Vision-controlled jetting for composite systems and robots – Nature

The results of this work were created with our contactless manufacturing system, which allows for a high print throughput independent of the structure that is to be printed. Our method allows us to place voxels of materials in freeform. The support material can be melted and washed away easily to allow for the creation of functional channels, cavities and hollow structures. In the following section, we describe the Vision-Controlled Jetting method and the evaluation methods that we used on our printed structures, systems and robots.

Vision-controlled jetting

The examples presented in this work were all 3D printed using a multimaterial additive manufacturing platform that utilized a vision-controlled jetting technology (Fig. 1 and Supplementary Videos 2 and 3). The platform has a scanning system, jetting system and positioning system that can now employ suitable material technologies, produce accurate print results and scale up in terms of size and throughput. The platform is composed of six subsystems described in detail in the following:

1. 1.

   The positioning subsystem moves the build plate to a certain location according to the commands that are issued by the print control software at a set velocity and along a set path.

2. 2.

   Each of the inkjet units contains four print heads (three units are shown in Fig. 1a), drive electronics, material feeds and a pressure control system to jet a specific material onto the build plate.

3. 3.

   The UV curing unit uses UV LEDs to cure the materials that have been deposited onto the build plate.

4. 4.

   The scanner unit uses its laser profilometry system to generate a high-resolution topographical map of the build surface.

5. 5.

   The print control software steers the printing system’s processes by utilizing the scanner data to generate adapted print layers as needed.

6. 6.

   Postprocessing removes the support material from the completed prints (Extended Data Fig. 2).

Positioning subsystem

The positioning subsystem controls the location of the build plate relative to the rest of the printer. An axial motion system utilizes a linear motor to move the build plate under the print hardware along the x axis. The y and z axes are driven by brushless DC motors. The y axis is used to shift the location in which the structure is built in such a way that it is aligned relative to the print head array. This axis allows the control loop to compensate for variations in nozzle performance. The z axis ensures that the build surface stays within working distance of the print heads as the build progresses. All the axes are operated as servos, using encoders with a 1 μm resolution for position and velocity control. The print velocity is limited by the deposition frequency and the resolution.

Inkjet units

Each inkjet unit contains all the hardware and electronics that are required to print a single material. Each unit contains four print heads (Fujifilm Dimatix SG1024-L), which are placed in a staggered array to fully cover the build plate. This allows a complete layer to be deposited each time the build plate passes under the inkjet units. The print heads have a native resolution of 400 dots per inch (DPI), and they eject droplets with a volume of about 70 pl at 15 kHz. The drive electronics translate the requested layer data into firing pulses for the nozzle actuators while the build plate is scanned under the print heads.

UV curing unit

After the deposition of a layer, the build plate moves under the curing unit (UV LED lamp) to initiate the polymerization of the printed material. One lamp is present on each side of the inkjet units to allow for bidirectional printing. The lamps emit 405 nm light at 16 W cm⁻².

Scanner unit

The scanner unit uses custom laser triangulation profilometry. A laser line is projected onto the surface of the build plate as it is passed under the scanner. An imaging system reads the shape of the laser line from a 32 × 2,048 laser line image, and it computes a two-dimensional height map with 2,048 pixels at each sampling interval. Each camera in the imaging system captures 6,000 laser line images per second. Four cameras are needed to cover the full width of the build plate, and each camera captures 9,000 images per scan to cover the length of the print. The two-dimensional height maps obtained from the individual cameras are assembled into a full 3D height map of the build surface. In total, each height map is computed from 2.36 × 10⁹ pixels within 2.5 s. The height maps are geometrically calibrated to a fixed pixel resolution of 32 μm × 64 μm × 20 μm and are provided to the feedback control system.

Print control software

The print control software orchestrates the activity of each subsystem to execute successful prints. When a build job has been defined, the print control software processes the file in a voxel representation by rendering the geometries at the printer’s resolution^51,52. The voxelization step uses ray tracing to quickly render the input geometries.

