**1. Introduction**

Composite materials are increasingly used in ultralight aerospace designs. Composite load-bearing structures allow design and manufacture unmanned aerial vehicles (UAVs) that weigh no more than a dozen kilograms and have wingspans exceeding 20 m. The strength of the composite materials is no longer a noticeable barrier. High-strength composite materials that significantly reduce weight are already widely available. However, other problems have not been solved so far. One of the most important is related to the loss of stability [1,2], which is the primary criterion for shaping such design structure. Among the other problems are the use of technologies to produce large-size structures from materials with thicknesses much less than 1 mm, and the challenge of designing large-size structures with extremely thin walls [3]. Forming a highly flexible structure, especially in aviation operating conditions—i.e., significant changes in geometry under varying load conditions—and aeroelastic phenomena associated with the operation of such an arrangemen<sup>t</sup> are issues that must be considered [4]. Structural shaping to meet the abovementioned requirements is becoming a fundamental issue. It is essential to analyze the spectrum of possible solutions at the concept and preliminary design stages in such conditions [5–9].

Time-consuming analyses and geometric modelling of structures make this process ineffective. Therefore, in such a particular case, using highly flexible thin-walled aircraft designs, knowledge-based engineering methods, and specific generative modelling was proposed [10,11]. This should allow the rapid generation of solutions and automate the support structure's modelling process based on simplified design criteria.

**Citation:** Skarka, W.; Jałowiecki, A. Automation of a Thin-Layer Load-Bearing Structure Design on the Example of High Altitude Long Endurance Unmanned Aerial Vehicle (HALE UAV). *Appl. Sci.* **2021**, *11*, 2645. https://doi.org/10.3390/ app11062645

Academic Editor: Marek Krawczuk

Received: 7 February 2021 Accepted: 10 March 2021 Published: 16 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

The generative model's operation is combined with a set of simplified verification methods based on Finite Element Method (FEM) used for a simplified structure model based on integrated load states. This method allows one to quickly evaluate individual solutions and choose a solution for further optimization analysis, which is carried out in the following design stages [10,11].

To solve the problem mentioned in the introduction, the authors propose a design methodology based on the use of the generative modelling technique, and the principles of forming highly flexible aircraft structures and thin-film composites with special attention focused on the FEM validation of the preliminary geometry [12,13].

The background of these issues is presented in Section 2 (state of the art) and the methodology proposed by the authors is described in Section 3—developed methodology. For the methodology verification, the example of a HALE UAV built by the authors was used, and verification details are presented in Section 4—use case. The article ends with conclusions and an assessment of the effectiveness of the methodology used.

#### **2. State of the Art**

In the design processes, strictly defined or even standardized activities are used, including the design of repeatable parts of the structure. A common practice is to make extensive use of these opportunities through Design Reuse (DR). Computer Aided Design (CAD) techniques make use of specific model solutions (CAD models) along with the principles of determining their features, which make the design process much easier [14].

The idea of using DR in the CAD systems environment gives rise to new possibilities, from simple parameterization to complex Generative Models (GMs). These methods can be applied to small parts, elements, and entire components, and products [10]. Such automation of constructing a geometric model in the CAD environment is also called Geometry-Based Design Automation (GBDA). It is not the only form of automation of the design process. Still, it stands out due to the widespread use of CAD systems in the engineering, automotive, aviation industries, etc. In individual CAD systems, tool-based forms of geometric modelling are quite different. These range from programming and scripting techniques, through extensive construction of various template complexity levels (high-level CAD templates), allowing for the integration of multiple forms of knowledge into the geometric model.

Automation itself accelerates the design process at some stage—e.g., it allows for faster generation of geometry and leads to the designer being able to study a much larger space of possible solutions simultaneously [1]. Such automation opens the way to integrating geometric modelling into optimization processes and, with complex multidisciplinary problems, to Multidisciplinary Design Optimization (MDO). However, it should be remembered that there is no definite methodology for the automation of design processes. It is also impossible to answer which of these works can be automated and with what techniques. This applies to the application of knowledge-based engineering methods. The proposed solutions are good practices rather than principles [15].

In a multidisciplinary approach, the geometric model plays a strategic role. So, it is possible to implement the MDO approach [16]:


In both approaches, a and b, neither the geometric model nor the data that are used in the optimization process are saved in the form of a geometric model. There is no need to make such a model for the calculations themselves, which significantly speeds up the optimization loop time and reduces the labor and computational demand. The aforementioned approaches are used when such a model, by definition, is not needed in methods that use mathematical models of phenomena or semiempirical or statistical methods. If the model is built based on the geometric data used for the calculations, it is used by experts for visualization and comparative purposes only, which helps to interpret the results.

