Autonomous Optimization The Impact of Autonomous Optimization on Foundries
MAGMA shows examples of developing the optimal gating and riser layout in combination with requiring the least amount of time and cost using the optimization tools of the casting process simulation software.
For the past two decades foundries have integrated casting process simulation into the development process for new casting programs. The limiting factor was that casting process simulation only used fixed process parameters to provide results for each calculated configuration. By using simulation trials and avoiding real world trails engineers were able to get a step closer to achieving a good gating, riser and process layout in combination with the making iron castings at the lowest possible cost. A classic conflict exists in the foundry industry as delivery times demanded by the customer are limiting the options of finding the best rigging and process setup for the required casting quality in combination with producing it at the lowest cost. The latest development in casting process optimization technology now takes the industry a step closer to obtaining an optimized casting design and process.
A correctly engineered gating and riser or tooling layout, designed under consideration of common place process variations, should be able to provide quality castings right the first time they are produced and avoid any further modifications of pattern/tooling and process during production. However, this requires an enormous amount of engineering time as engineers need to consider all possible layouts, simulate each of them individually, and analyze all the simulation results generated to eventually get to the final, optimal process layout. In the production world, time is typically the biggest constraint during the process layout phase. In most cases, the engineer can only analyze a few layouts manually to select the “good enough” one. This restricts the engineer from exploring other, possibly better, configurations, i.e. one that tolerates a wider process window.
A traditional single simulation considers only one set of process parameters and does not provide any information on the effect of process variations on the casting’s quality. Process variations have a significant influence on the casting quality. It is essential in finding the optimal rigging and process layout to understand each process parameter’s impact on the casting quality. With traditional simulation tools, a design of experiments can be done manually to understand the process window’s effect on casting quality. However, this requires an enormous amount of an engineer’s time to create the geometries, set up the simulations, and finally, evaluate all the results.
Any simulation tool requires an engineer to create the CAD geometry, enter the process parameters, start the simulation, and analyze the results. Based on these results, the engineer then modifies the CAD geometry and/or process parameters. This manual process of running individual simulations can be very tedious and time consuming. In addition, the results for different simulations are not compiled together to check the optimization progress and it is hard to evaluate all the simulations at the same time. In some cases an engineer runs about 50-60 simulations and still struggles to find the optimal layout.
Autonomous optimization tools integrated into the latest release of MAGMASOFT now allows an engineer to explore a much bigger design space to find the optimal gating and riser, as well as process layout by understanding the effect of process variations on casting quality with less effort and in less time. Unlike traditional simulation, autonomous optimization requires the engineer to go through the simulation setup only once, but many simulations will then be run covering the entire process window. The software autonomously (under “supervision” of the engineer) picks the possible designs, analyzes them and compiles all the results in the form of statistical data.
Several case studies are provided in this article describing how autonomous optimization benefits foundries by helping them to obtain the best process layout for the required casting quality and costs.
Just like when optimizing a casting process in the real-world, autonomous optimization uses the same three basic components:
- Variable Process Parameters (design variables)
- Selected Quality Criteria (output values), as well as calculated quantitative results
- Several Goals (objectives)
Until today, casting process simulation tools have been used by foundry engineers to select process parameters and evaluate simulation results. They then make manual changes to process parameters or geometries like runners, gates or tooling to get closer to achieving the objectives they set. This step-by-step approach can be described as manual optimization (figure 1).
In MAGMASOFT this process is now replaced by setting up an autonomous design of experiments (autonomous DOE) where variable design and process parameters are predefined (blue loop in figure 2). This creates a list of designs, which can be run automatically to i.e. change the number, location and size of risers or process related parameters or alloying elements. Integrating the goals (objectives) leads to a complete autonomous optimization (red) where simulated designs are automatically assessed in regards to how they contribute to the quality and cost objectives. Genetic algorithms and statistical tools are utilized to perform this task. Every simulated design is assessed regarding if and how it brings the casting design and/or process layout closer to the goal of making a quality casting at the lowest costs.
The following are selected examples of how this new approach supports foundry engineers to use their time more efficiently by minimizing their time spent on the computer, as well as how they develop their process understanding much faster than previously thought possible to lay out a robust casting process before the first casting is poured.