To print a layer, a feedback algorithm generates the layer data based on the scan data from the previous layer and from the input geometry’s voxel representation^53,54. The feedback algorithm aims to maintain the printing plane at a fixed distance from the scanner by reacting to the height of each voxel in the scan data. If the print is higher than the desired level, the feedback control can reduce the amount of ink that is deposited or skip printing at that voxel in the next layer. If the print height is too low for a given voxel, the feedback control will determine which material is missing according to the currently measured height. It will then increase the amount of ink that is deposited up to the maximum capacity of the print head.

The generated layer command is sent to the drive electronics of the print head. The drive electronics deposit the materials into their desired positions while the motion control system moves the build plate underneath the print hardware. This process is repeated as the parts are built up layer-by-layer until the build has been completed.

Postprocessing the prints

Completed builds are encased in a support material, which must be removed before the parts are ready for use (Extended Data Fig. 2 and Supplementary Video 3). The entire build is first placed in a convection oven and heated to 65 °C, where it is left overnight for the bulk of the support material to melt and drain away. The parts are then removed and placed in a tank of cleaning solution and heated to 65 °C, where they are sonicated for 20 min. The parts are then rinsed with water and allowed to dry in air. The drainage holes in the printed parts for this study are sealed using cyanoacrylate.

Print velocity

The inkjet units cover the print bed along the printer’s y direction (Fig. 1a). The print bed moves back and forth underneath the inkjet units in the x direction at velocity v_x (equation (1)). The velocity in the x-direction is dependent on the jetting frequency f_jet of the print head’s nozzles and on the resolution in the x direction r_x (equation (1)). The minimum droplet size determines the resolution r_x (here, 32 μm) and the actuation speed of the print head’s piezo nozzles limits the jetting frequency f_jet. The jetting frequency is adapted for each material to ensure best print performance.

Due to the exothermic nature of the curing process, the printed part is cooled for a certain amount of time t_cooling after each layer has been deposited. Therefore, the printer takes the time t_layer to print a single layer of a certain length (in the x direction) l_x and width (in the y direction) l_y. The length l_x is determined by the size of the full print bed and the distance that spans across all the inkjet heads, the UV lamp and the scanner. Since the print heads cover the whole print bed in the y direction, the layer time t_layer does not depend on the y extension of the print bed (equation (2)).

The height of each layer h_layer can be adapted, and it is dependent on the total number of deposited droplets in each x location. The layer’s height is also dependent on the volume of the jetted droplet V_droplet and the resolution in the x direction r_x (here, 32 μm) and y direction r_y (here, 64 μm). The print head’s speed in the z direction v_z determines the overall print speed (equation (3)). In contrast to other printing methods, the speed for this type of inkjet deposition system does not depend on the printed object’s geometry in the y direction. The speed is, however, dependent on the resolution in the x direction r_x and the resolution in the z direction r_z (here, 20 μm), that is, it depends on the layer’s height h_layer (equation (3)).

Inserting equation (1) and equation (2) into equation (3) describes the relation of resolutions to print speed (equation (4)).

The user can adjust the print velocity v_z (here, 16 mm h⁻¹) by adjusting the jetted droplet’s volume. The droplet’s volume can be tuned by adjusting the fluid’s rheological characteristics or by changing the print head’s operating parameters (such as the piezo actuation waveform or jetting temperature). The total print duration is determined by the build job’s width in the x direction and height in the z direction. A slicer software arranges all parts to be printed in a single build job.

Part packing density on build plate

Many parts can be placed on a single build plate due to the high packing density of the print process (for example, hundreds of parts in Extended Data Fig. 2). In contrast, powder-based print processes pose thermal constraints that do not allow parts to be placed close to each other. While powder-based systems typically only pack about 15% to 20% (ref. ⁵⁵), VCJ, as a form of inkjet material deposition, can accommodate packing densities above 40%.

