PIX 3D Printed Car Chassis
Additive Manufacturing and Generative Design
Additive Manufacturing (AM) is way of making parts through a joining or solidification process with material being added layer by layer. A typical AM system consists of a heat source, a raw material feedstock and a motion system.
The application of Additive Manufacturing has now gone beyond prototyping and is now gradually entering production phase. In the automotive field, Local Motors and Divergent3D have both shown that it is possible to manufacture most of the car parts through AM. Bigger companies like General Motors are now taking notice and trying to incorporate into their vehicles. This trend can be attributed to the fact that AM systems offer huge customization, can quickly adapt to changes in part designs and can bring down supply chain costs drastically.
AM can also be cheaper than traditional manufacturing. If the part has a highly complicated geometry, very high build-to-fly ratio when manufactured conventionally and is made from expensive raw material (Titanium alloys, for instance), then AM can be the cheaper option.
In order to fully exploit the advantages offered by AM, we need to look hard and rethink the way cars are designed. On average, there are about 30,000 parts in every vehicle and printing 30,000 parts is not realistic or practical. This is where Generative Design becomes necessary. Generative Design offers different design solutions for parts and components and if used correctly, can reduce the number of parts. For instance, Generative Design can create an optimized chassis for a vehicle as a single part instead of hundreds. GD can also create parts that are lighter and yet show better or equal structural performance when compared to traditional design.
The main challenge that hinders full acceptance of AM in the automotive industry is high cost and time requirement. Powder based AM systems are still extremely expensive and slow when considering metallic 3D Printing and do not yet justify mass production of vehicles.
Wire Arc Additive Manufacturing — The possible solution
In order to bring down the cost of additive manufacturing for cars, PIX is turning to WAAM (wire arc additive manufacturing).WAAM is an emerging AM technology that uses a combination of an electric arc as heat source and a metal wire as feedstock. The motion system for the arc can be provided by a robot or a CNC machine. In its most basic form, it is the application of robotic welding technology to additive manufacturing. Today, this technique is being given serious consideration by many companies, especially those in the aerospace field. PIX (China), Norsk Titanium (Sweden), Gefertec (Germany), Digital Alloys (USA) are a few companies that are focusing heavily in this technology.
Few years back, WAAM was not given enough consideration by the manufacturing industry as WAAM still can’t produce net shape parts (ready-to-use). Moreover, the layer height for WAAM is about 1–2 mm, leading to a significantly high surface roughness. Heavy distortion due to heat is another major problem. These physical limitations imply that WAAM cannot yet compete with selective laser melting (powder bed) systems on surface finish and producing very complex parts.
Where WAAM trounces powder bed systems is in manufacturing cost and size of parts. Once the initial investment is made in setting up robot and the welding equipment, the primary costs are that of the raw material (welding wires) and electricity. Powder bed systems, on the other hand, are much more expensive with some metallic powders costing as much as gold for the same weight. These systems are still quite far from being economical to produce on a large-scale.
The size of parts made by powder-bed systems is limited by the size of bed. For WAAM, the size is limited by the reach of a robotic arm. The largest CNC and industrial robot systems are larger than the largest powder-bed systems. So, if one wants to print really big and low complexity parts, WAAM can do this quicker and many times cheaper compared to powder bed systems.
Gearing Generative Design towards WAAM — The technique
One of the goals of PIX is to make Generative Design structures by AM, mainly for automotive application. Specifically, PIX aims to develop a car chassis through Generative Design and manufacture through WAAM. The major challenge lies in incorporating WAAM constraints into Generative Design.
During the design of the chassis, several methods were implemented to tackle this challenge. This included performing structural optimization with tools like Autodesk Generative Design, PTC Frustum and Altair Inspire. These software are good at performing structural optimization while considering manufacturing constraints for milling, casting, powder bed based AM etc. But, it is very difficult to take into account of WAAM constraints which are much stricter. So, we came up with two ideas:
Develop our own generative design algorithm and tools that can consider WAAM constraints and then design structures using this algorithm.
Manually modify the generative design results to make it possible to manufacture by WAAM.
The Slime Mold Algorithm
Now, let’s discuss the development of the GD algorithm. This algorithm draws inspiration from the growth of a slime mold organism in nature. The slime mold starts as a single-celled organism. As it grows, it begins branching out a dense network of connections. Then, based on where it finds food, it optimizes the network by keeping only those branches that most efficiently connect to the food sources and pruning back the rest.
This approach is much like the problem of path planning in real life. Researchers have shown that the slime mold can solve complex maze and food problems. When the organism is put in a maze, the network changes its shape to connect two exits by the shortest path. The most famous example is the railway design in Tokyo. Engineers have spent more than 100 years optimizing the rail system, but the slime organism took just 26 hours to reach the same conclusion.
If we think of a mechanical structure to consist of a set of points connected with a small number of lines, it is possible to model the slime mould’s behavior. This is done in the following way.
First, a set of points — load points, support points and evenly distributed model boundary points are defined in space (Fig. 3). These points behave like food sources for a slime mold. Next, all possible and valid connections (lines) between the points are established (Fig. 4b). Then, a real-valued weight parameter (in the domain [0,1]) is given to each point. Now the structure for each design iteration consists of members that are sampled based on the following rule, as seen in Danil Nagy (2018).
Locate the vertex with the highest weight.
Build a structural member that connects this vertex to the next vertex with highest weight.
Decay the weight of both points by multiplying it by a decay parameter
For each design iteration, static FEA (finite element analysis) is carried out to measure the performance of the structure based on maximum deformation and stress. These performance measurements allow us to set up an optimization problem in which the mass becomes the objective function and the performance measurements can be used as constraints. For instance, the constraint can be that maximum deformation cannot exceed 5mm.