**Figure 1.** Multidisciplinary Design Optimization (MDO) framework without geometric model in the loop.

**Figure 2.** MDO framework with geometric model in the loop.

In the second group of the optimization approaches (c, d), we deal with a geometric model in the calculation process. The geometry used for calculations is taken from the model, or the geometry data used in the calculations are visualized through the visualization interface. Typically, these methods are used when some form of geometric structure model is required—e.g., the finite element method (FEM) or Vortex Lattice Method (VLM) or any calculation method using geometry input [11]. Depending on the calculation method and the form of the model used, the form of the model itself may be simplified or integrated into a complex CAD model. In both the first and second forms, generative models can be used to assist in the generation of geometry.

Especially for the latter and multidisciplinary problems, the optimization computation models can differ significantly from each other. Two types of Simultaneous Analysis and Design (SAND) and Nested Analysis and Design (NAND) models stand out from the several different possible configurations of multidisciplinary computation. With a SAND configuration, the disciplinary optimizer simultaneously determines the value of both design and state variables at a disciplinary level. In contrast, in the case of a NAND framework, the optimizer only determines the design variables, with state variables needing to be calculated at each iteration [17].

Automation of the design process in aviation applications and the application to the supporting structures of flying objects are the subject of intensive research. This is not only because of:


Such integrated approaches are also called MDO (multidisciplinary design optimization) [7,16,18], Model-Based Design (MBD) [8,19], Model-Based Optimization (MBO) [20].

Usually, CSM methods are used as high-fidelity models in the detailed design phase and sometimes also as medium-fidelity models in the preliminary design stage on the simplified geometry. Although, in the conceptual design phase there is often a need to obtain alternative concepts load-bearing structures. On the other hand, relying on simplified geometry encounters a severe problem precisely because of the large amount of work involved in generating alternative concepts of support structures, which often exceed the time required for numerical analyses and their interpretation. Such a bottleneck is usually eliminated using methods of automating the process of generating a geometric model. It is necessary to use a model with a high degree of detail, considering the aspect of manufacturing technology, design dependencies, the engineering correctness of the model, and structural consistency. Generative modeling yields good results [10].

Generative modelling (GM) is one of the most popular knowledge-based engineering techniques used in the design field [8,15,21]. The main idea behind GM is to elaborate the CAD model, which will automatically, or at least semiautomatically, generate the model's geometry based on the design requirements and integrated knowledge [9,15,22]. The general schema of GM functioning is presented in Figure 3.

**Figure 3.** The general schema of the GM functioning [10,15,22].

The essential advantage of using the GM, which is one of knowledge-based engineering (KBE) methods of design automation, is the possibility to reduce the time a designer spends on some routine design tasks and put more effort into creative parts of the project, as

presented in Figure 4 [15,22,23]. Thanks to this, it is possible to develop a new approach to some problems, which may be better and more efficient than those currently used. On the other side of the coin, development of the GM is time-consuming and usually requires the assistance of an expert in the field [15,21]. Moreover, the number of knowledge engineers who are essential during knowledge acquisition and structuralizing is limited [15,19,23]. However, the effort put into the GM development may bring more benefits in the long-term perspective, especially in the number of projects that can be realized in a shorter time and with less effort, as presented in Figure 4 [9,15,23].

**Figure 4.** Benefits of the knowledge-based engineering (KBE) approach [10]. Copyright 2007 Elsevier.

## **3. Developed Methodology**

The principal authors' purpose is to simplify the conceptual stage of the aircraft design as much as possible by using generative modelling techniques combined with FEM analysis. The result of these actions is the systematic approach presented in Figure 5 in a simple schema. Because the design process based on the GMs is slightly different from the traditional design method, the authors tried to adapt the generation procedure to specific requirements of the FEM validation.

**Figure 5.** The general schema of the developed methodology.

At different stages of the project development, different approaches to defining material data were used. In the initial stages, the isotropic material model was used, and in the next stages the anisotropic model was used for composite materials. The latter was based on general manufacturer data in the initial stages and then the tests performed on samples with a given structure.

### *3.1. CAD Model Development*

As we can see in the schema in Figure 5, the first stage of the process was the generation of the load-bearing structure's primary geometry. In this process, the series of the GMs were in use. Each component, such as a spar, a rib or a stringer, was designed individually with its own set of parameters.