Print materials

Three materials were printed together to produce the final parts: soft, rigid and support. A thiol-ene elastomer was used to print soft flexible components (Fig. 2). A rigid formulation of thiol-ene was used as the load-bearing structure. A phase-change material (wax) was used as a support structure. The phase-change material is jetted in a molten state at an elevated temperature and hardens as it cools after deposition. The material melts upon reheating above 60 °C, allowing for easy removal (Extended Data Fig. 2 and Supplementary Video 3). Additionally, VCJ also supports the print of epoxies. Two epoxy formulations⁵⁶ have been developed: a tough epoxy (Extended Data Table 1c) and a chemically resistant epoxy (Extended Data Table 1d).

Multimaterial prints

Multimaterial fabrication depends on the chemistries in use. In general, multimaterial parts must consist of materials from the same polymer family to ensure adequate bonding when mixed or placed in direct contact with each other. Incompatible materials can refuse to bond, causing separation, or inhibit curing. If multimaterial parts with incompatible materials are needed, it is possible to separate the two material regions with a thin separator of support wax (single voxel) to ensure full cure. This separation benefits from the use of mechanical interlocking between the two material regions to prevent material separation after the support is removed (Extended Data Fig. 1g–i).

Testing standards and material characterization

We used standardized testing to evaluate the printable materials compared to the state-of-the-art materials. In the following, we describe the standards used in this work.

Modulus of resilience using ASTM 2632

We investigated the modulus of resilience of the materials directly from the printer according to ASTM 2632 (ref. ⁵⁷) with three samples per material. ASTM 2632 specifies the test parameters for impact resilience of solid rubber from the measurement of the vertical rebound of a dropped mass from 16 inches in height.

Assignment of a glass transition temperature T
_g by DMA using ASTM E1640-18

We conducted the DMA and assigned a glass transition temperature T_g according to ASTM E1640-18 (ref. ⁵⁸). The DMA was performed on soft thiol-ene (Fig. 2g, Extended Data Fig. 4) and compared with the two acrylates Tango Black Plus and Agilus 30 (Fig. 2g).

Viscoelastic behaviour

The viscoelastic behaviour of the materials was quantified by recording stress-strain cycles going from 0% to 140% displacement and back to 0% at a stain rate of about 0.53 s⁻¹. The hysteresis of the material that relates to its viscoelasticity can be inferred by the area enclosed by the stress-strain cycle. We tested three samples of soft thiol-ene and Tango Black Plus, and two samples of Agilus 30.

Outdoor weathering using ASTM G154 Cycle 1

ASTM G154 (ref. ⁵⁹) mimics outdoor weathering in addition to UV exposure. The test reproduces the weathering effects that occur when materials are exposed to sunlight and moisture (rain or dew) during real-world usage. Rather than just an exposure to humidity, this test causes water droplets to form on the parts’ surface, modelling dew formation.

The testing standard ASTM G154, Cycle 1 exposes all samples to 0.89 W (m² nm)⁻¹ UV irradiation at a wavelength of about 340 nm from a UVA-340 lamp. The exposure cycle consists of 8 h UV at (60 ± 3) °C Black Panel Temperature followed by 4 h Condensation at (50 ± 3) °C Black Panel Temperature. The test samples were removed and tested after 250 h, 500 h, 750 h and 1,000 h.

Material characterization

In contrast to processes that require a planarizer, the contactless VCJ process enables printing of chemistries that continue to cure after the discontinuation of irradiation. This includes thiol-ene and epoxy chemistries.

The soft thiol-ene material⁶⁰ has a Shore hardness of 32 A, tear resistance of 5.6 kN m⁻¹ and elongation at break of 200% (Extended Data Table 1a). In addition, the material’s exposure to the outdoors was simulated according to ASTM G154, Cycle 1. After the outdoor weathering, material tests were conducted following ASTM D638: Type IV, 50 mm min⁻¹ (Extended Data Fig. 3d–i). Another thiol-ene resin was used to print rigid components. The rigid thiol-ene has a tensile strength of 45 MPa, tensile modulus of 2.1 GPa and elongation at break of 15% (Extended Data Table 1b).

