1 Introduction

Modern additive manufacturing approaches have the potential to add significant flexibility to manufacturing processes. In additive manufacturing, an object is built up step by step from raw material. While 3D printing is well established for plastics, additive manufacturing is more complex for metals, resulting in a plethora of manufacturing approaches with different characteristics.

A well-established approach is wire arc additive manufacturing (WAAM), where a metal object is welded from scratch onto a metal ground body using wire filler [1]. Using an industrial welding robot, this approach allows relatively large quantities of metal to be processed in a short period of time [2,3,4,5,6]. WAAM is based on arc welding, where an electric current induces the heat necessary to melt both the wire and the surface to which the wire material is attached. Various shielding gases are used to protect the weld from interaction with the air. In addition to solid wires, cored wires are also increasingly used in wire-based additive manufacturing [7,8,9]. A typical manufacturing system is shown in Fig. 1.

Fig. 1
figure 1

Welding robot system at Clausthal University of Technology [10]

Full size image

Arc-based additive manufacturing offers great potential for industrial applications today due to new developments in the fields of robotics, welding technology, and computer-assisted manufacturing [6, 11,12,13]. Two core topics are currently the focus of research, in addition to others. One is the production of geometry, geometry errors, etc. There are still open questions here, particularly with regard to path planning and the dependence of the geometry on the selected process variables. The second issue is the adjustment or determination of the achievable properties, as these are crucial for the use of the technology in industry. For low-alloy steels in particular, the mechanical properties that can be achieved depend on the thermal history of the material [14,15,16,17,18].

The combination of both areas into a geometry and property-oriented additive manufacturing has so far been little discussed in the literature for arc-based welding processes. Some work has been done by Chaurasia et al. [19,20,21], Richter et al. [22,23,24], and others [25] on the usage of camera systems for tracking the temperature field of the weld pool. The main outcome of these works is that tracking the temperature field can lead to increased geometrical and microstructural homogeneity.

The correlations between cooling conditions and emerging properties can serve as a starting point for such a consideration, and therefore, the production of components with locally adjusted properties, such as strength and toughness, is the overreaching goal Fig. 2.

Fig. 2
figure 2

WAAM part idea of different localized properties

Full size image

The correlation between the material properties and the selected energy input and the number of thermal cycles is already known for a number of materials [26]. In particular, for low-alloy steels, which have an allotropic phase transformation, the course of the temperature history is of decisive importance for the properties generated in the component. The temperature history is dependent on three central factors: the energy input, the interlayer temperature (in additive manufacturing determined by the time until over-welding), and the heat conduction conditions, which are determined by the component geometry. The melt pool size or the melt pool volume also depends on these three boundary conditions.

In the presented work, an approach is presented to realize property-oriented additive manufacturing from the interaction of property-oriented path planning and a system for controlling the melt pool size. By controlling the weld pool size, the cooling of the material can be adjusted within certain limits and a local adjustment of the microstructure, and the mechanical properties can be achieved.

The first part of the work will briefly describe a new approach to realizing property-oriented path planning, Fig. 3. This is based on a rudimentary thermodynamic simulation of the build-up process, which has been kept simple enough to allow an execution in a very short time, to obtain a path planning via an adapted SAT solver, which takes into account defined cooling conditions in certain component areas [10]. In this publication, the results of the system identification are used for investigations of the material properties. The changes in the microstructure that result, in a first approach, from the step changes in the process parameters.

Fig. 3
figure 3

Schematic of the process steps that need to be taken to realize property-oriented control in WAAM

Full size image

Finally, a microstructural characterization will show that different microstructure morphologies and properties can be realized in a component by combining property-oriented path planning and the use of a control loop to regulate the melt pool size.

1.1 Property-oriented path planning method

One method for the realization of material property-oriented path planning was suggested by Ehlers et al. [10]. In that paper, it was shown how to search for welding paths with a satisfiability solver (SAT). The basic idea is to discretize the object to be manufactured into blocks and represent the connections between these blocks in the form of a graph. Two blocks that can be produced consecutively are connected by an edge in the graph. For their initial study of planning welding paths under cooling time constraints in 3D, the discretization of the object parts into cubes was performed, but the approach presented is not restricted to cubes. A search was then performed in the graph for a fragmented Hamiltonian path, i.e., one consisting of several independent paths that together visit all vertexes. Additional constraints on the paths encode the process-related requirements. The approach was to check whether the planned fragmented path satisfied the required t8/5-times after finding a satisfactory assignment to all variables. The cooling conditions were calculated by a simplified simulator. Figure 4 shows a visualization of the simulator state in the middle of a welding process of a simple geometry.