Once the optimization problem is set up, it is solved using a gradient-free genetic algorithm. The algorithm tries to minimize mass as well as maximum deformation of the structure. Because these two objectives (mass and deformation) compete with each other, it is up to the designer to decide which objective to give more importance to.
This algorithm is now applied to the design of a 2D automotive frame component. Fig. 3a below shows the chassis of a PIX 4.0 vehicle, an autonomous and electric vehicle project within the company. Fig. 3b shows the base floor that holds several components such as the battery, control system, master cylinders for braking systems, computer etc.
The red and the green points represent load and fixed support points respectively. The loads are taken as weights of the electronic components as well as two humans and are distributed among the red points based on their location with respect to the components. Sample points are placed along the boundary to allow the genetic algorithm to search a large design space for an optimal solution. Once the optimization problem is fully defined, GA searches for an optimum and some of the results are shown below. The algorithm has created more branching at places where the loads are highest.
Because all the structural members in the results are without much complexity, it is possible to fabricate this structure by WAAM. The next step is to study how this algorithm can be extended to 3D structures. PIX is working on this problem.
Generative Design for Chassis
The other approach we used is to apply commercially available structural optimization software such as Autodesk Generative Design, Frustum and Altair Inspire to compute an approximate optimized chassis structure. Structural topology optimization is technique of finding optimum structure that best satisfy design goals under a set of constraints.
While most of these software use the SIMP (Solid Isotropic Material with Penalization) method, Autodesk GD uses the Level-Set method. This method has the advantage that the final solution is smooth and does not suffer from checker-boarding pattern observed in SIMP method, that lead to partial density voxels.
The chassis experiences several complex loading conditions and the best way to get an idea of the loading conditions is by testing — attaching sensors (strain gauges) to the chassis and running the vehicle in severe road conditions. These complex loading conditions make it very difficult for the optimization solver to achieve a converged result, causing the final structures to be very heavy.
The PIX team partnered with Matt Lemay, an Engineer working at Autodesk, who brought in fresh ideas and significantly contributed to the project. Matt Lemay specializes in Generative Design and has been helping many companies understand the benefits of and adopt this new technique. Along with Matt, we tried many ideas, like performing optimization in Autodesk GD without a starting shape, breaking down the chassis into several parts and performing optimization on each part separately. The PIX team is very thankful to Matt and Autodesk for their contribution to the PIX project.
Generative Design Inspired Results
The results calculated by these software do not meet the requirements of WAAM, as the resulting structures are far too complex and the feature thickness at some locations is too low despite considering appropriate thickness and overhang angle during problem setup.
Therefore, optimization needs to be continued to meet WAAM requirements. When designing the HackRod concept car frame, Mehdi Nourbakhs (2016) also used a re-optimization method to re-optimize the topology-calculated solution into a model that can be manufactured by 3D printing. This optimization method distributes a certain number of points on the surface of the generative design result. These points are then connected to other points in its proximity efficiently through straight structural members. The resulting new structure is much more suitable to be printed by WAAM than the Generative Design result.
The last method we tried was the traditional design method. The structure is first computed using a Generative Design software and then re-worked upon by the designer. This approach is quite time-consuming, but the final result can look more organic, smooth and appealing. The re-modeling is mainly done using software like MAYA or TSPLINE. The subdivision modeling algorithm is the main modeling method in Maya. It smooths the mesh by increasing the number of faces of the model, and is suitable for organic Model construction of morphology and nonlinear morphology. Subdivision modeling is often used by pioneering designers, and the resulting form has no blunt corners and is organic.
The design experience has given us insight into the difficulties of including WAAM specific constraints while designing structures using GD. Most commercial software tools for GD are great if the final goal is to make them using Powder-Bed Systems. But, in order to include WAAM constraints, the resulting GD structures need to be relatively simple. The lessons learnt during the process will help PIX in building next generation of automotive structures that will have significantly lesser parts, weigh lesser and have the ability to be made through inexpensive and fast metallic 3D printing technologies like WAAM. This is necessary to revolutionize car manufacturing, which has not seen much change since Ford first introduced the assembly line about a hundred years ago.
This article is contributed by engineers at PIX
Cui Qiang, Computational Design Engineer. PhD candidate at Tsinghua University and master at Beijing University of technology. Majored in Industrial Design Engineering. Winner of multiple competitions.
Siddharth Suhas Pawar Mechanical Engineer. Majored in Aerospace Engineering, Structures and Materials, graduated from University of Michigan.
References
[1]Nourbakhsh, M. , Morris, N. , Bergin, M. , Iorio, F. , & Grandi, D. . (2016). Embedded Sensors and Feedback Loops for Iterative Improvement in Design Synthesis for Additive Manufacturing. Asme International Design Engineering Technical Conferences & Computers & Information in Engineering Conference.
[2]Danil Nagy, Zhao, D., & Benjamin, D. (2018). Nature-based hybrid computational geometry system for optimizing component structure.
[3]Janssen, P., Loh, P., Raonic, A., & Schnabel, M. A.(2017). A Bio-inspired Stigmergic Algorithm Tool for Form-Finding.
[4]Lorensen, W. E. , & Cline, H. E. . (1987). Marching cubes: a high resolution 3d surface construction algorithm. ACM SIGGRAPH Computer Graphics, 163–169.
[5] Bernhard, M., Hansmeyer, M., & Dillenburger, B. (2018). Volumetric modelling for 3D printed architecture. In L. Hesselgren, A. Kilian, O. Sorkine Hornung, S. Malek, K.-G. Olsson, & C. J. K. Williams (Eds.), AAG — Advances in Architectural Geometry(pp. 392–415).
[6] Preisinger, C., & Heimrath, M. (2014). Karamba — a toolkit for parametric structural design. Structural Engineering International, volume 24(24), 217–221(5).