The thiol-ene step-growth polymerization utilized in this work consists of an ABAB system alternating between poly-thiols and poly-enes. This polymerization approach results in a highly regular polymer chain structure, which combined with the high molecular weight, achieved through careful formulation, results in a highly elastic polymer. The high elasticity of the polymer can be seen in the large change of storage modulus before and after the glass transition temperature T_g in the DMA (Extended Data Fig. 4).

The contactless VCJ process also permits the printing of further resin families, for example, 100% UV-cationic cured epoxy materials. Epoxies are particularly attractive for several reasons, including low shrinkage, high chemical resistance and excellent UV stability. The tough epoxy presents an ultimate breakdown strength of 53.8 MPa, an elastic modulus of 2.5 GPa, an elongation at break of 7.1%, a Shore hardness 78D, Izod impact strength of 33.8 J m⁻¹ and a heat deflection temperature at 0.45 MPa of 76 °C (Extended Data Table 1c). In addition, the outdoor stability of the epoxy was tested per ASTM G154, Cycle 1, followed by ASTM D638, Type IV, at 50 mm min⁻¹ (Extended Data Fig. 3d).

The chemically resistant epoxy has an ultimate tensile strength of 59.2 MPa, an elastic modulus of 2.7 GPa, an elongation at break of 2.5%, a Shore hardness 81D and a heat deflection temperature at 0.45 MPa of 130 °C (Extended Data Table 1d). This epoxy is also resistant to chemicals and solvents (Extended Data Fig. 3a,c).

The adhesion between cast soft and rigid thiol-ene was tested via lap shear ASTM D 3163-01 (Extended Data Fig. 7). A shear strength of (1.08 ± 0.10) MPa was determined for five tested samples.

Printed systems and robots

Robotic hand

The printed robotic hand resembles a human hand with bones whose shapes have been extracted from open-source magnetic resonance imaging data³⁷. The joints connecting the bones are modelled to resemble the human anatomy. The printed tendons are attached to the bones in locations approximating the anatomically correct insertion areas of the muscles. Rigid guides are modelled as extrusion from the bone to guide the tendons to ensure the forces are delivered to the attachment point. Each printed tendon is connected to a servo motor (DYNAMIXEL XL430-W250-T, ROBOTIS Co. Ltd.). One end of multifilament fishing line is knotted to the end of the printed tendon and the other end of the fishing line is spooled onto a reel of the servo motor.

Each fingertip and the palm of the hand are fitted with a sensor pad that measures pressure. This printed sensor pad is a cavity with a thin membrane that is connected through a long, printed tubing. Each printed tubing coming from the sensor pad is externally connected to a commercial pressure sensor (015PG2A3, Honeywell International Inc.) with a sensor range of 0 kPa to 25 kPa. The sensor signal is read out by a microcontroller (Arduino DUE, Arduino S.r.l.).

The hand’s controller runs on a computer. The motor’s actuation patterns and control sequences are written in Python, and the sensor signal from the microcontroller is read out via a serial connection. The control loop for the hand allows the closing of the individual fingers until contact is sensed through the printed sensor pads.

The hand was evaluated by testing its compliance, dexterity and ability to grasp objects. The fingers’ compliance was tested through the manual bending of the joints and hitting the hand with a hammer. The dexterity of the hand was evaluated by controlling the tendon-actuation to make contact between the tip of the thumb and another fingertip of the same hand. The object grasping tests were executed according to a multistep grasping algorithm (Extended Data Fig. 5). Several objects were placed in front of the hand. The closure of the hand was started as soon as contact was sensed at the palm sensor. The fingers then closed until their fingertips sensed contact with the object to be grasped.

