Les travaux présentés dans cette thèse concernent des problématiques d'architecture logicielle multi-domaines pour le développement d’applications distribuées. Ces applications sont caractérisées aujourd’hui comme des systèmes ouverts, complexes, hétérogènes et à large échelle. Les approches traditionnelles, telles que l’approche orienté objet, n’offrent plus un paradigme de conception suffisant pour appréhender la complexité de tels systèmes. Ces nouvelles tendances ont conduit à l’émergence d’approches de plus haut niveau telles que les approches orientées services, composants ou agents. Chacune de ces approches offrent des intérêts et des caractéristiques propres dans le développement d’applications distribuées. Les services offrent une abstraction et une interopérabilité à large échelle. Abstraction dans le sens où un service permet de spécifier un élément fonctionnel sans préciser comment cet élément est implémenté. Les composants sont une approche robuste basée sur la composition et la réutilisation d’éléments clairement définis par leurs interfaces. Les agents sont eux des éléments présentant un comportement dynamique dirigé par un but et des interactions de haut niveau avec les autres agents formant l’application, vue comme une organisation de services collaboratifs. D’un point de vue conceptuel, le service peut donc être perçu comme le modèle « métier » de l’application, alors que les composants et les agents constituent un modèle d’implémentation. L’étude de ces différents domaines et des modèles associés, a montré que les approches composants et agents sont complémentaires, les points forts d’une approche représentant les faiblesses de l’autre. Face à ce constat, il nous est paru intéressant d’intégrer ces deux approches, au sein d’une même démarche de conception. Cela permet, d’une part, qu’une approche puisse bénéficier des intérêts de l’autre et d’autre part, d’utiliser conjointement des agents et des composants dans la conception d'une même application. La démarche que nous avons adoptée consiste à considérer les services comme pivot d’interaction afin de rendre possible l’interopérabilité des agents et des composants. Pour supporter cette démarche, nous avons défini un processus de conception basé sur l’Ingénierie Des Modèles qui contient quatre modèles conceptuels (Domain Specific language) dont l’intérêt est de mettre l’accent sur les concepts de services et d’interaction. Nous avons ainsi défini un modèle de services, un modèle de composants et un modèle d’agents. Enfin, un modèle mixte appelé CASOM, Component Agent Service Oriented Model, permet de spécifier une application via une combinaison des trois domaines précédents. Ensuite, des règles de correspondances ont été définies entre les quatre modèles pour pouvoir par exemple transformer une spécification agents en une spécification composants ou mixte. L'implémentation de ces transformations a été réalisée en langage ATL (ATLAS Transformation Language). / The presented work considers problems related to multi-domain software architecture for the development of distributed applications. These applications are large-scaled, heterogeneous, open and complex software systems. Traditional approaches such as object-oriented are no longer sufficient to represent such complex systems. These trends lead to the emergence of higher-level approaches such as service-oriented, components or agents. Each one of these approaches offers interests and characteristics in the development of distributed applications. Services provide an abstraction and interoperability in a large scale. Abstraction is in the sense that a service can specify a functional element without specifying how this element is implemented. The components are a robust approach based on composition and reusability through their clearly defined interfaces. Agents are elements which are characterized by dynamic goal directed behaviours and high-level interactions with other agents forming the application, seen as an organization for collaborative services. From a conceptual point of view, the service can be seen as the “business” model of an application, while components and agents are the implementation models. The study of these different domains, with their related models, showed that the components and agents approaches are complementary; the strengths of one approach overcome the weaknesses of the other. Therefore, we are interested in the integration of these two approaches in a single design approach. This allows an approach to benefit from the interests of the other, on one hand and the use of agents and components jointly in the design of an application on the other hand. To reach our objective, we consider services as pivot of interaction between agents and components. The result of our analysis leads us to develop a design process based on Model-Driven Engineering which contains four conceptual models (Domain Specific Languages) with the main interest of focusing on the concepts of services and interaction. We then defined a service, component and agent models. Finally, a hybrid model called CASOM, Component Agent Service Oriented Model, was proposed that allows application specification via a combination of the three domains. Then, mapping rules have been defined between the four models in order to transform agents specification into components specification or mixed. The implementation of these transformations was done in ATL language (ATLAS Transformation Language).

The monitoring and control of low voltage distribution grids has historically been disregarded due to the unidirectional flow of power. However, nowadays the massive integration of distributed energy resources into distribution grids, such as solar photovoltaics, distributed storage, electric vehicles and demand response programs, presents some challenges. For instance, the unidirectional top-down power flow is being replaced by power flows in any direction: top-down and bottom-up. This paradigm shift adds extra regulatory, economic, and technical complexity for the Distribution System Operators (DSO). Thus to overcome the possible operational constraints, thermal limits, or voltage problems in the grid, an update of the existing electricity infrastructures is required. In response to this new situation, this thesis investigates the cost-effective communication and control architectures that are required for active low voltage grid monitoring and control applications, considering the regulatory constraints and the efficient utilization of the assets from a DSO’s perspective. The solutions include: i) optimal sensor placement configuration to perform low voltage state estimation, ii) optimal metering infrastructure designs for active low voltage monitoring applications, iii) coordinated control strategies to allow the integration of microgrid-like structures into the distribution grids, iv) optimal placement of actuators for operating the control strategies, v) a multiagent-based control solution for self-healing and feeder reconfiguration applications, and vi) a framework model and simulations to assess the reliability of the ICT infrastructure that enables the monitoring and control applications. As concluding remark, since the deployment of technology at low voltage grids is restricted to assets owned by the DSO, the operability of the grid is limited. This condition makes it so that the required communication and control enhancement solutions shall prioritize cost-effectiveness over comprehensiveness and complexity. Thus, the results from the presented studies show that it is essential to perform thorough cost-benefit analyses of the potential improvement solutions for each grid, because this will allow deploying the right technology only at the necessary locations. / <p>QC 20171106</p><p></p>

