Approach toward the application of mobile robots in laser materials processing

Laser material processing (LMP) has been one of the established manufacturing processes for years. This can be justified with the large variety of processes such as laser cutting, welding, structuring, or also additive manufacturing by means of lasers as well as the possibility to produce single and series components quickly and directly from CAD data.^1–3 

Machine tools are typically adapted or industrial robots are used as the kinematic system for LMP.^4,5

However, these tools do not make much use of the advantages of LMP, such as the force-free process. It should also be noted that the size of these standard systems must scale with the size of the parts.⁶ Accordingly, the use of resources and costs increases with the size of the parts. Innovative kinematic systems such as mobile robots (MRs), which move on the components during processing, could provide a remedy.⁷ This means that the size of the MR is independent of the size of the component to be processed.

The aim of this paper is to work out how such an MR for LMP should be designed in principle and whether the hypothesis can be validated prototypically.

In the literature, there are already some approaches to the use of MRs in material processing. Here, MRs are mostly equipped with an articulated arm robot. The MR provides a mobile platform for movement in the working environment, while the actual processing work is carried out by the jointed-arm robot. Due to the design, these are usually cost-intensive solutions. The accuracy of the mobile platform is rarely specified, but in Ref. 8, it is given as ±10 mm.^8–10 

MRs are also used to make production lines more flexible or to support manufacturing processes. Here, however, a combination of MRs and jointed-arm robots is also used. Accuracies of the MR are also given in Ref. 11 with ±10 mm.^11,12

In LMP, however, only a few approaches to the use of MR can be found. In Ref. 13, an MR equipped with an articulated arm robot for laser metal deposition is presented. Here, however, the MR serves only for the general positioning of the system and does not perform any movement during the laser process.

Based on the results of the literature research, the following problem can be derived: No mobile robot system (without additional kinematics) exists yet, which seems to be suitable for laser materials processing. To investigate the described potential, nevertheless, an MR for LMP will be built, presented, and investigated in the following.

Within this paper, the following research questions will be answered:

• What are the restrictions for mobile robot systems caused by the laser process?

• Which functional structure results from the given restrictions?

• What kind of kinematics and which core components are derived from the functional structure and how must they be designed?

• Can the theoretical proof of the suitability of the mobile robot system for LMP also be demonstrated prototypically?

To answer the initial research questions, the methodology of the agile product development will be applied, which has already been used in the product development process for several years(Ref. 14, pp. 106–108). The core idea of the agile product development is to identify and prioritize issues in the product development process (Ref. 14, pp. 106–108). The issues can be derived from the functional structure of the system and prioritized on the basis of the development risk (Ref. 14, pp. 107–109). To answer the last research question, a laser welding process is carried out.

The experiments can be divided into two subtests. In the first part, the experiments to determine the accuracy of the mobile robot (see Fig. 1) and the sensors used take place. For this purpose, a command trajectory, a parallelogram with side lengths of 500 and 1500 mm (see Fig. 2), is traversed, and all sensor data are written down. Since the mobile robot moves over the component, a component length of 2000 mm is assumed. The coordinate systems of the sensors are synchronized before the start of each measurement series. To determine the accuracy of the MR itself, a reference measuring device is used, and the data are evaluated. To determine the accuracy of the sensors, the measurement error between the sensor data and the reference measuring device is determined. The error is calculated as the Euclidean distance. Furthermore, three different process speeds (see Fig. 2) are investigated. This series of experiments is performed without a laser process on an aluminum sheet.

As a reference measurement device, a laser tracker (LT) type “Leica Absolute Tracker AT930” from Hexagon Metrology GmbH is used. These devices offer an exact determination of the movement of the MR. The LT offers an absolute angular performance of ±15 + 6 μm/m (maximum reference error of 39 μm in the presented experiments).¹⁵ The reflector of the LT is mounted in the upper area of the MR.