Walking robot

The printed walking robot prototype is an eight-channel system with two sets of two channels for actuating groups of three legs (Extended Data Fig. 6a). One channel supplies the top joints and one the bottom joints of the group of three legs. Applying pressure to these channels bends the legs at the respective joint. Pressure patterns symmetric to the centre plane of the robot allow the robot to locomote in a forward and backward direction. The pressure patterns are adapted to provide more pressure to one half than the other, leading to the robot turning left or right. Another set of two channels is used to actuate the robot’s arm. One actuator is located at the joint intersection with the body. The other actuates the ‘forearm’. Finally, two channels connect to a gripper. One channel supplies the gripper with actuation pressure, the other connects the sensing pad to a pressure sensor. The sensing pad is a cavity at the fingertips of the gripper. Reading pressures at these channels allows us to reason about the forces and thereby the contact made between the tip of the gripper and the contacting object.

We connect the supply channels of the robot to seven channels of a 16-channel proportional valve terminal (MPA-FB-VI, Festo Vertrieb GmbH & Co. KG). The valve terminal has individually addressable channels that command pressures between 0 kPa to 250 kPa at a flow rate per channel of up to 380 l min⁻¹. The sensing channel is connected to a pressure sensor (015PG2A3, Honeywell International Inc.) with a sensor range of 0 kPa to 25 kPa. A microcontroller (Arduino DUE, Arduino S.r.l.) receives the sensor signal and streams the measurements to the serial port. The pressure patterns and control sequences are written in Python, and the sensor signal from the microcontroller is read from its serial port. We demonstrate the walking robot’s ability to locomote, grasp and sense using different objects. Experimental still images (Extended Data Fig. 6b) and video recordings (Supplementary Video 5) are available.

Heart pump

The functional heart pump is a multimaterial print that operates as two pressurized-air-driven liquid pumps (Supplementary Video 6) resembling the double ventricle of a mammalian heart. Two openings are located at the bottom of the heart to allow pressurized air to compress the membranes of each artificial ventricle. This compression corresponds to a heart muscle shrinking the volume of a ventricle. The ventricle’s volume is connected to a liquid supply system through a one-way inlet valve and a one-way outlet valve. These valves resemble the three-leaved heart valves that can be found in the aortic valve, the tricuspid valve and the pulmonary valve. The outer shell of the heart approximates a mammalian heart. Each ventricular chamber is fitted with a printed sensor pad that allows the sensing of the heart’s frequency. The sensor pad connects to a sensor channel in the heart. The channel is connected to a pressure sensor (015PG2A3, Honeywell International Inc.) with a sensor range of 0 kPa to 25 kPa. The sensor signal is decoded on a microcontroller (Arduino DUE, Arduino S.r.l.). A reciprocating syringe pump system is used to actuate the printed pump.

To test the flow rate of the heart and the functionality of the sensor, an experimental setup like the circulatory system found in mammals was used (Supplementary Video 6). Three 10 l translucent buckets were connected to the heart. The left bucket resembled de-oxygenated, old blood, the bucket in the middle resembled the lung’s blood volume and the right bucket was for oxygenated blood leaving the heart. To measure the flow rate, we recorded the change in weight of the buckets over time. The sensed frequency in the sensor pads was compared to the frequency of actuation of the syringe pump.

Multimaterial metamaterial structure

Going beyond the limited properties of a single material in bulk, metamaterials can be freeform constructed from multiple materials to provide features not found in a homogeneous material block. We can adjust by design the stress-strain curve of a material using a truss-based configuration. The links of the truss are made of soft materials and the nodes of the truss are additionally reinforced with rigid, spherical elements. This configuration allows for more distinct changes in material stiffness beyond a given level of strain.

We printed metamaterials from soft and rigid thiol-ene with different link and node diameters and tested the resulting cubes of the metamaterials using a compression testing machine (Instron 5943, Illinois Tool Works Inc.) and a high-speed camera (FASTCAM Mini AX200, Photron). Each metamaterial construct was placed in the testing area of the compression testing machine and was compressed from 0 mm to 18.2 mm in relative displacement.