NTU Singapore: Robotic Post-Processing System Removes Residual Powder from 3D Printed Parts

Researchers from Nanyang Technological University in Singapore wrote a paper, titled “Development of a Robotic System for Automated Decaking of 3D-Printed Parts,” about their work attempting to circumvent a significant bottleneck in 3D print post-processing. In powder bed AM processes, like HP’s Multi Jet Fusion (MJF), decaking consists of removing residual powder that sticks to the part once removed. This is mostly completed by human operators using brushes, and for AM technologies that can produce hundreds of parts in one batch, this obviously takes a long time. Manual labor like this is a significant cost component of powder bed fusion processes.

An operator manually removing powder (decaking) from a 3D printed part.

“Combining Deep Learning for 3D perception, smart mechanical design, motion planning, and force control for industrial robots, we developed a system that can automatically decake parts in a fast and efficient way. Through a series of decaking experiments performed on parts printed by a Multi Jet Fusion printer, we demonstrated the feasibility of robotic decaking for 3D-printing-based mass manufacturing,” the researchers wrote.

A classic robotic problem is bin-picking, which entails selecting and removing a part from a container. The NTU researchers determined that 3D perception, which “recognizes objects and determining their 3D poses in a working space,” would be important in building their bin-picking system. They also used a position-controlled manipulator as the baseline system to ensure compliant motion control.

The NTU team’s robotic system performs five general steps, starting with the bin-picking task, where a suction cup picks a caked part from the origin container. The underside is cleaned by rubbing it on a brush, then flipped over, and the other side is cleaned. The final step is placing the cleaned part into the destination container.

Proposed robotic system design for automated decaking.

Each step has its own difficulties; for instance, caked parts overlap and are hard to detect, as they’re mostly the same color as the powder, and the residual powder and the parts have different physical properties, which makes it hard to manipulate parts with a position-controlled industrial robot.

“We address these challenges by leveraging respectively (i) recent advances in Deep Learning for 2D/3D vision; and (ii) smart mechanical design and force control,” the team explained.

The next three steps – cleaning the part, flipping it, and cleaning the other side – are tricky due to “the control of the contacts” between the parts, the robot, and the brushing system. For this, the researchers used force control to “perform compliant actions.”

Their robotic platform made with off-the-shelf components:

  • 1 Denso VS060: Six-axis industrial manipulator
  • 1 ATI Gamma Force-Torque (F/T) sensor
  • 1 Ensenso 3D camera N35-802-16-BL
  • 1 suction system powered by a Karcher NT 70/2 vacuum machine
  • 1 cleaning station
  • 1 flipping station

The camera helps avoid collisions with the environment, objects, and the robot arm, and “to maximize the view angles.” A suction cup system was found to be most versatile, and they custom-designed it to generate high air flow rate and vacuum in order to recover recyclable powder, achieve sufficient force for lifting, and firmly hold the parts during brushing.

Cleaning station, comprised of a fan, a brush rack, and a vacuum outlet.

They chose a passive flipping station (no actuator required) to change part orientation. The part is dropped down from the top of the station, and moves along the guiding sliders. It’s flipped once it reaches the bottom, and is then ready to be picked by the robot arm.

Flipping station.

A state machine and a series of modules make up the software system. The machine chooses the right module to execute at the right time, and also picks the “most feasible part” for decaking in the sequence.

The software system’s state machine and modules perform perception and different types of action.

“The state machine has access to all essential information of the system, including types, poses, geometries and cleanliness, etc. of all objects detected in the scene. Each module can query this information to realize its behavior. As a result, this design is general and can be adapted to many more types of 3D-printed parts,” the researchers explained.

The modules have different tasks, like perception, which identifies and localizes visible objects. The first stage of this task uses a deep learning network to complete instance detection and segmentation, while the second uses a segmentation mask to extract each object’s 3D points and “estimate the object pose.”

Example of the object detection module based on Mask R-CNN. The estimated bounding boxes and part segmentations are depicted in different colors and labelled with the identification proposal and confidence. We reject detection with confidence lower than 95%.

“First, a deep neural network based on Mask R-CNN classifies the objects in the RGB image and performs instance segmentation, which provides pixel-wise object classification,” the researchers wrote.

Transfer learning was applied to the pre-trained model, so the network could classify a new class of object in the bin with a high detection rate.

“Second, pose estimation of the parts is done by estimating the bounding boxes and computing the centroids of the segmented pointclouds. The pointcloud of each object is refined (i.e. statistical outlier removal, normal smoothing, etc.) and used to verify if the object can be picked by suction (i.e. exposed surfaces must be larger than suction cup area).”