In the second part of the experiment, a laser welding process is performed to test the MR under process conditions. For this purpose, a 0.5 mm thick sheet of stainless steel (AISI 304-1.4301) is welded onto a 2 mm sheet of stainless steel and the welding seam is evaluated metallographically. The process parameters were determined in previous tests. The parameter for the laser power $P L$ and the process speed $v P$ can be found in Table II. The laser beam diameter of the machining plane is $d L$ of ∼200 μm. The laser used for the experiments is a 1000 W single-mode fiber laser of type redPOWER® QUBE 1 kW from SPI Laser. For all welding experiments a lightweight processing head of type FiberMINI-R QBH 75/200® from Optoprim respectively Laser Mechanisms, Inc. is used. The metallographic images of the welds are taken at 200× magnification using a VHX-6000 digital microscope from Keyence. The specimens are separated, ground, and polished before taking the images.

In the following, the described research questions will be answered one after the other. First, the restrictions resulting from the laser processes will be discussed. In the second step, the functional structure of the system is derived from this. This results in the type of kinematics and core components to be further investigated. After the design of the core components and the description of the overall prototype, the validation under real process conditions will be discussed.

The main task of kinematic systems in LMP is to perform a relative movement between the laser optics and the workpiece. In this context, the laser beam must follow a defined trajectory, which is derived from the subsequent component geometry (Ref. 16, p. 454).

Accordingly, a kinematic system must be able to perform any change in the direction at any time. Furthermore, it can be assumed that the programmed trajectory should be traversed as precisely as possible to achieve the highest possible component.

Several options are available to quantitatively evaluate accuracy. In this paper, the general tolerances defined in DIN ISO 2768-1 should be used. These are ±2 mm in the tolerance class “medium” for nominal dimensions between 2000 and 4000 mm.¹⁷ 

Another important process parameter is the process speed. Since the MR is to be used universally for various LMP operations, a wide range of different process speeds must be considered here. According to Ref. 18, a process speed starting at 20 mm/s can be assumed for laser welding. Laser hardening processes are already carried out at a process speed of up to 70 mm/s (Ref. 19, p. 179). For laser cutting, process speeds of up to 170 mm/s are realistic (Ref. 20, p. 404). Accordingly, a maximum process speed $v P$ of up to 200 mm/s can be assumed as reasonable for the MR in order to be able to cover future developments in process and laser system technology.

The restrictions can be summarized as follows to answer the first research question:

• Movement in all directions must be possible;

• Accuracy in the range ±2 mm; and

• Speed in the range of 200 mm/s.

The functional structure of the MR can be derived from the general structure of robot systems found in the literature. Siciliano et al. (2009) define actuators, sensors, and controllers as the essential elements of a robot (Ref. 21, pp. 2–3). Kinematics describes the basic structure of robots, which consist of rigid bodies connected by joints (Ref. 22, pp. 11–13).

Since all elements are also part of MR, the kinematics, actuators, sensors, and controllers will each be defined as a functional structure and, thus, as questions to be answered early in the product development process. The investigation of the control system is part of the further research work and is not presented in this paper. The results concerning the integration of the optical components for LMP are also considered individually.

Regarding the kinematics, the following question can be derived:

• What kind of kinematics is suitable for an MR for LMP, especially under the background of the later field of application?

Regarding the selection of a suitable kinematic system for the MR for LMP, different types can be distinguished in the literature. On the one hand, a distinction is made between legged robots (Ref. 22, p. 1203), track robots (Ref. 22, p. 1540), and wheel-driven (Ref. 22, p. 575) mobile robots. Furthermore, within these groups, a distinction can be made between MRs that have holonomic or nonholonomic characteristics.

A nonholonomic robot, here, for example, an autonomous car, cannot perform any lateral movement. This means that not every direction is freely and directly accessible (Ref. 23, pp. 45–46). MRs with omnidirectional drive systems, on the other hand, are considered holonomic robots (Ref. 22, p. 578). Accordingly, only MRs with omnidirectional drive systems will be considered further. No literature can be found on the topic of the legged omnidirectional MR in material processing, so this area will not be investigated further. Although there are approaches to the use of omnidirectional track robots,²⁴ the majority of research focuses on wheeled omnidirectional robots. In this respect, these will be considered further here.

