. Graphics
24
(
3
),
553
560
(
2005
).
10.
J. A.
Sethian
, “
A fast marching level set method for monotonically advancing fronts
,”
Proc. Natl. Acad. Sci. U.S.A.
93
(
4
),
1591
1595
(
1996
).
11.
R.
Kimmel
and
J. A.
Sethian
, “
Computing geodesic paths on manifolds
,”
Proc. Natl. Acad. Sci. U.S.A.
95
,
8431
8435
(
1998
).
12.
N. N. “ModuleWorks CNC simulation and toolpath verification CAD/CAM components,” see http://www.moduleworks.com/products-services/simulationverification/.