Web based Simulation Workflow Optimization algorithm applied in aeronautics: Manufacturing of control surfaces

Abstract

Simulation workflow optimization has become an important investigation area as it allows users to process large scale & heterogeneous problems in distributed environments in a more flexible way. The most characteristic category of such problems comes from the aerospace industry. In this work a specially developed Simulation Workflow Optimization (SWO) algorithm that is based on heuristic optimization techniques (Genetic Algorithms) and delivers an optimized workflow implementation of an initial plan or workflow schedule, will be applied on an aerospace manufacturing problem in order to demonstrate its potentials. The algorithm has been developed under the iProd EU project and the application use case refers to the manufacturing of an airplane tail rudder from FAE. The SWO tool helps the user to select the optimal use of local and external resources that will satisfy the product requirements under the specific time & cost constraints. The tool is customized for the specific domain/application and it is remotely invoked via web GUI & services under the iProd collaborative framework.

Full text

6th International Conference from “Scientific Computing to Computational Engineering”
6th IC-SCCE
Athens, 9-12 July, 2014
© IC-SCCE
WEB-BASED SIMULATION WORKFLOW OPTIMIZATION ALGORITHM
APPLIED IN AERONAUTICS: MANUFACTURING OF CONTROL SURFACES
Moussas V.C.1,2, Tsahalis J.1 and Tsahalis H.-T.1
1 Paragon S.A.
Karaoli & Dimitriou 13, Galatsi, GR-11143, Athens, Greece
e-mail: jtsahalis@paragon.gr, web page: http://www.paragon.gr
2 School of Technological Applications, Tech. Educ. Inst. (TEI) of Athens,
Ag. Spyridonos Str., Egaleo 12210, Athens, Greece
e-mail: vmouss@teiath.gr, web page: http://users.teiath.gr/vmouss/
Keywords: Optimization methods; scheduling; simulation workflow; evolutionary algorithms; web services;
distributed tools; heterogeneous network;
Abstract. Simulation workflow optimization has become an important investigation area as it allows users to
process large scale & heterogeneous problems in distributed environments in a more flexible way. The most
characteristic category of such problems comes from the aerospace industry. In this work a specially developed
Simulation Workflow Optimization (SWO) algorithm that is based on heuristic optimization techniques (Genetic
Algorithms) and delivers an optimized workflow implementation of an initial plan or workflow schedule, will be
applied on an aerospace manufacturing problem in order to demonstrate its potentials. The algorithm has been
developed under the ‘iProd’ EU project and the application use case refers to the manufacturing of an airplane
tail rudder from FAE. The SWO tool helps the user to select the optimal use of local and external resources that
will satisfy the product requirements under the specific time & cost constraints. The tool is customized for the
specific domain/application and it is remotely invoked via web GUI & services under the ‘iProd’ collaborative
framework.
1 INTRODUCTION
Data and knowledge management technologies are of strategic importance for industrial innovation, provided
they are integrated in the company processes, in the organizational structure, and can be flexibly adapted to
company evolution. In particular the Product Development Process (PDP) of manufacturing companies requires
the efficient management of huge amounts of data from different sources and their integration in the sub-
processes that compose the product chain. The efficient use of information lifecycle, by the large adoption of
virtual testing and by the inter-functional management of related data in the product management is expected to
become a strategic advantage for the innovation race.
The aim of the EU Project “Integrated management of PROduct heterogeneous Data” (iProd) [1] is to
improve the efficiency and quality of the Product Development Process developing a flexible, service oriented,
customer driven software framework that will be the backbone of computer systems associated with current and
new development processes. To achieve these goals, iProd relies on knowledge management (KM), knowledge
based engineering (KBE) and process integration & automation technologies and optimization. The knowledge
base along with a reasoning engine support information sharing, collaboration across companies, and promote
efficient decision taking. The iProd framework will impact the product development process in order to reduce
drastically product development costs by means of an optimized testing process, support knowledge and
competencies extraction, structuring and sharing also with suppliers, improve focus of new product development
with a fast and structured management of competitor and market analysis data (figure 1).
The work for PDP improvement involves the development and application of test planning and optimization
methodologies, which are part of the iProd Reasoning Engine, their end result being detailed optimal workflows
for applications areas such as Aerospace, Automotive and Appliances.