Picking and cleaning modules are made of multiple motion primitives, the first of which is picking, or suction-down. The robot picks parts with nearly flat, exposed surfaces by moving the suction cup over the part, and compliant force control tells it when to stop downward motion. It checks if the height the suction cup was stopped at matches the expected height, and then lifts the cup, while the system “constantly checks the force torque sensor” to make sure there isn’t a collision.

Cleaning motion primitives remove residual debris and powder from nearly flat 3D printed parts. The part is positioned over the brush rack, and compliant force control moves the robot until they make contact. In order to maintain contact between the part and the brushes, a hybrid position/force control scheme is used.

“The cleaning trajectories are planned following two patterns: spiral and rectircle,” the researchers explained. “While the spiral motion is well-suited for cleanning nearly flat surfaces, the rectircle motion aids with removing powder in concave areas.”

A combination of spiral and rectircle paths is used for cleaning motions. Spiral paths are in red. The yellow dot denotes the centroid of the parts at beginning of motion. Spiral paths are modified so they continue to circle the dot after reaching a maximum radius. The rectircle path is in blue, parameters include width, height, and direction in XY plan.

The team tested their system out using ten 3D printed shoe insoles. Its cleaning quality was evaluated by weighing the parts before and after cleaning, and the researchers reported the run time of the system in a realistic setting, compared to skilled human operators.

In terms of cleaning quality, the robotic system’s performance was nearly two times less, which “raised questions how task efficiency could be further improved.” Humans spent over 95% execution time on brushing, while the system performed brushing actions only 40% of execution time; this is due to a person’s “superior skills in performing sensing and dexterous manipulations.” But the cleaning quality was reduced when the brushing time was limited to 20 seconds, which could mean that the quality would improve by upgrading the cleaning station and “prolonging the brushing duration.”

Additionally, humans had more consistent results, as they are able to adjust their motions as needed. The researchers believe that adding a cleanliness evaluation module, complete with a second 3D camera, to their system would improve this.

Average time-line representation of actions used for cleaning.

“We noted that our robot ran at 50% max speed and all motions were planned online. Hence, the sytem performance could be further enhanced by optimizing these modules,” the team wrote. “Moreover, our perception module was running on a CPU, implementations of better computing hardware would thus improve the perception speed.”

While these results are mainly positive, the researchers plan to further validate the system by improving its end-effector design, optimizing task efficiency, and adapting it to work with more general 3D printed parts.

Discuss this and other 3D printing topics at 3DPrintBoard.com or share your thoughts below. 

The post NTU Singapore: Robotic Post-Processing System Removes Residual Powder from 3D Printed Parts appeared first on 3DPrint.com | The Voice of 3D Printing / Additive Manufacturing.

ABB Robotics Adds 3D Printing to RobotStudio Software

Industrial robotic arms are no strangers to 3D printing. Their great dexterity and flexibility have made them increasingly powerful platforms for a variety of 3D printing processes. Swiss tech giant ABB has caught on and has added 3D printing capabilities to the latest edition of its RobotStudio simulation and offline programming software.

Numerous customers have already begun using ABB robots for 3D printing purposes. One of the most notable may be MX3D, which outfitted its industrial robotic arm with a wire arc welding system to 3D print large-scale metal objects, including a bridge in Amsterdam. Viridis3D also uses a standard ABB robot to perform a unique take on 3D printing sand for metal casting.  You’ll also come across countless researchers taking advantage of these robots for new 3D printing applications.

Clearly catching on to the growing adoption of its tools in the additive manufacturing space, ABB is hoping to make it easier for users to do so. Part of the PowerPac portfolio, RobotStudio removes the need for manual programming for 3D printing. According to the company, the software’s new 3D printing feature will allow users to program ABB robots for AM in just 30 minutes. This includes such 3D printing processes as welding, concrete or printing with polymer granules.

ABB highlights the fact that any 3D printing slicer software can be converted into ABB’s simulation environment and robot code. The company suggests that this process is faster than plotting a toolpath for a traditional printing system.

Established over 130 years ago, ABB has roughly 147,000 employees spanning more than 100 countries. Despite its size and legacy, it is not the only manufacturer competing in the 3D printing space. Other companies who have seen their robotics featured in AM processes include Yaskawa, Comau, KUKA, Universal Robots, FANUC and Schunk.

Of these, FANUC is the only one to sell a system specifically for 3D printing, a collaborative wire arc welding robot. Comau and KUKA are comparable to ABB in terms of the way that their robots have been used for AM, while Universal Robots sells less expensive, less industrial machines that have been used by the likes of Voodoo Manufacturing.