Regarding the number of wheels, only four-wheeled mobile robots are considered for use in LMP. This can be justified by the high possible speed and advantages in terms of stability (Ref. 22, pp. 582–585).

Furthermore, the wheel diameter must be considered. This has a direct influence on several factors for a given revolution of the motor in a given time. With a larger wheel diameter, a longer distance is covered at a higher speed. However, the minimum possible movement of the MR increases per incremental revolution of the motor. Since for the later application in LMP a trajectory as exact as possible is required, a wheel diameter as large as possible should not generally be selected here. Accordingly, a medium diameter is aimed at for the MR planned here. Omnidirectional wheels are available in different sizes; here, a diameter $d W$ of 100 mm should be selected, and this also corresponds to standard systems in the literature.²⁵ 

The schematic diagram of the MR with an omnidirectional drive is shown in Fig. 3.

Based on these results, the third research question can be answered as follows: A wheeled MR with an omnidirectional drive system should be used as a kinematic system for LMP. The core components are the actuators, sensors, and control. The dimensioning of the components will be presented in the following sections.

Regarding the actuators, the following questions can be derived:

• Which motor is suitable for the MR?

Furthermore, it must be estimated which motor can be suitable for the MR. In particular, the torque of the MR must be taken into account. According to Ref. 28, the torque for acceleration $M M , a$ can be calculated according to formula (3). For this purpose, the mass moments of inertia of the motors $J M$ as well as of all wheels $J W , 4 W$⁠, the mass of the MR $m MR$⁠, the coefficient of efficiency $η$⁠, the wheel diameter $d W$⁠, as well as the speed change $Δ n a$ and the acceleration time $Δ t a$ are used.

The parameters used to calculate the acceleration torque are shown in Table I.

Since the MR must accelerate from a standstill to the process speed $v P$⁠, the rotational speed $n P$ is equated with the change in the rotational speed $Δ n a$⁠. The acceleration time $Δ t a$ is set to 0.5 s.

The mass of the MR $m MR$ estimated here is 50 kg.

With these parameters and formula (3), the acceleration moment can be calculated to be 1.13 Nm. A safety factor of 50% is calculated to ensure the driving capability. This results in a minimum torque of the stepper motors of 1.7 Nm.

Furthermore, the minimum motion of the MR system must be considered for the smallest possible motion of the stepper motor. The background to this is that stepper motors can only execute a rotation around the defined step angle (Ref. 31, p. 577). According to the kinematics model [see formula (1)], the translatory movement of the omnidirectional wheel in x and y directions, starting from the rotatory rotation of the motor, corresponds to the translatory movement of the MR. The smallest possible movement of the MR shall be less than 10% of the tolerance requirements to ensure the tolerance requirements. This has been defined as ±2 mm in Sec. III A. Accordingly, a minimum movement $x MIN$ of the MR smaller than ±200 μm must be possible.

On this basis, the second question can be answered, and it can be stated that a stepper motor with a minimum torque of 1.7 Nm should be suitable for the MR. Accordingly, the ST5918L2008-B stepper motor from Nanotec is selected for the kinematic system.²⁹ The stepper motor has a torque of approximately 1.9 Nm and can be operated in the microstep mode. This allows the division of a full step of 1.8° into several partial steps. With this motor, a subdivision into 64 partial steps is possible, and accordingly, the minimum possible revolution is calculated to be 0.028°. Converted, a minimum translatory movement of the MR of 24.5 μm is possible.

The question about the motion of the MR system without slip can only be validated by a prototype and will be answered at the end of this article.

Regarding the sensor technology, the following questions can be derived:

• What kind of sensor system is suitable for an MR for LMP, especially under the background of the accuracy requirements?

• Can the motion of the MR be recorded with this sensor system in such a way that the measurement data can be used as input for a control system?

For the sensors to be used here, the same accuracy requirement and associated measurement resolution of ±200 μm from Sec. III D should be taken as a basis.