Moussas V.C., Tsahalis J. and Tsahalis H.-T.
Figure 1. The iProd Framework
A main function of the reasoning engine is the optimization of the physical and virtual tests of a PDP. Product
testing and verification procedures need optimization techniques in order to achieve the most efficient schedule
of both simulated and physical tests required. Today’s Product Development Processes (PDPs) are becoming
more and more decentralized and distributed. As a result, the corresponding physical & virtual (i.e. human &
simulation) test workflows are also becoming more complex as well as, distributed and heterogeneous and thus,
the overall PDP optimization problem becomes more complex, with multiple & contradictory objectives and
requires powerful and/or specially designed optimization tools [2].
Under iProd framework a special tool for Simulation Workflow Optimization (SWO) was developed in order
to support the optimization of virtual (simulation) workflows in cases of distributed and heterogeneous networks
of collaborating systems. The aim is to present a flexible web based tool [3] that will be able to promote a
simulation workflow optimization method, make it available to a remote application or another service, and
support a wider automated collaboration between heterogeneous design & simulation tools (figure 2).
Figure 2. The role of SWO in the PDP process
A more detailed description of the SWO module can be found in [4]. In this work we will present the results of
PRODUCT
DEVELOP
MENT
PROCESS
(PDP)

Moussas V.C., Tsahalis J. and Tsahalis H.-T.
the SWO tool when used for a sample complex and heterogeneous manufacturing process from the aerospace
domain that requires a significant number of virtual tests.
2 THE FAE USE CASE
Aircraft box structures are a perfect compromise between weight and price. The conceptual design process of
these structures is a typical Multi-disciplinary Design and Optimization effort, mostly performed by human
engineers. The iterative nature of MDO turns the development of such components into a long and costly
process. Fokker Aerostructures (FAE) is focusing on a number of Key Technologies, like thermoplastic
composites and on Operational Excellence to offer their customers short lead times and affordable costs on such
highly innovative products. The current climate for suppliers of aircraft structures is very competitive. Besides
low cost and low weight, aircraft OEMs demand a shorter time-to-market. This requires the ability to develop
aircraft parts quickly and cost effectively.
The process of developing an aircraft part usually starts with a Request for Proposal (RFP) by the aircraft
integrator. After the proposal phase, two phases follow (figure 3) in which the design is detailed further: the
preliminary and the detailed design phase. In each phase models are used with different level of fidelity. The
challenge in each of these phases is the iterative nature between design, stress analysis, sizing and performance
analysis. After a design is made, linear and non-linear stress analysis is performed on the design to check if stress
limits are not exceeded. Based on that analysis the structural elements are sized in order to stay within limits.
Based on the sizing a performance analysis can be made with respect to weight and cost.
Another challenge lies in the use of a common geometric model of the aircraft part, usually by means of a
CAD model. This CAD model is used for Finite Element Analyses and also provides information for the
recurring and non-recurring cost estimations. It is extremely important that all disciplines use the same basis.
Currently the design of a rudder is time-consuming and costly. It requires a lead time in the order of several
months. Also time and resource constraints do not allow for many design iterations, such that a feasible but sub-
optimal design is achieved.
Figure 3: As-is design process of a rudder at Fokker Aerostructures
The Fokker Aerostructures (rudder) use case involves the search for an optimal rudder design through
investigation of multiple repeated simulations of varying configurations while trying to satisfy conflicting
requirements such as: customer satisfaction, weight minimization, manufacturing cost minimization.
As shown in figure 3, a number of physical and virtual test procedures are required for the design & analysis
of the product. Each “virtual test” (shown in blue) is composed of a large number of simulations & analysis runs
that need to be executed, in order to find a better or the optimal product design.
The duration of these virtual tests varies from a few days to several weeks or months. For instance, the
“Update FEM analysis” subtask requires 60 days to complete all required simulations (figure 4). By using SWO
on each one of these virtual tests and by optimizing each simulation workflow, a significant amount of time can
be saved, or alternatively, a larger set of product variations could be tested, both resulting to increased customer
satisfaction. The only requirement is the detailed knowledge of the specific task and the availability of the
involved resources together with the objective that the designer wants to achieve.