Faster Process Understanding
In this example autonomous optimization provides correlation between process parameters and quality criteria leading to faster process understanding. Specifically, the impact of the gating design on the distribution of re-oxidation inclusions (surface cleanliness) on the surface of a steel casting is evaluated. The software ran 12 different gating designs, which were previously prepared in a CAD system. This autonomous DOE was set up and then selected all of the 12 different gating geometries one after another. It then performed the simulations without any interaction of the software user. Once all simulations were concluded, the software offered the engineer immediate and comprehensive insight into all simulation results and how they impact the chosen quality criteria through a newly introduced assessment and evaluation tool. It allows for the fast selection of good and bad designs because all results can be assessed at the same time. No longer does the engineer need to go back into each simulation, bring up the results, and then manually compare them to results of other simulation runs. They’re all available at the same time.
The different designs are shown in the bar diagram in figure 3 and placed in order according to their surface cleanliness. The gating system of a good one (#3) and a bad one (#10) are displayed in figure 4, adjacent to its respective cleanliness simulation result. In addition, the turbulence during the filling process is shown, which can be measured through the free surface area value. The design with the most surface inclusion also shows the most turbulent filling pattern. This graph confirms the commonly accepted correlation between a quiet filling pattern and better cleanliness. Autonomous optimization creates the opportunity to quickly and safely assess quantitative quality criteria. The comparative evaluation of autonomous optimization results detects correlations between process parameters and quality criteria. The engineer can now the gating configuration out of these 12 options that best meets quality and cost requirements of this particular casting.
Engineering Time Reduction
The chill and padding configuration of a ductile iron planetary gear carrier is optimized in this example (figure 5). The feeding of the critical flange area is provided by a combination of top risers and several chills outside and below this area. An additional variable is the option to increase the wall thickness of the area below the risers (padding), leading to several different layouts.
Several parameters were varied to assess which of them has a significant impact on the shrinkage porosity in the critical area. Specifically the size of the padding and the chill configuration outside and below the critical area. This lead to an autonomous design of experiment (DOE) with 32 designs, which were run autonomously by the software. In detail, the software created all geometries of the 32 different designs by itself. It was given dimensional ranges it was allowed to change chill dimensions and locations, as well as how the padding could change by the engineer, but it then created (drew) all geometries by itself. This provides significant time savings for the foundry engineer, as only the first (starting) geometry had to be manually created. Every other design was done by the software and then run as well.
The results of this autonomous DOE (figure 6) show the impact each parameter variation has on the evaluated quality criteria (shrinkage porosity in the flange). Each marker in the scatter diagram represents one design (individual simulation).
The chill configuration has a significant impact on the shrinkage porosity. The layout with the maximum number of chills on the right of figure 6 always leads to small shrinkage defects. As expected, increasing the size of the padding also narrows the size range of defects, but its impact is smaller than the impact that changing the chills has.
It becomes clear which parameter the foundry engineer needs to focus on (in this case the chill configuration) to find the final and optimal configuration.
Robust Process Window 
In the high volume production of castings, it is essential to consistently meet the quality requirements. Process parameters can vary over time. Due to the complex interaction between these parameters, casting processes always operate in a process window. External factors, i.e. in the composition of the alloy or sand impact the casting process and require the adjustment of process parameters.
A high degree of automation in high volume production helps to reduce process variations and is not only economically beneficial. Detailed knowledge on how process parameter changes impact casting quality and economics is essential.
Autonomous DOEs and optimizations provide insight into the relationship between process parameters and casting quality. They assist the foundry engineer in having castings meet quality requirements and making them economically sound. Additionally they provide valuable information on how quality criteria are impacted by process parameter variations or other factors and how robust a casting process is. The utilization of autonomous optimization can thereby determine the best process configuration to assure quality castings are made over the entire process window.
The following example optimizes the riser configuration of an aluminum sand casting (Figure 7). The homogeneous temperature distribution confirms the successful prior optimization of the gating system, which lead to the synchronized filling of all cavities.
Each casting is filled through a “hot” riser. Additionally, two “cold” risers are located on top of each casting. First, the hot riser was optimized. An autonomous optimization with the objective “minimize shrinkage porosity inside the casting” and “minimize riser volume” was performed. A total of 45 designs (simulations of filling and solidification for each) were completed in less than 8 hours without any human interaction. Figure 8 shows the best design combining no shrinkage defects with the smallest riser volume.