The drive for the use of these machines is not just the fact that many of them can be used to fabricate large-scale structures, but also that they would be more easily integrated into an already existing manufacturing environment, where industrial robots are already the norm. Therefore, as AM becomes more widely adopted in a factory setting, robotics companies have a greater incentive to find ways to work with AM technology.

In other words, many makers of industrial robotic arms have begun to catch on to the growing adoption of their products for 3D printing. As AM grows, this number will surely increase, further motivating companies like ABB to accommodate their new customer base. It’ll be interesting to see what the next move by a large robotics company will look like.

The post ABB Robotics Adds 3D Printing to RobotStudio Software appeared first on 3DPrint.com | The Voice of 3D Printing / Additive Manufacturing.

Siemens and CEAD Develop Hybrid 3D Printing Robotic Arm

3D printing with continuous reinforcement fibers, like carbon fiber, is just now starting to come into its own, with numerous startups developing their own unique approaches to the concept. Their presence on the market is a different story, as many of these firms are still in the R&D or beta stages. Dutch firm CEAD, however, is already in the process of building and shipping its massive reinforcement fiber 3D printers and has already developed, in partnership with Siemens, a follow-up system that it will be showcasing at Formnext 2019: the AM Flexbot.

You may have caught our previous coverage of CEAD and its Continuous Fiber Additive Manufacturing (CFAM) process. The company’s first printer, the CFAM Prime, has a massive build envelope of 4mx2mx1.5m and is, according to CEAD, able to print thermoplastics reinforced with continuous glass or carbon fiber at a rapid average rate of 15 kg/hr. To maintain precision while depositing at such speeds, the system is controlled by Siemens’ Sinumerik 840D sl.

The firm’s new technology, the AM Flexbot, extends CEAD’s partnership with Siemens, resulting in a 3D printing process that uses a large-scale industrial robotic arm as a motion platform. The system features CEAD’s single screw extruder unit mounted onto a Comau robotic arm, all controlled by Siemens’ Sinumerik CNC with Run MyRobot /Direct Control software. For clarification: the AM Flexbot does not perform continuous fiber reinforcement.

CEAD parts for a small pedestrian bridge the company is making.

CEAD turned to Comau and Sinumerik Run MyRobot/Direct Control in order to maintain the precision necessary to accurately deposit the material to near-net-shape before milling parts to completion. The AM Flexbot is now available for purchase, but CEAD is aiming to expand its feature set. If the system can both 3D print and mill, it’s possible to imagine what additional capabilities it might have. This could include performing quality control metrology once a print has been completed, using robotic grippers to add external components, or working in tandem with other robotic arms to create even larger structures.

CEAD co-founder and Operations Director Lucas Janssen said of the product, “By using Siemens’ Sinumerik Run MyRobot /Direct Control together with a Comau robot arm in our latest solution, we are able to deliver a modular system scalable to fit our customer’s needs as many different functions can be added at any time. We are very pleased to work with Siemens and their reliable products.”

CEAD is not alone in the fiber reinforcement or robotic arm space. In terms of the former, companies like AREVO, Impossible Objects, Markforged, Desktop Metal, Anisoprint, and Continuous Composites are just a few who are either working on fiber reinforcement 3D printing or have commercial products readily available. For robotic arms, 3D Systems, MX3D, Stratasys, and EnvisionTEC are only the commercial companies who are either working on 3D printing with robotic arms or have solutions commercially available. In both spaces, there are countless research projects being performed, so CEAD will find itself with plenty of company in the near future.

The AM Flexbot will be on display at the Siemens booth (D81, Hall 12.1) at Formnext 2019, but if you don’t catch it there, Siemens will also be including a CEAD 3D printing system at its Additive Manufacturing Experience Center (AMEC) in Erlangen, Germany.

The post Siemens and CEAD Develop Hybrid 3D Printing Robotic Arm appeared first on 3DPrint.com | The Voice of 3D Printing / Additive Manufacturing.

Virginia Tech inventor granted patent for 3D printing vending machine

Kevin D. Kline, an inventor at Virginia Tech, has been granted a patent for a “3D printing vending machine” which can be used for continuous manufacturing. According to the patent document “In one embodiment, the present invention relates to assemblies that automatically remove printed components to permit the continuous and automatic manufacturing of additional components.” […]

Researchers Decrease Support Structures for Models Through Multidirectional 3D Printing

An illustration for the idea of the algorithm: (a) a progressively determined planar clipping results for generating the optimized base planes, and (b) the inverse order of clipping planes results in a sequence of regions to be fabricated where the printing direction of each region is the normal of its base plane. The orientation of a printing head is fixed during the procedure of physical fabrication. The parts under fabrication are reoriented to realize the multidirectional 3D printing.