Moussas V.C., Tsahalis J. and Tsahalis H.-T.
Figure 4. Sample list of various virtual tests in a Gantt form.
3 SWO APPLICATION
For the specific aerospace industry and the rudder manufacturing use case, five main parameters were
selected as they may characterize the proposed designs: the material, the geometry or complexity of the rudder,
the weight, the task duration and the cost.
The SWO tool was assigned to optimize a “virtual test”, the one that is dealing with the Finite Elements
analysis (“Update FEM Analysis”). It is a test that requires many smaller jobs to run and it is using a more
complex infrastructure of parallel resources that may allow more space for optimization. SWO calculates an
optimal simulation workflow for those jobs that satisfies the user objectives. The major task details and subtask
parameters involved are divided in three categories: the user objectives, the use case specific and the SWO tool
specific. Typical relationships between controlling parameters and resulting indices & scores are shown in figure
5. All parameters are retrieved from the KB where the use case details are stored and updated. Some of the
parameters are fixed but others are user defined, such as (Table 1):
a) The objectives that describe the main goal the designer is trying to achieve,
b) The use case parameters that describe the availability and the capability of the resources, and,
c) The SWO tool parameters that control the function of the Genetic Algorithm
Table 1. User Defined Parameters for the FAE use case
USER PARAMETERS
DESCRIPTION
OBJECTIVES PRIORITY
Minimize Weight
Set the user priority of minimizing the product weight.
From 1 (highest) to 5 (lower), default = 1.
Minimize Cost
Set the user priority of minimizing the product cost.
From 1 (highest) to 5 (lower), default = 1.
Minimize Duration
Set the user priority of minimizing the required time.
From 1 (highest) to 5 (lower), default = 1.
USE CASE OPTIONS
Avail. FE Tracks
Set the Maximum number of available FEM Parallel Computing Tracks. From 2 to 8,
default = 8, 1 = no parallel support.
Avail. Tokens
Set the Maximum number of Tokens available to run the simulations. From 300 to 1000,
default = 1000.
Avail. Memory
Set the Maximum GB of available Memory.
From 30 to 100, default 100.
GA TOOL OPTIONS
Population
Set the Size of Initial Population for the Genetic Algorithm (GA). Default = 13. (<15)
Generations
Set the Maximum number of Genetic Algorithm (GA) generations. Default = 120. (<200)
Figure 5. Relationships between controlling parameters and final indices & scores.

Moussas V.C., Tsahalis J. and Tsahalis H.-T.
4 SWO RESULTS
The iProd framework offers the supporting platform for PDP improvement and the connection to the SWO
tool. The entire setup is connecting several distributed modules and web services in one user interface. The user
selects the domain and the product of interest, and proceeds with the Correlation matrix, Work Breakdown &
Task Planning tasks (figure 6). At the end the user is provided with a workflow of physical (human) & virtual
(simulation) tasks to execute.
Before executing the simulation tasks on its computer resources, the user can now use SWO and attempt to
further optimize internally these tasks. From the Virtual Test Manager tab the user selects the SWO tool and uses
the SWO GUI to prepare and submit the request.
Figure 6. The SWO page in the software integration framework of iProd.
The next steps for the user are: load the UC data from the KB, adjust the user defined parameters and then pass to
the execution phase by calling the optimization service. When the SWO run is finished, the following information
is returned (figure 7): A table containing the top 5 simulation cases selected by the SWO tool that best satisfy the
UC data and the user criteria, and, a plot showing the convergence of the GA algorithm to these solutions.
Figure 7. The first SWO results with the top 5 of the calculated solutions.
The user may ask to view in more detail the SWO results, in tables showing the GA populations of the
simulation cases selected at specific phases of the optimization procedure. The three tables contain the
population from the final solution (100% convergence) as well as the populations (fittest cases) from two earlier
runs, at a convergence rate of 60% and 30% (Figure 8). The tables contain the IDs that define the various
simulation tests and the estimated objective values.