The measurement of wheel revolutions by encoders is listed as an internal position measurement. Here, it is to be considered, however, that likewise, a summation of the errors can develop. In addition, slippage occurring between wheels and underground cannot be determined by the encoders. To reduce the inaccuracies, it is recommended to use several encoders.³² 

Various encoders are available for the stepper motors described above. The matching incremental NOE2-05-K06 encoder offers a measuring resolution of 4000 pulses per revolution.³³ This corresponds to a measuring resolution of 0.09°. Converted to the translatory motion of the MR, using formula (6), this results in a translatory resolution of 78 μm. This means that the encoders should, in principle, be suitable for position determination in accordance with the given requirements.

Camera systems can be used as an external sensor. These are mounted on the MR maker, which are captured by several cameras. In this way, a three-dimensional movement of the MR is calculated. A suitable camera sensor is the motion tracking system from Qualisys, which offers a 3D resolution of about 110 μm.³⁴ Thus, the measurement system seems to be just suitable for an application in this case.

Thus, the first question of the sensors can be answered as follows: There are numerous sensors for the position determination of MRs. With regard to accuracy requirements on the process side, the position measurement via internal encoders and an external camera system should be suitable. The second question, like the question regarding the actuators, must be answered with a prototype.

Regarding the optical components, the following question can be derived:

• How/where is the laser optics integrated into the system?

Since MRs with omnidirectional drives have the ability to move from all positions in all directions, the motion of the MR can be described as translation as well as a twisting around the center of gravity (Ref. 26, p. 513). For four-wheeled MRs with an omnidirectional drive, the center of gravity is in the middle of the MR (Ref. 26, p. 523). To reduce the computational and control effort, the x and y positions of the laser optics are placed in the center of gravity of the MR. This point will be referred to as the tool center point (TCP) (see Fig. 3) in the following. This is possible because the laser tool used is a rotationally symmetrical tool for which the rotation about the normal to the working plane is irrelevant.

To answer the third and fourth research questions, the following result can be summarized: An MR with omnidirectional wheels driven by stepper motors seems reasonable. Internal encoder measurements and an external camera system should be used to determine the accuracy. The complete experimental setup of the MR is presented in the following.

The resulting experimental setup of the MR is shown in Fig. 1. The laser optics (1) (type “FiberMINI® II,” Laser Mech) is mounted in the center, and the setup is generally realized on an aluminum frame (5). The dimensions of the system are about 450–600 mm. The movement is performed by the four omnidirectional wheels (7) (100 mm Mecanum Wheels, Nexus Robotics Ltd.), each with a stepper motor (type “ST5918L2008-B-NEMA23,” Nanotec Electronic GmbH & Co. KG). The motors are driven by a motor controller (3) (type “drylin® D1,” igus® GmbH). The movement is calculated on a pocket-sized computer (4) (type LattePanda 3 Delta, LattePanda). The supply of process media (2) and power (6) is carried out in the upper area or laterally.

The revolutions of the wheels are recorded by an encoder (7) (type NOE2-05-K06, Nanotec Electronic GmbH & Co. KG) and read out and stored with an Arduino. As an external sensor, the Qualisys motion tracking system will be used.

This section describes the results of prototype validation. In particular, the question of whether the MR can move without slippage is addressed here as well as whether the sensor data can be recorded with sufficient accuracy. Furthermore, the result of the laser welding test is presented.

In the first experiment, the accuracy of the MR is investigated. The result is shown in Fig. 2. It can be seen here that the command trajectory cannot be traversed at any process speed. The attained geometry is measured by the reference measurement device, the laser tracker (LT).

The Euclidean error at the end of the command trajectory (command: x = 0, y = 250) is between 85.52 mm (50 mm/s), 98.02 mm (100 mm/s), and 105.12 mm (200 mm/s).

However, it can also be seen that the Euclidean error to the lower side of the parallelogram (command: x = 0, y = −250) is significantly larger with 220.66 mm (50 mm/s), 213.07 mm (100 mm/s), and 188.03 mm (200 mm/s).

It can also be seen that the programmed corners, especially at 100 and 200 mm/s, are only approached rounded.

