Multi-paradigm modeling settings inherently and frequently cause inconsistencies between models in dierent languages involved in the design. The proper management of inconsistencies starts with choosing the appropriate formalisms to characterize inconsistent situations. In this paper, we introduce a formalism, which allows reasoning about inconsistencies in the context of the design process. We support our ideas by examples from a case study on the design of an unmanned aerial vehicle.
The design and evolution of mechatronic and cyber-physical systems involves
multiple stakeholders with dierent views on the system as a whole. Because
of overlapping elements and properties between the various views and models,
there is a need to manage the dependencies and possible inconsistencies between
these models [
The process of inconsistency management (ICM) breaks down into three main
activities. First, overlap between dierent models is characterized ; based on this
information, inconsistencies can be detected and consequently resolved. Dierent
authors approach this process with various granularity and see the role of
characterization (often referred as inconsistency denition ) dierent. Spanoudakis et
al. do not emphasize the role of proper inconsistency characterization [
In this paper, we present a novel formalism to characterize dependencies among dierent models. We argue, that because inconsistencies arise from models being used in complex design processes, their characterization is to be approached from a process oriented point of view. This approach considers not just static relations among models, but also the dynamics of the process. Although we do not explicitly investigate the subsequent activities of detection and resolution, we give specic directions on how these should be addressed to exploit the full potential of our formalism.
Our work is motivated by the specic nature of a multi-paradigm settings in
the mechatronic domain. Although ICM is an adequately studied eld in
software engineering, mechatronic design introduces challenges that make software
focused ICM techniques inecient in mechatronics. This is due to the
involvement of models of physics , such as models of nite element methods (FEM) or
computational uid dynamics (CFD). Typically, ICM techniques targeting pure
software systems, even if being situated in a multi-model/multi-paradigm
setting, focus on syntactic (structural) inconsistencies arising from the involvement
of detached linguistic meta-models [
In a design process with physics involved, however, semantic inconsistencies,
arising from semantic overlaps of various models, have a signicant impact as
well [
The rest of the paper is organized as follows. In Section 2 we motivate our work using an example scenario from a case study on the design of an unmanned aerial vehicle (UAV), and present state-of-the-art techniques targeting dierent aspects of the scenario. In Section 3, we introduce the core ideas of our approach. Section 4 discusses the related work. Finally, Section 5 briey discusses the results and the planned future directions. 2
Process and dependency modeling: the state of the art In this section, we motivate our work by an example scenario and present the state-of-the-art solutions tackling well-dened parts of it. 2.1
We illustrate our approach using an example from a case study on the design of an unmanned aerial vehicle (UAV). In the example scenario, the wings of the UAV are designed using models in various formalisms. The aim is to formalize the relations among the properties derived from these models either by syntactic or semantic links (e.g. by simulation).
The design process is sequential and it starts with modeling the overall geometry using a CAD tool. After validating the model against the requirements, a nite element method is used to analyse the system. Dierent physical properties are predicted using a geometric and nite element analysis, for example the mass of the UAV. The design team than creates a plant model of the UAV using a lumped parameter model. This plant model is computationally less intensive and ideal for creating a control model for the UAV. Finally, a control model is designed.
By explicit modeling of the design process, valuable information can be gained on data- and control ows along the elementary design activities. Additionally, by explicitly modeling dependencies, inconsistencies among models in the process can be identied. The state-of-the-art addresses these two challenges separately. We argue, that combining these techniques yield signicant advantages in managing model inconsistencies. 2.2
The Formalism Transformation Graph and Process Model (FTG+PM)
formalism [
The work interprets
semantic overlaps as
dependencies among elements of
dierent models and captures these
in a correspondence model,
called the dependency model
(DM). Figure 2 shows such
a DM. Synthesis properties
(SP) and analysis
properties (AP) are distinguished
to capture wether properties
represent a system alternative
at design time, or a result of Fig. 2: A dependency model as presented in [
In this paper, we do not distinguish between synthesis and analysis properties and dependencies, as the formalism presented later is applicable for each of these. 3
Our approach combines the ideas presented in Section 2.2 and Section 2.3. First, we generalize and extend the dependency modeling technique and redene it in the context of processes modelled using the FTG+PM formalism. 3.1
The nature of model dependencies in design processes is often well-known for participating stakeholders. In fact, this knowledge often goes beyond the level of simple inuence-like relationships: dependencies encompassing sensitivity and exact mathematical relationships are also common in practice because of domain knowledge and experience.