Fig. 4
figure 4

Temperature field in a fast temperature simulation of a simple WAAM part [10]

Full size image

1.2 Microstructure and thermal history

Understanding the microstructure evolution during additive manufacturing is key to realizing certain material properties. In the case of continuous cooling or heating, the corresponding CCT or CHT diagram describes the metastable microstructure formation. In general, however, these data cannot be directly extrapolated to additive manufacturing conditions due to the, in many cases, different cooling conditions between WAAM, multilayer welding, and the conditions under which the CCT/CHT diagrams are generated. Repeated thermal cycling around certain temperatures will result in a different microstructure.

The influence of a cyclical temperature evolution on microstructure formation during welding was analyzed by Samardžić et al. [27] and Huang et al. [26] using a welding simulator and a quenching and forming dilatometer, respectively. These physical simulators allow the application of transient, particularly cyclic, temperature profiles. The resulting dynamic transformation temperatures for multiple temperature cycles can be determined. Microstructural investigations and hardness measurements were also carried out. Celin et al. [28] investigated the microstructure development in different areas of the heat-affected zone during the multipass welding of a fillet weld for a high-strength fine-grained structural steel. The multipass welding process results in thermal cycles corresponding to those of the WAAM process. The microstructure during the actual welding process was compared with that obtained from dilatometer simulations using a temperature profile similar to that during the welding process. It was concluded that dilatometer studies are well suited to simulate the microstructure evolution in the heat-affected zone under cyclic temperature conditions. Several working groups have investigated microstructure evolution during the WAAM process. Some results are published in [16, 29,30,31,32,33,34,35]. For low-alloy steels, in general, they showed that at least four different types of microstructures could occur. In the top layers, a microstructure comparable to single run weld metal is present. Beneath the microstructure changes and is comparable to multi run welds. Due to the extensive reheating beneath this zone, the microstructure changes again and forms a microstructure comparable to a normalized steel sheet. The fourth zone is the transition area between the weld metal and the substrate. Here, again, a microstructure comparable to multilayer welded joints occurs. The underlying relationships need to be understood and are the basis of a local adjustment of material properties. Afterwards, the temperature field distribution can be controlled to achieve certain material properties. In the following, the realization of a measurement and control system is presented, and the first results for local adjusted material properties are presented.

2 Materials and methods

2.1 Experimental setup

The experimental setup consists of three core components, the welding robot, the visual detection of the melt pool size and temperature, and the control system. These are presented individually below, except for the welding robot (a common kuka robot).

2.1.1 Visual detection of the melt pool size and temperature

The determination of the melt pool expansion and the temperature is the basis for a local adjustment of the mechanical properties. Only on this basis is it possible to develop a control system for specific heat input as a function of heat dissipation. The system presented below is based on work that has already been published [22,23,24]. A two-color pyrometer measures the temperature of an object via the ratio of the light emission of several wavelength ranges. These wavelengths are used as reference points. Under the condition that the object behaves like a black or gray radiator at the selected wavelengths, the emissivity need not be considered [36]. In our solution, two narrowband optical filters (central wavelength of the narrowband filters: 680 nm and 860 nm) are used in front of the cameras to block out the arc. The spectrum of a low alloy steel with the corresponding filter wavelength is shown in Fig. 5. The selection of the correct filter must be material-dependent due to the different emitted wavelengths of each material in metal vapor–based plasma. The basic radiation of the argon plasma stays the same, but in GMA welding, the metal vapor is dominant.

Fig. 5
figure 5

Result of the measurement of the spectrum from 530 to 1000 nm with possible wavelengths λ1 and λ2 for the ratio pyrometer [22]

Full size image

The emitted wavelengths of the metal vapor plasma depend on its composition. In GMA welding, this in itself depends to a large extent on the filler metal used. These filters are matched to the emission spectrum of the plasma generated by the arc and define the support points. This enables optical temperature measurements of the molten pool. The jitter of the two triggered digital camera images used to determine the surface temperature is much smaller than the exposure time of the images. Therefore, with the temporal resolution of the camera's exposure time, the synchronized cameras capture the same instant. The general setup of the measurement system is shown in Fig. 6. The data processing is described in [22,23,24]. The results are gray images of the weld pool and derived temperature field images, Fig. 7. In combination with the solidification temperature, the melt pool area can also be derived from the images.