In Fig. 4, the Euclidean error of the camera system (Qualisys) relative to the laser tracker is plotted over the normalized position. Here, it can be seen that the error is almost always above the expected value of 110 μm at all process speeds.

In Fig. 5, the Euclidean error of the encoder relative to the laser tracker is plotted over the normalized position. Here, it can also be seen that the error is almost always above the expected value of 110 μm at all process speeds. The maximum Euclidean error of the encoder measurement is 127.53 mm (50 mm/s), 97.52 mm (100 mm/s), and 68.84 mm (200 mm/s).

The maximum Euclidean error of the camera measurement is 5.10 mm (50 mm/s), 5.15 mm (100 mm/s), and 7.74 mm (200 mm/s).

In Fig. 6, the error of the camera system and the encoders at a process speed of 200 mm/s is plotted for the first 10% of the attained trajectory. Here, in contrast to other diagrams in Figs. 4 and 5, the Euclidean error of the encoders is significantly below the Euclidean error of the camera system.

To answer the question of whether the mobile robot system can perform a laser welding test, the results are presented below.

Four different parameter sets (see Table II) are applied for execution. Here, the laser power $P L$ and the process speed $v P$ are varied.

In Fig. 7, the mobile robot system is shown during the laser welding test. Here, the sheets that are to be welded together are placed on a laboratory table and the mobile robot moves over the sheets.

The images of the metallographic evaluation of the samples are shown in Fig. 8. Here, it can be seen that a continuous weld seam can be made between the two sheets for all parameters. Sample 2 is the only one in which a collapse of the sheet surface can be observed, whereas in samples 3 and 4, the surface bulges.

Thus, the last research question can be answered, and it can be shown in principle that a mobile robot system is suitable for laser materials processing and that a weld seam can be produced.

The selected kinematic system as an omnidirectional setup seems to be suitable for a mobile robot system for laser materials processing. However, especially at higher speeds, the programmed corners of the command trajectory are rounded, so that the component geometry cannot yet be followed. However, this can be attributed to the controllers of the motors, which are programmed in such a way that the process speed should be as constant as possible.

The stepper motors selected and used seem suitable in principle for the mobile robot system for laser materials processing, and the calculated torque is sufficiently dimensioned. However, it can be determined in the tests that there is a lot of slip between the mobile robot and the metal surface. Due to this slip, the command trajectory of the mobile robot system could be traversed, and there is a large deviation. It can be observed that the slip is direction dependent. This can be explained by the fact that not all wheels always turn in the same direction in different directions. When the mobile robot travels in the x direction, all four wheels turn in the same direction, but when it travels in the y direction, two wheels turn against each other. When moving in 45° direction, only two of the four wheels are used. This can result in different force transmission. It can also be noted that not all of the four wheels are always in contact with the component surface. This can be caused by an inaccurate assembly of motors and wheels. This can also be reflected in different force transmission. Both points taken together can also lead to twisting of the mobile robot, especially during 45° travel.

The sensor systems investigated, an external camera system and an internal encoder measuring system, cannot meet the requirements.

The inaccuracies of the camera system are with max. 7.74 mm clearly smaller than those of the encoders and can be attributed to the following reasons. On the one hand, the metallic surface of the component used has a high degree of reflection. The infrared radiation used by the camera for measurement is, therefore, strongly reflected. The cameras influence each other through their positioning, so that the measurement result could be influenced. Furthermore, the inaccuracies can be attributed to the manual alignment of the coordinate systems of the reference measuring device and the camera system, since this cannot be carried out with extreme precision and repeatability. Furthermore, the camera system restricts the processing space of the mobile robot system, which is disadvantageous for a later application on the largest possible component. In addition, the costs for the camera system are significantly higher than those for the internal measuring system using encoders.

With the internal measuring system using encoders, the rotation of the stepper motors is precisely recorded. Using the kinematics model, the x and y motions can be calculated. However, since the encoders measure the rotation of the stepper motors and not the motion of the mobile robot itself, they cannot measure the slip between the wheels and the component. Furthermore, a relative measurement is performed, which means that the measurement errors add up over time. Accordingly, a measurement error of max. 127.53 mm can be justified.