Explicit modeling of this information enables more complex formal
reasoning about model dependencies as compared to [
L2: sensitivity. In engineering practice, it is a common scenario that not
only the semantic relationship is known among various model elements, but
also the sensitivity. By simulation and analysis, further knowledge can be
gained on sensitivity information [
L3: exact mathematical relationship. The highest level of precision is a mathematical relationship of the depending model elements. We foresee algebraic relationships being the typical examples here. For example, in Section 2.1, the mass of the UAV can be derived as the mass of all the model elements featuring a mass property, i.e. the airframe, the wings and other hardware. Other types of mathematical relationships, such as dierential equations can also be used as an L3 relationship. 3.2
A metamodel for modeling dependencies with levels of precision
To enable modeling with levels of precision, we propose a metamodel shown in
Figure 3. The metamodel builds on [
Figure 4 shows the process model of the example scenario, extended by dependency information. Three dependency properties between the language of geometrical properties and the language of physical properties are depicted using directed edges of L1 L3 levels of precision. (Every directed dependency edge is typed in the FTG, which is not presented hereby.) For example, there is an L3 relationship between the mass of the wing and the mass considered in the control model, in this direction. This dependency link allows propagating changes from the geometrical model through an exact mathematical relationship to the control model. In the other direction, however there is only an L1 relationship, meaning if the mass in the control model is changed, the changes cannot be propagated directly to the geometry model, but there is an indication that they should be.
The generality of this approach is shown using Figure 5. The gure property c property d shows the relation between properties dependency of submodels. By using slicing tech- property a property b niques, a relation is established be- has tween a part of the model and a certain property. For example, in our ex- submodel B’ ample we could be interested in only submodel A’ slice the mass of the wing of the UAV. By model A model B slicing the model to only contain the wing, we can use the geometrical and Fig. 5: Properties related to submodels material information to establish a re- by using slicing. lation with only the wing segment of the CAD model. The slicing can also occur during the analysis step. This leaves the relation between a property and the submodel implicit. For example, because the analysis is black-box. Leaving the relation between the submodel and the property implicit could have an inuence on how resolution strategies can cope with an inconsistency. 3.4
We show two typical uses of our technique to motivate its usefulness in complex design processes.
Inconsistency characterization, detection and resolution Our approach
is situated in a conceptual inconsistency management framework as an
important formalism to build inconsistency management processes upon. Since the
dependency model of our approach constitutes a graph (more specically: a
hypergraph), characterizing inconsistencies as graph patterns over the graph of
dependencies, seems to be a natural t. Additionally, graph patterns can be
evaluated at run-time very eciently and therefore, it gives a well-performing
technique to detect inconsistencies [
Qamar et al. use L1 dependencies to identify potentially inconsistent states [
L2 dependencies extend this approach by giving additional hints on the magnitude of inconsistency in the output property. Therefore, L2 dependencies enable reasoning about the quality of the process topology and optimize it for consistency. (Section 3.4)
Since L3 dependencies extend the former two by dening an exact mathematical relationship among model elements, they enable a semi-automated detection of inconsistencies by continuously evaluating these relationships. Process optimization Dependency information can help restructure processes into more ecient alternative topologies.
A typical example of such an optimization can be making sequential design processes parallel in order to speed up processes, by identifying activities having no data- or control dependency, but featuring explicitly modelled semantic dependencies. Apart from the former two constraints, the semantic dependency model ensures inconsistencies will be managed over parallel branches as well. Techniques based on change propagation, incremental synchronization typically focus on sequential sub-process topologies and therefore, they fall short to tackle inconsistencies in parallel settings. L1 L3 dependencies all support ICM over parallel branches, although the eciency of ICM is determined by the level. Figure 6 shows the parallel equivalent of the PM in Figure 4. concept:DesignCon
cept :DesignConcept concept:Physical
PropertyLanguage :GetPpheyrtsieicsalProssttiiffffnneessDDeepp::LLm31assDep:L3ccoonnttrroollSSuurrffaacceeDDeepp::LL31
massDep:L1
L2 relationships can help further optimizing processes for decreasing the amount
of inconsistencies encountered, e.g. by moving high-impact activities in the earlier
phases of the process and therefore, potentially reducing the amount of required
re-iterations. The lifting of design choices to a higher level of abstraction to allow
for parallel design also hints towards the use of viewpoint contracts in the design
process, similar to Toerngren et al [
Finally, organizing processes in a topology where higher level dependencies point forward in the process increases the eciency of ICM techniques based on change propagation. 4
Characterization of model inconsistencies has been approached using various
techniques, such as description logic based [
Triple graph grammars (TGG) have been widely used as a formal
underpinning to model-based ICM techniques [
Design structure matrices (DSM) [
Herzig et al. provides support for automating the task of inconsistency
characterization by exploring semantic relationships using similarity metrics [
Qamar et al dene ve levels of detail of modeling dependencies [
In this paper, we introduced a novel formalism for characterizing and detecting
inconsistencies among design models. Our work was motivated by the specic
nature of multi-model settings in the mechatronic domain. The approach extends
the dependency modeling technique of [
The main advantage of our work is that inconsistencies can be characterized using a graph-like structure, which enables using state-of-the-art graph querying tools to evaluate (or enforce) model consistency.
As a future work, we plan to further evolve the formalism by including authorization and ownership models of design processes and therefore, enable reasoning about the limits of automation of inconsistency resolution techniques and including explicitly modelled manual steps in the resolution process. Acknowledgement This work has been carried out within the framework of the MBSE4Mechatronics project (grant nr. 130013) of the agency for Innovation by Science and Technology in Flanders (IWT-Vlaanderen).
The authors wish to thank the insightful comments of Johan Vanhuyse, Tuur Benoit, Klaas Gadeyne and Maarten Witters.