Fig. 6
figure 6

Optical setup of the ratio pyrometer. The two beams in blue and red represent the imaging paths of two points in the weld pool on the cameras [22]

Full size image
Fig. 7
figure 7

Captured gray images, resulting temperature field and measured weld pool area [22]

Full size image

2.1.2 Control of the weld pool area

Fig. 8 shows the experimental setup and the list of components and used process variables. A hollow cylinder with a diameter of 120 mm was manufactured for the initial study. The process variables were chosen to give the control system a sufficiently wide range of possible adjustments. In addition to the melt pool, the temperature in front of the weld pool and the interpass temperature were measured using a conventional pyrometer. In this way, rising interpass temperatures can be compensated or the process can be set on hold when the interpass temperature exceeds certain limits to assure a stable and controllable process. The mechanism for interpass control is described in further detail in [37]. The limits of the manufacturing process must be taken into account as additional nonlinearities. It is possible to reduce the energy input over a sufficiently large range without reaching the process limit and without introducing nonlinearity. The fabrication of the cylinder was chosen to have a stable observation on one point, without any influences of changing the direction of welding and/or observation. This allows a longer welding time without additional ignition processes. Interruptions in the production process led to the complex interrelationships of material and energy input already mentioned and are therefore avoided. A detailed description of the possibilities and the operation of the system is given in [38].

Fig. 8
figure 8

Experimental setup [38]

Full size image

The melt pool area can be used as the controlled variable. The aim of the control system developed is to keep the melt pool area as constant as possible at a given set point. Initially, only the wire feed rate was used as a setting variable. In further work, the system will also take into account other manipulated variables, such as arc voltage correction and welding speed.

2.2 Material and initial welding conditions

A low-alloy filler metal (DIN EN ISO 14341-A G 50 7 M21 4Mo / A5.28 ER80S-D2) using an M21 shielding gas (82% Ar, 18% CO2) was selected as the material for the investigations. The chemical composition is shown in Table 1. The substrate for the tests was 14-mm-thick plates of S355.

Table 1 Chemical composition of the filler
Full size table

The typical microstructure in single pass welds of this filler consists of acicular ferrite and grain boundary ferrite, Fig. 9 [39, 40].

Fig. 9
figure 9

Typical weld metal microstructure of the filler (Le Pera etchant) [39, 40]

Full size image

In the presented results, the possibility of locally achieving two different microstructures will be demonstrated by setting two different set points of the wire feed speed. The first one for a “small” weld bead with a wire feed speed of 4.5 m/min (heat input ca. 330 J/mm) and the second one a “large” weld bead with a wire feed speed of approximately 5.4 m/min (heat input ca. 400 J/mm). The set points were synchronized with the rotation of the substrate to match the same position in every layer, Fig. 11. In a second step, the control loop was closed, and the targeted weld pool area was set to 90 mm². The design and boundary conditions of the controller have been described in [37, 38]. There, it is shown that the controlled system can be described as a quasi-linear system with a static nonlinearity and can be described using a Hammerstein model. The block diagram of the controlled system is given in Fig. 10.

Fig. 10
figure 10

Block diagram of the closed loop control for the manufacturing process using the Hammerstein structure [38]

Full size image

The identification and design of the system leads to a controller variable K of:

$$K(s)=\frac{5.7s+1}{1.9s}$$
(1)

2.3 Microstructural characterization

Following the welding tests, the additively manufactured hollow cylinders are cut, and the cross-section of the small and large bead parts is made, Fig. 11. These were then embedded in a polymer, abraded, and polished. The final polishing was performed with a grain size of less than 1 μm in the lapping suspension. When grinding, care was taken to ensure that the area affected by cutting was reliably removed. Subsequently, the samples were etched with Nital. The microstructure images were taken after preparation of the samples with a Leica CTR-6000 reflected light microscope at different magnifications. Hardness testing was performed in accordance with ISO 6507. A line of microindentations was made to determine the average hardness distributed over the cross-section.

Fig. 11
figure 11

Manufactured cylinder with changing process variables

Full size image

3 Results

Fig. 11 shows one of the sample geometries. The surface is uneven, typical of WAAM-made samples. The width of the layers between the small and large beads are slightly different. The difference in width deviates between 0.2 and 1 mm. Especially in the upper layers, the difference between the two parts increases due to the different heat input to heat dissipation ratios.

This change in the geometry leads to the assumption that a different cooling behavior has been reached and that the microstructure should be significantly different.

3.1 Microstructure

The microstructure of approx. the 30-mm high hollow cylinder wall structures can be subdivided into four different zones, Fig. 12. The bottom layer, which is just above the substrate, will not be discussed in detail in this paper due to the relatively large heat conduction compared to the wall-like structure in higher layers and the dilution with the base material. A microstructure typical of multilayer welds can be seen in the bottom layers.