The integration of optical components into the center of gravity or TCP of the mobile robot system proves to be useful during the tests. Even if the robot twists due to the reasons already described, the laser optic is always in the center of the system and only performs a translational movement.

Furthermore, it can be shown that the planned welding test can be carried out successfully.

In this paper, an approach to the use of (MRs for LMP is presented. For this purpose, restrictions are derived from LMP that an MR must fulfill. Based on these restrictions, a kinematic model is selected, and suitable actuators and sensors are determined. Here, an omnidirectional MR driven by stepper motors was evaluated as suitable. An internal encoder-based system and an external camera system were selected as possible sensors. For the integration of the laser optics, it was worked out that in the best case, it should be integrated into the TCP of the MR.

To validate the theoretical results from the first part of the paper, an experimental part was connected. For this purpose, a prototype of an MR for LMP was built from the parts and components identified in the first part. The accuracy of the MR itself is determined with a high accuracy reference measurement device. To validate the sensors, the measurement error between the sensors and the reference measuring device was determined. It should be noted that the targeted accuracy of the MR of ±2 mm could not be met. This can be explained, among other things, by the slip between the robot and the component surface. Since no control system is available at the present time, this error cannot be compensated. The position information required for control can be provided by the sensor system. During these tests, it was found that the external, camera-based measuring system has a maximum Euclidean error of 7.74 mm. In comparison to the internal measuring system via encoder, which achieved a maximum Euclidean error of 127.53 mm, this is significantly more accurate. However, it must be pointed out here that this is a comparatively expensive measuring system, which is limited in terms of measuring space.

In a second series of tests, a laser welding test was carried out with the MR. Two sheets of stainless steel with thicknesses of 0.5 and 2 mm were welded together. It was found that an adequate weld seam could be produced with the specified parameters.

In summary, the research questions of the paper can be answered and the potential of mobile robots for laser material processing can be shown by means of laser welding experiments. However, the accuracy of the mobile robot system and the two measurement systems is not yet sufficient for use as a kinematic system. However, further steps should be taken.

To further improve the accuracy of the MR, a position control algorithm was to be integrated into the system. Furthermore, the accuracy of internal position determination should be improved so that it can be used for the control. Further validation with regard to laser materials processing must be carried out in order to further demonstrate the potential.

In addition, further laser components can be integrated into the MR to open up further areas of application. One example is the deflection of laser radiation by means of a laser scanner. This would allow further degrees of freedom to be implemented and exploited compared to the fixed optics used.

Should it be possible to improve the accuracy and integrate further laser components, the mobile robot system could be used to cover various applications. One example is the laser cleaning of large-area components for weld seam preparation. Laser welding and cutting of large components, for example, in shipbuilding could also be possible. Laser structuring of large components such as aircraft wings also appears possible.

This study was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy—EXC-2023 Internet of Production—Grant No. 390621612. The authors acknowledge the financial support by the Federal Ministry of Education and Research of Germany in the framework of Research Campus Digital Photonic Production (Project No. 13N15423).

The authors have no conflicts to disclose.

Thomas Kaster: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Funding acquisition (supporting); Investigation (equal); Methodology (equal); Project administration (equal); Resources (equal); Software (supporting); Supervision (lead); Validation (equal); Visualization (equal); Writing – original draft (lead); Writing – review & editing (lead). Jan-Hendrik Rissom: Conceptualization (supporting); Data curation (equal); Methodology (equal); Software (lead); Validation (equal). Leon Gorissen: Data curation (equal); Visualization (equal); Writing – review & editing (equal). Philipp Walderich: Methodology (equal). Jan-Niklas Schneider: Conceptualization (equal); Resources (equal); Writing – review & editing (equal). Christian Hinke: Conceptualization (supporting); Funding acquisition (lead); Project administration (equal); Resources (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.

The source code for the mobile robot experiments and supplementary information are openly available in “An approach towards the application of mobile robots in laser materials processing - Supplementary Information” at https://s.fhg.de/kaster2023b, Ref. 35.