In most planar-layer based 3D printing systems, material collapse is prevented on large overhangs by adding support structures to the bottom. But support structures in single-material 3D printing methods have some major issues, like material waste and the possibility of surface damage. This can be helped by introducing rotation and turning the hardware into a multidirectional system, where models are subdivided into separate regions and each one is 3D printed along a different direction.

L-R: Snowman models fabricated by an FDM 3D printer and the team’s multidirectional 3D printing system by adding only one rotational axis on the same 3D printer.

A team of researchers from Tsinghua University, TU Delft, and the Chinese University of Hong Kong developed two types of multidirectional 3D printing hardware systems: one modified from an off-the-shelf FDM 3D printer with an added rotational degree-of-freedom (DOF), and the other implemented on an industrial robotic arm to simulate a tilting table for two rotational DOFs. They outlined their work in a paper titled “General Support-Effective Decomposition for Multi-Directional 3D Printing.”

The abstract reads, “We present a method to fabricate general models by multi-directional 3D printing systems, in which different regions of a model are printed along different directions. The core of our method is a support-effective volume decomposition algorithm that targets on minimizing the usage of support-structures for the regions with large overhang. Optimal volume decomposition represented by a sequence of clipping planes is determined by a beam-guided searching algorithm according to manufacturing constraints. Different from existing approaches that need manually assemble 3D printed components into a final model, regions decomposed by our algorithm can be automatically fabricated in a collision-free way on a multi-directional 3D printing system. Our approach is general and can be applied to models with loops and handles. For those models that cannot completely eliminate support for large overhang, an algorithm is developed to generate special supporting structures for multi-directional 3D printing. We developed two different hardware systems to physically verify the effectiveness of our method: a Cartesian-motion based system and an angular-motion based system. A variety of 3D models have been successfully fabricated on these systems.”

The researchers wanted to create a 3D printing system that would be able to “add rotational motion into the material accumulation process” to ensure fewer supports, if any. To do so, they created a general volume decomposition algorithm, which “can be generally applied to models with different shape and topology.”

“Moreover, a support generation algorithm has been developed for multidirectional 3D printing,” the researchers explained. “The techniques developed here can speedup the manufacturing of 3D printed freeform models by saving the time of producing and removing supports.”

Progressive results of fabricating models on 4DOF multidirectional 3D printing system and a 5DOF system realized on a robotic arm.

The research team’s paper made several technical contributions, including their support-effective algorithm, which is based on beam-guided search and can be applied to 3D models with handles and loops. In addition, they also summarized decomposition criteria through their multidirectional 3D printing process and created “a region-projection based method” for generating supports for multidirectional 3D printing.

There are, however, some drawbacks involved when changing from one 3D printing direction to another, such as slowing down the process, which is why the researchers “prefer a solution with less number of components, which can be achieve by considering the following criterion of clipping.”

A comparison of decomposition results obtained from three schemes introduced in this paper.

“After relaxing the hard-constraint of support-free into minimizing the area of risky faces as described in JG, the scheme of generating support is considerately vital while both feasibility and reliability should be guaranteed,” the researchers wrote. “To tackle this problem, we propose a new pattern called projected supports that ensures the fabrication of remained overhanging regions through a collision-free multi-directional 3D printing.”

The decomposed and 3D printed results fabricated by the system with 4DOF and 5DOF in motion.

The team applied their algorithm to several models, and were able to reduce, and even eliminate in some cases, the need for support structures. In addition, their method’s “computational efficiency” was on par with general 3D printing time.

“We present a volume decomposition framework for the support-effective fabrication of general models by multidirectional 3D printing,” the researchers concluded. “A beam-guided search is conducted in our approach to avoid local optimum when computing decomposition. Different from prior work relying on a skeletal tree structure, our approach is general and can handle models with multiple loops and handles. Moreover, a support generation scheme has been developed in our framework to enable the fabrication of all models. Manufacturing constrains such as the number of rotational axes can be incorporated during the orientation sampling process. As a result, our algorithm supports both the 4DOF and the 5DOF systems. A variety of models have been tested on our approach as examples. Hareware setups have been developed to take the physical experiments for verifying the effectiveness of our system.”

Co-authors of the paper are Chenming Wu, Chengkai Dai, Guoxin Fang, Yong-Jin Liu, and Charlie C.L. Wang.

Discuss this research and other 3D printing topics at 3DPrintBoard.com or share your thoughts below.