Fig. 12
figure 12

Microstructure comparison of three different zones and two heat inputs

Full size image

Fig. 12 shows an overview of the cross-sections of two different heat inputs. In addition, Fig. 12 shows an overview of the resulting different microstructures of the four zones. In the upper zone, the microstructure is typical for a single-layer weld or cover layer typical of this filler, cf. Fig. 9. Typically, this filler forms acicular ferrite with small amounts of grain boundary ferrite in single pass welds.

Looking at the two heat inputs, there are clear differences in the microstructure in the top layer with regard to the amount of grain boundary ferrite and the grain size of the accicular ferrite. The microstructure from the area with higher energy input, i.e., the larger the weld pool is, the coarser the microstructure. This corresponds to the current state of knowledge on the behavior of this group of materials.

The coarser structure due to the increased energy input also continues in the other two identified microstructural areas. Below the top layers, the accicular ferrite initially dissolves, and globular ferrite or bainite is formed. This dissolution increases towards the bottom layers due to the more frequent reheating, culminating in a ferritic-pearlitic microstructure in the third zone in the high-energy input region with significantly larger grains than the low-energy input, Fig. 13. In the case of a lower energy input, a ferritic-bainitic microstructure forms, Fig. 14.

Fig. 13
figure 13

Microstructure comparison of the most reheated region for two heat inputs

Full size image
Fig. 14
figure 14

Microstructure comparison of the most reheated region for two heat inputs, SEM images

Full size image

In summary, it can be concluded that the formation of the microstructure corresponds to the current state of knowledge [16, 41,42,43,44,45]. Moreover, the small difference in energy input leads to significant changes in the microstructure morphology. This is especially true in regions with a high number of reheating events.

3.2 Hardness

Figure 15 shows the distribution of the hardness over the cross-section of the additively manufactured material. The different microstructures in the specimen result in different hardnesses of the material. First, a reduction from the top to the middle of the specimen can be seen. This is due to the dissolution of the ferritic ferrite and the formation of ferritic-perlitic or ferritic-bainitic microstructures. From the middle to the bottom, the hardness then increases again due to the increased amount of bainite or perlite formed.

Fig. 15
figure 15

Hardness distribution within the additive-manufactured steel

Full size image

Furthermore, due to different heat inputs, on average, the hardness at the low-energy input side is approx. 20HV0.1 above the high energy input side. This behavior was also expected and is in line with the current state of research [15, 16].

3.3 Closed-loop control

By closing the control loop and thereby regulating the melt pool size to approx. 90 mm², the effects of an increasing interpass temperature on the geometry expression could be significantly reduced. Fig. 16 shows the comparison between the cross-section of the sample geometry with and without the control loop activated. Due to the closed control loop, a reduction in the surface waviness and a reduction in the deviation of the width of the specimen can be achieved. Furthermore, it can be seen that due to the closed control loop, the width of the component is almost constant over the entire height.

Fig. 16
figure 16

Sectional view of the manufactured parts with open-loop (left) and closed-loop (right) control. The red-dashed line shows the measurement points of the micro-identification to determine the hardness distribution [38]

Full size image

Figure 17 shows the area of the weld pool, the welding power used and the interpass temperature during the tests. A hollow cylinder with 30 layers was produced in each case, and manufacturing one layer took 40 s The closed-loop controller with integral behavior reduces the welding power to 5% as the interpass temperature rises, and after 20 to 25 layers, constant temperatures are reached. Furthermore, the closed control loop allows the weld pool area to be kept constant. Without closed-loop control, this is not the case. To keep the area of the weld pool constant, the welding power must be reduced. This also lowers the measured interpass temperature.

Fig. 17
figure 17

Comparison of disturbance response for open-loop (black) and closed-loop (blue) control [38]

Full size image

4 Summary and outlook

The presented work reviews the first realization of a system for property-oriented arc-based additive manufacturing and shows the implementation on a first simple example. The implementation of the system includes the following:

  • A fast calculation system for weld path planning

  • The implementation of a monitoring system for the weld pool

  • The setup of a control system for the size of the weld pool area

On the basis of the example produced, it was possible to demonstrate that the possible controlled variables for the welding process are sufficient to be able to set significantly different microstructures and properties of the material. With this basis, the geometric fidelity of WAAM components can now be significantly improved in addition to the targeted adjustment of the microstructure and properties. In conclusion, it can be said that the results obtained so far make it possible to locally adjust properties in the WAAM process.