Keywords

1 Introduction and Related Work

Fiber reinforced plastics (FRP) are, due to their good weight specific mechanical properties used in high performance applications with minimum weight requirements [1, 2]. However, the production of FRP parts, is labor intensive with little automated processes in place [3]. Especially low bending stiffness of woven fiber mats, as well as their fragile nature lead to challenges in robotic handling.

For efficient processes, it would be preferable to start a production process from a stack of raw fabric plies, as this allows the cutting and preparation in bulk [4]. When using such a stack, the first and foremost task is to separate single sheets of fabric material from this stack for further handling and manipulation. While processes such as pinch gripping or freeze gripping have been shown to be capable of singulating single plies from a stack, they can lead to undesirable alterations on the gripped ply or the remaining stack underneath [5].

In contrast, low pressure suction grippers are capable of handling even fragile fabrics without damage [6]. However, the porous nature of most textile materials makes stack singulation very difficult due to the second or third ply still being effected by the grippers’ vacuum suction [7].

Cubric et al. have successfully used vacuum grippers for grasping textile materials, however, they concluded: ‘It has also been found that the application of this vacuum gripper is not suitable for taking one layer of fabric from material bundle’ [8].

In further research, it was shown that parameters such as area mass density, gripper position, suction cup geometry, and supply pressure have a major influence on the successful handling of non-woven textiles. However, no attempt has been made to quantify or model any of these influences [9].

The author in [6] has successfully used a vacuum suction gripper for separating single plies of woven carbon fiber mats from a stack, by controlling the electrical contact resistance of the carbon fiber pressing against the suction cup. The main deficit of this approach is its dependence on the conductivity of the handled materials.

In this paper, a robotic gripping system capable of reliable single ply separation for woven technical textiles is presented, talking the aforementioned deficits. To achieve this task, the gripping system is split up into relevant subsystems and their influences on the gripping process are modeled, allowing the generation of a model capable of predicting the suction pressure inside the gripper depending on chosen process parameters. This model in turn enables:

  • Measuring the number of plies adhering to the gripper at any given time

  • Determination of a suction pressure at which the gripper will be most likely to grip a single ply from the stack

2 Gripping System Overview

The main end effector used in this research consists of four Schmalz SCG 1xE100 low pressure suction grippers. Every individual gripper element (see Fig. 1) consists of:

  • A vacuum generating Coandă ejector.

    High pressure air supplied to the pressure inlet (1) is accelerated through a small slit. The high velocity airstream adheres to the outlet walls of the ejector curving away from the suction chamber. This high speed airstream leads to a low pressure zone inside the chamber (4).

  • A screw-on perforated plate (2) that encloses the suction chamber and ultimately acts as the interface to any gripped fabric.

  • A vacuum pressure sensor monitoring the suction pressure \({p}_{U}\), through a bypass opening (3) directly into the suction chamber.

3 Subsystem Modelling

Since the pressure in the suction chamber is the only process parameter that can be measured in this gripper setup, an attempt is made to model the suction pressure as a function of other effective process parameters, as shown by the Ishikawa diagram in Fig. 2.

Fig. 1
figure 1

Left Stylized 3D Cut through an SCG Gripper, Right Gripper cross section

Full size image
Fig. 2
figure 2

Ishikawa diagram of major influences on the Suction pressure inside the gripper

Full size image

For a detailed analysis of the gripping related subsystems, the system is divided into 4 subsystems effecting the generated suction pressure: Vacuum generation, Perforated plate, Material Properties and Load-case dependent leakage.

3.1 Vacuum Generation

The vacuum generated by the gripper is mainly dependent on the input pressure supplied to it. This input pressure is controllable from 0 to 4 bar by adjusting a voltage-controlled pressure control valve at the input port. Another major influence on the vacuum generated by the Coandă ejector is the flow restrictions to the incoming suction airstream.

In this section the generation of a predictive model is discussed, which allows the determination of the suction pressure inside the gripper \({\mathrm{p}}_{\mathrm{U}}\) at any given time dependent on the input pressure \({\mathrm{p}}_{\mathrm{N}}\) and the volumetric flowrate of the suction airstream \({\mathrm{Q}}_{\mathrm{V}}\).

To determine this vacuum generation characteristic, the volumetric flow rate of air \({Q}_{V}\) entering the gripper is recorded, as well as the suction pressure \({\mathrm{p}}_{\mathrm{U}}\), while varying the supply pressure and placing different flow resistances in the suction path (see Fig. 3 left). To ensure a smooth almost laminar airflow at the location of measurement for volumetric flowrate, a 30 cm long pipe is added to the bottom of the gripper-restriction assembly.