The scatter chart (figure 9) displays shrinkage porosity versus riser volume. The optimal design is colored in green. The chart also shows that there are other options to produce a defect free casting (horizontal group of markers) but they all would require a larger (more expensive) hot riser.
The optimal design provides the optimal combination of getting a defect free casting at the lowest cost, assuming the other casting process parameter are kept constant. Autonomous optimization can also be used to evaluate these other process variations and determine if this gating and riser configuration delivers sound castings for the entire process window. Figure 9 shows the “Hot Spot FS time” result, evaluating the feeding performance for the optimal layout (left) and a slightly modified design (right). Both designs produce castings with no shrinkage porosity. However, the deemed optimal design shows the tendency of extending the critical feeding area into the casting, which could be cut off when other process parameters (i.e. the pouring temperature) change even slightly. The modified design doesn’t show this tendency and is therefore providing a more robust setup. This robustness criterion can also be autonomously optimized in the software.
Simultaneous Optimization of Geometries and Process Parameters
The section size of runner systems for high pressure die castings is usually smaller, the farther it is away from the biscuit. The runner design of this example (figure 10) is autonomously optimized by varying the locations of the first and second runner contraction over the length of the main runner (see uninterrupted and dashed lines in figure 10).
An autonomous DOE with 98 distinct designs was performed were all geometry changes were autonomously generated, simulated, and then assessed. Figure 11 shows a comparison between the worst (left) and best (right) design using the local filling time results as quality criterion. The objective of the autonomous optimization was to minimize the difference of the local filling time between the two cavities. Only the right side of the system was simulated to utilize the symmetry of the layout and save calculation time. The best design (right) shows the desired nearly simultaneous filling of both cavities.
The quality of high pressure die castings is strongly related to the coordination between the runner design and the shot curve parameters. So while this autonomous optimization changed the location of runner contractions, it simultaneously needed to vary the start of the acceleration phase of the shot curve. The piston velocity profile over the piston travel is shown in the upper section of figure 12.
The start of the acceleration phase always lays in-between the vertical green and blue lines. The lower section of figure 12 shows the main effect diagrams depicting the impact of changing the start of the acceleration phase (upper diagram) and the impact of moving runner contraction location (lower diagram) on the difference in filling of the casting cavities. Each marker in a main diagram represents one average value of all measurements of the autonomous DOE for a specific acceleration point or runner contraction location. The incline of the lines represents the significance of the respective process parameter on achieving the objective. The steeper the incline, the greater the positive impact.
This example confirms that changing one process parameter always impacts other process parameters or tooling features, which need to be considered simultaneously when optimizing a process.
Several examples were used to show how the complete integration of autonomous optimization in the casting process simulation tool MAGMASOFT can be used to save engineering time, gain process understanding faster, and to create robust casting process windows by autonomously optimizing process parameters and casting process related design features. The software, for example, finds the best process parameter combinations for runner and gate dimensions, as well as locations and dimensions of risers and chills. Foundry engineers can use autonomous optimization to simultaneously achieve quality and cost targets.
The goal of retaining the user-friendliness of this simulation tool when integrating this new methodology was achieved by implementing semi-automatic options for the generation of parametric geometry variations and statistical tool in combination with genetic algorithms. The simultaneous assessment of results derived from the autonomous optimizations enable the foundry engineer to easily compare and evaluate them. Dependencies between design and process variables, quality criteria and objectives are clearly visualized.
30 years after the introduction of casting process simulation, foundry engineers now can combine single simulations, autonomous DOEs and autonomous optimizations to gain better process understanding and to establish robust casting processes making quality castings at the lowest cost possible.
 Hahn, I., Sturm, J.C.: Versuchspläne in der gießtechnischen Simulation, GIESSEREI 96 (2009), Nr. 7
 Bramann, H., Pavlak, L.: Innovatives Produktdesign und robuste Prozessauslegung durch virtuelles Experimentieren mit der Gießprozess-Simulation. GIESSEREI 102 (2015), Nr. 2
 Dieckhues, G., Gummersbach, T., Schmidt, D.: Feeder Design for an Aluminium Sand Casting, Vortrag auf dem Internationalen MAGMA User Meeting, Potsdam, Oktober 2014