Moussas V.C., Tsahalis J. and Tsahalis H.-T.
Figure 7. The detailed SWO results with the fittest individuals at three instances.
The final step for the user is to display the proposed plan of simulation tests based on the above results (figure
8). From the four plots, the first plot shows the entire simulation plan as proposed by SWO, showing the
estimated duration and its relation to the initially expected schedule.
The three other plots bellow show in detail the three phases of the simulation plan. Each phase contains
Linear & Non-Linear FE Analysis runs of the simulation test selected by the tool and shows how they have been
arranged to run (submitted for execution) by the user. The potential use of parallel tracks is also shown in these
plots.
Additionally, the user may retrieve the above details in a text file reporting the proposed list of simulation
runs to execute and their schedule, in order to prepare the corresponding simulation workflows for execution by
the virtual test manager (Figure 9).
From the above results it is clear that by using SWO and the parallel FEM capability to optimize the
simulation workflow a significant amount of time can be saved (aprox. 40%) that would permit either to finish
earlier or to try more solutions.
5 CONCLUSIONS
In this document, the application of the Simulation Workflow Optimization tool developed in the iProd
project in the domain of the aerospace manufacturing was presented. The Simulation workflow optimization is
used to optimize the individual simulation parameters and configurations for the specific application, based on
the overall PDP requirements and objectives. The result of SWO is a proposed list of simulation tasks that will
converge to the optimal solution but in a shorter time. The significant amount of time saved will permit the user
either to finish earlier the design phase or to try more solutions towards customer satisfaction.
6 ACKNOWLEDGMENTS
The work presented in this paper has been performed under the EU-funded R&D project IPROD with
contract number FP7 FP7-ICT-2009-5-257657.
REFERENCES
[1] iProd project: Integrated management of product heterogeneous data, http://www.iprod-project.eu
[2] Lee H. and Kim S.-S. (2001), “Integration of Process Planning and Scheduling Using Simulation Based
Genetic Algorithms”. Int J Adv Manuf Technol 18:586–590, 2001.
[3] Tsahalis, J., Tsahalis, H.-T. Moussas, V.C. (2013) “Optimization of a Heterogeneous Simulations
Workflow”, Proceedings of the 5th IC-EpsMsO, Athens, 3-6 July, 2013.
[4] Tsahalis, J., Moussas, V.C., Tsahalis, H.-T. (2014) “An Algorithm for Distributed Heterogeneous Simulation
Workflow Optimization”, Proceedings of the 6th IC-SCCE, Athens, 9-12 July, 2014.

Moussas V.C., Tsahalis J. and Tsahalis H.-T.
Figure 8. The proposed plan of simulation tests and the corresponding Gantt charts.