This generates a three dimensional dataset of 200 specific operating points. The linear regression model fitted to this data set yields a characteristic equation (see Fig. 3 right):

$${p}_{U}\left({Q}_{V},{p}_{N}\right)={p}_{N}\cdot {a}_{10} -{Q}_{V}\cdot {a}_{01}-{p}_{N}^{2}\cdot {a}_{20}- {p}_{N}\cdot {Q}_{V}\cdot {a}_{11}-{Q}_{V}^{2}\cdot {a}_{02}$$

With \({a}_{10}=4279\frac{Pa}{bar};{a}_{01}=11.28\frac{Pa min}{l};{a}_{20}=286.6 \frac{Pa}{ba{r}^{2}};\)

$${a}_{11}=0.3824\frac{Pa min}{bar l}; {a}_{02}=0.01701\frac{Pa mi{n}^{2}}{{l}^{2}}$$

With an adjusted coefficient of determination of: \({\mathrm{R}}_{\mathrm{adj}}^{2}=0.9299\)

Fig. 3
figure 3

Left Cross-section through the experimental setup for the measurement of vacuum generation characteristics, Right Interpolated vacuum generation characteristic model

Full size image

3.2 Perforated Plates

The suction chamber of these grippers is enclosed by a perforated plate, which also acts as an interface to the gripped textile. The design of this interface however is not predetermined, allowing for adjustments specific to any use case. To determine the influence of these suction plates, the experiment described above is repeated with 9 different suction plate designs shown in Fig. 4. These suction plate designs vary in the size of individual orifices \({D}_{H}\) as well as the combined cross sectional area of the orifices.

By 3D Printing these test specimen and measuring the volumetric airflow through them at varying suction pressures a characteristic resistance graph for every suction plate can be determined, as shown in Fig. 4 right.

The literature on airflows through perforated plates [10, 11] states that the pressure loss of lamina flow through such a plate is supposed to be:

$${p}_{Ambient}-{p}_{S}=\frac{1}{2}Eu \rho {V}^{2}$$

with

\({p}_{S}\):

Vacuum pressure in the suction chamber

Eu:

Euler number/ pressure loss coefficient

\(\rho \):

Fluid density

V:

pipe bulk-mean velocity

Fig. 4
figure 4

Left Perforated plate geometries used in this study, Right Accompanying characteristics. All subplots are equally scaled \({p}_{U}=0..10000 \text{Pa}\) and \({Q}_{V}=0..600 \text{l}/\text{min}\)

Full size image

With \({Q}_{V}^{2}\) being proportional to \({V}^{2}\) and \({p}_{U}\) as the left hand pressure differential the equation can be restated with a complex pressure loss coefficient \(\xi \) under the assumption of constant fluid density (which nearly holds apart from changes in environmental conditions) as follows:

$${p}_{U}=\xi {Q}_{V}^{2}$$

While an increase in plate porosity clearly coincides with an increase in airflow and consequentially reduction in suction pressure, the influence of the hole diameters is not monotonous.

3.3 Air-Permeable Textiles

An airstream through a porous media, such as a woven fabric, leads to a pressure drop, which can be described by the following expression [12]:

$$\Delta p={C}_{1}\frac{{Q}_{V}}{A} +{C}_{2}{\left(\frac{{Q}_{V}}{A}\right)}^{2}$$

With

\(\Delta p\):

the pressure loss across the media

\({C}_{1}\):

linear loss coefficient

\({C}_{2}\):

quadratic loss coefficient

With low flow rates \({Q}_{V}\) the quadratic term can be dropped, leading to the equation used in the norm DIN EN ISO 9237 [13].

For any material used the air permeability \(R\) can easily be determined as described by the norm [13] by measuring it at a certified test stand. This value R is the area normalized airflow through a textile at a 200 Pa pressure differential. Therefore, the Value of \({C}_{1}\) can easily be determined and substituted into the pressure differential equation above: \({C}_{1}=\frac{200 Pa}{R}\Delta p=\frac{200 Pa}{R}\cdot \frac{{Q}_{V}}{{A}_{F}}\)

Combining this equation with the pressure drop across the suction-plate yields:

$$\Delta p=\frac{200Pa}{R}\cdot \frac{{Q}_{V}}{{A}_{F}}+\xi {Q}_{V}^{2}$$

This characteristic can be intersected with the vacuum generation model shown in Sect. 3.1. This approach however achieves only moderate success in correlating to real measured data as shown in Fig. 5.

Fig. 5
figure 5

Left Suction pressure over supply pressure for 9 different suction plates. Orange line calculated model, Blue line: smoothed measurement data. Right Cantilever test for the determination of simulation parameters and simulated deformations in good agreement (local Von-Mises-Stress shown in false color mapping)

Full size image

While this theoretically derived model shows some general correlation to the overall characteristics of the gripper, one cannot assume perfect prediction of any interaction and therefore the perfect prediction of suction pressure inside the gripper is not achievable without the use case specific experimental determination of gripper characteristics.

3.4 Leakage Currents

As shown in Fig. 1 (right) deformations in the fabric can lead to air streams not completely passing through the fabric. These stray air currents will not apply holding forces on the textile and therefore play a huge role in grip security and the separation of textiles from the gripper at low supply pressures. This subject is addressed by introducing a scalar load equivalent term which will be obtained by a simple static mechanical simulation.