The only common thing that connects all components is the geometrical input in the form of airfoil curves for the next sections of the wing. In the proposed approach, each section of a wing is designed independently, and as a section, we mean a part of a wing which contains the same type of an airfoil profile. Using the shared geometrical data makes any change in the geometrical input forces automatically rebuild all related elements without user attention.

The same approach was used in the stage of further development of the models. If the FEM validation results were satisfying, the user could add additional features related to the manufacturing method or specific requirements of the used material or technology.

In the presented approach, the GMs for additional features use existing geometry, from the previous stage of the design process, as the geometrical input. That creates a hierarchical feedback loop that forces automatically rebuilding all related elements—i.e., change of the airfoil causes revamp of the spar. Changes in the spar make changes in the spar features. The sequence of changes in this hierarchy is shown in Figure 6.

**Figure 6.** The hierarchical order of automatic rebuilding.

In the GM elaboration process it is essential to decompose the design object into smaller and simpler objects to generate. The authors prepared a series of hierarchical diagrams where all features were extracted and described by parameters, required geometrical inputs, rules, etc. In Figure 7, the main idea of the decomposition process is presented.

**Figure 7.** The idea of the decomposition process.

For all stages of model development, the numerical validation procedure was the same. The basic idea of the numerical validation process is presented in Figure 8. The proposed methodology is based on three standard stages: (1) preprocessing, where we worked with a CAD model to obtain the numerical representation of the validated model; (2) computation, which in our case consists of three stages: frequencies analysis, structural analysis, and buckling analysis; (3) postprocessing, where a user can decide if the model fulfils all required assumption or need some changes and to be revalidated based on the results.

**Figure 8.** The methodology of the Finite Element Method (FEM) validation.

At different stages of the project development, a different approach was used to define material data. In the conceptual design, a very simplified approach is usually used—the material is defined as isotropic for analytical calculations and the calculations are to indicate the expected shape and cross-sections of the supporting structures. If FEM verification calculations are carried out at this stage, then simplified models and material data are also defined in this way. Material data were selected depending on the intended target material and many different options were explored. In preliminary design for composite elements, the preliminary internal structure of the reinforcement was selected and the material data were determined based on the general material data for the reinforcement class. Only in the next stages, and in particular during the production of the prototype, were samples made to fully identify the material data of the given composite structures; only such data give accurate results. Each subsequent stage of approximation brings the results closer to those expected in reality.

### *3.2. Numerical Model Preparation*

The basis of Finite Element Analysis(FEA) analysis is the CAD model, and depending on the design stage, the model has different levels of detail. So, it was necessary to elaborate a process after which the input geometry for analysis will mostly be the same. This issue forced the authors to add additional steps that proceed with the central part of the study.

To unify the input geometry, the authors decided that the CAD model must be adequately prepared in the preprocessing stage. This preparation includes:


After the simplification process, the model must be divided into segments that correspond to the composite structure. An example of the simplification and segmentation process is shown in Figure 9.

**Figure 9.** An example of the simplification and segmentation process. On the left is the model before the process, and the right is the model after the process.

> In the segmentation process, each composite structure acquires a unique color code, which allows to quickly interpret the model material structure and define the material properties in the future steps of the analysis. The final phase of the geometry preparation is the extraction of the mid-surface for each component. This process provides the consistency geometry input for the loads and constraints definition in the later stages.

> In general, composite structures can be calculated using a few different approaches. In the basic strategy, the layer modelling technique is used, that using 2D or 3D finite elements, is dominant. In the presented method, the authors decided to use only 2D shell elements. This decision is since in thin-layer composites, the thickness is the smallest dimension, and the ratio to the other dimensions is quite large.

> The next step in the numerical model preparation process is a simplification of contact areas, especially where ribs or other elements are joined to outer structures. In case for

a perpendicular connection, the authors reduced the connection to a single component which contains layers of both base laminate and rib laminate. An example of the contact area simplification is presented in Figure 10.

In case of a parallel connection, simple nods joining is enough and there is no need to create a new component, as in the case of perpendicular connections. After this stage, the discrete model was ready to apply loads and constraints.

#### *3.3. Loads and Constraints Definition*

Additionally, to unify the load application process, all forces and momentums were applied to the center of mass, laying on the cord line. Additionally, to discretize the continuous load, the authors decided to use loads in selected points—mostly the joints where ribs connect with a spar. The idea behind the load application is presented in Figure 11. The values of such distributed loads are calculated from the aerodynamic forces and mass loads, based on different flight scenarios.

**Figure 11.** Loads and reactions in the wing section.