Moussas V.C., Tsahalis J. and Tsahalis H.-T.
SIM SCHEDULE DESIGN CASE TO SimWF ESTIMATED RESULTS
FEtrack StartTime Material Geometry MeshDetail Weight Cost Duration
LIN 2 1 M1 G070 N61 36.6 kg 8.1 k$ 475 min
LIN 3 943 M2 G039 N49 34.3 kg 11.8 k$ 382 min
LIN 3 1 M2 G069 N61 32.9 kg 9.1 k$ 475 min
LIN 4 1 M2 G096 N61 31.7 kg 9.4 k$ 475 min
LIN 5 904 M2 G118 N51 30.7 kg 15.6 k$ 397 min
LIN 3 476 M2 G136 N60 29.9 kg 8.8 k$ 467 min
LIN 1 903 M2 G152 N53 29.2 kg 14.4 k$ 413 min
LIN 5 1 M2 G159 N61 28.9 kg 17.2 k$ 475 min
LIN 1 483 M2 G160 N54 28.8 kg 9.4 k$ 420 min
LIN 5 476 M2 G161 N55 28.8 kg 8.1 k$ 428 min
LIN 4 476 M2 G169 N60 28.4 kg 9.3 k$ 467 min
LIN 2 476 M2 G169 N61 28.4 kg 9.3 k$ 475 min
LIN 1 1 M3 G190 N62 22.9 kg 14.9 k$ 482 min
FEtrack StartTime Material Geometry MeshDetail Weight Cost Duration
NLN 2 1325 M1 G070 N61 36.6 kg 8.1 k$ 3800 min
NLN 1 15765 M2 G039 N49 34.3 kg 11.8 k$ 3056 min
NLN 3 1325 M2 G069 N61 32.9 kg 9.1 k$ 3800 min
NLN 2 5125 M2 G096 N61 31.7 kg 9.4 k$ 3800 min
NLN 3 12661 M2 G118 N51 30.7 kg 15.6 k$ 3176 min
NLN 2 8925 M2 G136 N60 29.9 kg 8.8 k$ 3736 min
NLN 2 12661 M2 G152 N53 29.2 kg 14.4 k$ 3304 min
NLN 3 5125 M2 G159 N61 28.9 kg 17.2 k$ 3800 min
NLN 1 12405 M2 G160 N54 28.8 kg 9.4 k$ 3360 min
NLN 1 8981 M2 G161 N55 28.8 kg 8.1 k$ 3424 min
NLN 3 8925 M2 G169 N60 28.4 kg 9.3 k$ 3736 min
NLN 1 5181 M2 G169 N61 28.4 kg 9.3 k$ 3800 min
NLN 1 1325 M3 G190 N62 22.9 kg 14.9 k$ 3856 min
FEtrack StartTime Material Geometry MeshDetail Weight Cost Duration
LIN 5 18821 M1 G150 N53 32.6 kg 7.9 k$ 413 min
LIN 5 19234 M1 G174 N53 31.4 kg 6.1 k$ 413 min
LIN 2 18821 M2 G160 N58 28.8 kg 9.4 k$ 451 min
LIN 3 19272 M2 G161 N46 28.8 kg 8.1 k$ 359 min
LIN 4 19249 M2 G161 N51 28.8 kg 8.1 k$ 397 min
LIN 4 18821 M2 G161 N55 28.8 kg 8.1 k$ 428 min
LIN 3 18821 M2 G161 N58 28.8 kg 8.1 k$ 451 min
LIN 1 19288 M2 G162 N43 28.8 kg 13.1 k$ 335 min
LIN 1 18821 M2 G167 N60 28.5 kg 12.1 k$ 467 min
LIN 2 19272 M2 G190 N49 27.5 kg 8.8 k$ 382 min
FEtrack StartTime Material Geometry MeshDetail Weight Cost Duration
NLN 3 23262 M1 G150 N53 32.6 kg 7.9 k$ 3304 min
NLN 1 23390 M1 G174 N53 31.4 kg 6.1 k$ 3304 min
NLN 2 19654 M2 G160 N58 28.8 kg 9.4 k$ 3608 min
NLN 1 26694 M2 G161 N46 28.8 kg 8.1 k$ 2872 min
NLN 3 26566 M2 G161 N51 28.8 kg 8.1 k$ 3176 min
NLN 2 23262 M2 G161 N55 28.8 kg 8.1 k$ 3424 min
NLN 3 19654 M2 G161 N58 28.8 kg 8.1 k$ 3608 min
NLN 1 29566 M2 G162 N43 28.8 kg 13.1 k$ 2680 min
NLN 1 19654 M2 G167 N60 28.5 kg 12.1 k$ 3736 min
NLN 2 26686 M2 G190 N49 27.5 kg 8.8 k$ 3056 min
FEtrack StartTime Material Geometry MeshDetail Weight Cost Duration
LIN 1 32246 M1 G190 N64 30.6 kg 6.6 k$ 498 min
LIN 2 32246 M1 G191 N64 30.5 kg 13.4 k$ 498 min
LIN 3 32246 M1 G193 N64 30.4 kg 9.9 k$ 498 min
FEtrack StartTime Material Geometry MeshDetail Weight Cost Duration
NLN 1 32744 M1 G190 N64 30.6 kg 6.6 k$ 3984 min
NLN 2 32744 M1 G191 N64 30.5 kg 13.4 k$ 3984 min
NLN 3 32744 M1 G193 N64 30.4 kg 9.9 k$ 3984 min
Figure 9. The proposed list of simulation runs