For the generation of such a scalar load equivalent value 6 limp materials are selected and their mechanical properties needed for simulation are determined, as shown in Table 1:

Table 1 Material properties and method of acquisition
Full size table

Figure 5 (right) shows the good match of simulated cantilever test with the real one for one example material.

To quantify the deformations of a single textile sheet with a single scalar value, a circular path around the gripper is defined where the deformation of the simulated textile is measured around the circumference of said path \(d(s)\) with the normalized path length \(s\). This in turn allows the calculation of the load equivalent term L.

$$L={\int }_{0}^{1}d{\left(s\right)}^{2}ds$$

.

While these deformations have little effect on the suction pressure inside the gripper when the textile is securely gripped, they show big influences on the separation of layers from the gripper. To verify this load equivalent 8 different load cases were defined with varying material geometries and gripper configurations. Five of these variations are shown in Fig. 6 (right). After grasping the material the operating pressure is reduced until the ply drops off. Recording the supply pressure at which separation occurs \({s}_{10}\) a correlation between \({s}_{10}\) and the load-equivalent-term \(L\) can clearly be seen in Fig. 6 (left).

Fig. 6
figure 6

Left Correlation of Load-Equivalent L and separation-pressure \({s}_{10}\), Right Five of our considered example load cases

Full size image

While the specific linearity factor is dependent on the selected suction-plate and material combination this linearity factor can be determined experimentally and can later be used to determine separation-points for a minimal pressure at which a single ply is gripped successfully, thus being a good starting-point for a minimal supply pressure for stack singulation.

4 Robotic Workstation

As mentioned above, the goal is to proof the concept of robotic single ply separation from a given textile stack. Therefore, a robotic test stand is built made up of a UR5e cooperative robot and a gripper assembly, containing 4 low pressure suction grippers each equipped with a −1 to 1 bar differential pressure sensor, which can be seen in Fig. 7 Control structure for the robotic workstation left.

Fig. 7
figure 7

Control structure for the robotic workstation

Full size image

4.1 Control Structure

To control the test-stand, an OPC-UA server running on a Beckhoff PLC is used as well as a control backbone of the singulation routine on a connected personal computer running a Matlab control code. The control architecture can be seen in Fig. 7.

Before programming the routine, the experiments described in Sect. 3 are conducted to generate:

  1. 1.

    A model on the suction-pressure expected when gripping a single layer, which is used for the determination of the number of gripped materials.

  2. 2.

    A material model useable in simulation.

  3. 3.

    A model on the minimum supply pressure where a single sheet just so adheres to the gripper based on the simulation derived load equivalent at this gripper

These experimental results can be used in the following control strategy for stack separation:

  1. 1.

    Move to a location just above the stack.

  2. 2.

    Move down until contacting the stack.

  3. 3.

    Activate every gripper with a supply pressure just above the minimum supply pressure for single layer handling. Obtained from analysis described in Sect. 3.4

  4. 4.

    Slowly move upwards clear of the stack.

  5. 5.

    Increase the supply pressure in every gripper to a detection pressure (e.g. 3 bar)

  6. 6.

    Analyze resulting suction pressure and compare it to the expected value for single layer handling.

  7. 7.

    If the values of step 6 differ, either discard all gripped sheets or replace them atop the stack

  8. 8.

    Else, lower the supply pressure to a safe handling pressure (e.g. 1 bar) and move the single gripped sheet to its destination.

4.2 Experimental Validation

Performing the control process described above it is possible to repeat the stack separation process indefinitely by moving a stack to a new location and back again.

In a test with a sample size of 500, it was possible to successfully detect and separate a single material layer from the stack 99.6% of the time. Both errors were due to the material sticking to the stack when lifting the gripper at step 4. This error is recognized at step 6 and could have been corrected in a production process.

The same series of tests for separating two layers of material together was successful in 66% of the cases. Problems observed when going for multiple layer handling were mainly:

  • The gripper grasping three materials instead of two without realizing this mistake at step six, due to large variations in the suction-pressure with more than one material adhering to the gripper.

  • The second grasped layer dropping off of the gripper at the initial lifting procedure at step 4 (this error was successfully)

5 Conclusion and Outlook

In this work, attempts at modeling the governing interactions in a low pressure suction gripper when handling porous materials are presented.

While it was not possible to generate a fully parametrical model for predicting the pressure inside a suction gripper at any given time, the approach nonetheless allows to derive a process for automated stack separation with an overwhelming success-rate of >99%.

This level of confidence would even allow for usage of this process in real production environments. Multiple challenges came up when going for multi-layer handling operations, while these might not be as important for production facilities as single ply separation, the furtherance of the understanding on those processes remains an interesting field of research.

Incorporating and fusing other sensor principles, such as immediate force measurement at the Gripper-Textile Interface as well as optical feedback could further improve the success rate and error tolerance for such delicate handling processes.

Furthermore incorporating artificial intelligence into the control structure could improve results and might allow for dynamic reaction on slightly alternated environmental parameters.