Los sistemas multi-agente (SMA) son sistemas distribuidos donde entidades autónomas llamadas agentes, ya sean humanos o software, persiguen sus propios objetivos. El paradigma de SMA ha sido propuesto como la aproximación de modelo apropiada para aplicaciones como el comercio electrónico, los sistemas multi-robot, aplicaciones de seguridad, etc. En la comunidad de SMA, la visión de sistemas multi-agente abiertos, donde agentes heterogéneos pueden entrar y salir del sistema dinámicamente, ha cobrado fuerza como paradigma de modelado debido a su relación conceptual con tecnologías como la Web, la computación grid, y las organizaciones virtuales. Debido a la heterogeneidad de los agentes, y al hecho de dirigirse por sus propios objetivos, el conflicto es un fenómeno candidato a aparecer en los sistemas multi-agente. En los últimos años, el término tecnologías del acuerdo ha sido usado para referirse a todos aquellos mecanismos que, directa o indirectamente, promueven la resolución de conflictos en sistemas computacionales como los sistemas multi-agente. Entre las tecnologías del acuerdo, la negociación automática ha sido propuesta como uno de los mecanismos clave en la resolución de conflictos debido a su uso análogo en la resolución de conflictos entre humanos. La negociación automática consiste en el intercambio automático de propuestas llevado a cabo por agentes software en nombre de sus usuarios. El objetivo final es conseguir un acuerdo con todas las partes involucradas. Pese a haber sido estudiada por la Inteligencia Artificial durante años, distintos problemas todavía no han sido resueltos por la comunidad científica todavía. El principal objetivo de esta tesis es proponer modelos de negociación para escenarios complejos donde la complejidad deriva de (1) las limitaciones computacionales o (ii) la necesidad de representar las preferencias de múltiples individuos. En la primera parte de esta tesis proponemos un modelo de negociación bilateral para el problema de / Sánchez Anguix, V. (2013). Complex negotiations in multi-agent systems [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/21570 / Palancia

[pt] A área da saúde desponta como um terreno fértil para a pesquisa científica em Tecnologia da Informação. As atividades de pesquisa na área permitem abordar diversas questões para promover o desenvolvimento tecnológico. Além disso, em relação aos aplicativos para dispositivos móveis, as principais lojas de aplicativos registraram um crescimento significativo no número de aplicativos de saúde disponíveis. Esses crescimentos encontram uma de suas causas na realidade pandêmica que enfrentamos desde o início de 2020. Para lidar com os diferentes desafios de saúde apresentados atualmente, o uso de sistemas multiagentes tem sido considerado uma boa abordagem para lidar, por exemplo, com distribuição, pró-atividade e autonomia dos sistemas. Embora várias plataformas conhecidas utilizem agentes de software, algumas delas não oferecem suporte adequado para o desenvolvimento de agentes, como a plataforma iOS. Assim, nesta dissertação propusemos um framework BDI para iOS que visa apoiar o desenvolvimento de aplicativos móveis de saúde com agentes de software. Esse framework é chamado de Swift Agent Development framework for health (SADE4Health) e usa recursos nativos do iOS, como o Healthkit, que é um repositório central de dados de saúde e condicionamento físico para acessar e compartilhar dados, mantendo a privacidade e o controle do usuário. Para mostrar como o framework suporta o desenvolvimento de novos aplicativos iOS com agentes de software, são explicados os passos mínimos necessários para criar um agente usando os recursos de saúde oferecidos pelo iOS, assim como um cenário de uso modelado a partir deles. Além disso, é apresentado um cenário de uso relacionado ao monitoramento remoto de sinais vitais de pacientes, ilustrando como desenvolver uma instância do framework proposto. Por fim, uma avaliação empírica com desenvolvedores iOS para medir a usabilidade do framework trouxe importantes achados. / [en] The healthcare area is emerging as a fertile ground for scientific research in Information Technology. Research activities in this field allow us to address several issues to promote technological development. In addition, regarding mobile device apps, the leading app stores registered significant growth in the number of available eHealth apps. One of the causes of this growth derives from the pandemic reality we have faced since the beginning of 2020. In order to confront the different health challenges presented currently, the use of multi-agent systems has been considered a good approach, dealing with, for example, distribution, pro-activity and autonomy of systems. Although several known platforms that use software agents, some of them do not offer appropriate support to develop agents, such as the iOS platform. Thus, in this dissertation we have proposed a BDI framework for iOS that aims to support the development of health mobile apps with software agents. This is known as Swift Agent Development Framework for Health (SADE4Health) and it uses native iOS resources, such as Healthkit, comprising a central repository for health and fitness data to access and share data while maintaining the user s privacy and control. To show how the framework supports the development of new iOS apps with software agents, the minimum necessary steps to create an agent using health features offered by iOS are explained, as well as a modeled use scenario based on it. Furthermore, a use scenario related to remote monitoring of patients vital signs that illustrates how to develop an instance of the proposed framework is presented. Finally, an empirical evaluation with iOS developers to measure the framework usability brought important findings.

In this thesis, we develop methods leading towards the distributed control of spatio-temporally constrained systems. Overall, we focus on two different approaches: a model predictive control approach and an approach based on ensuring set-invariance via control barrier functions. Developing a distributed control framework for spatio-temporally constrained systems is challenging since multiple subsystems are interconnected via time-varying state constraints. Often, such constraints are only implicitly given as logic formulas, for example in Signal Temporal Logic (STL). Our approach to dealing with spatio-temporal constraints is as follows. We aim at combining the computational efficiency of low-level feedback controllers with planning algorithms. Low-level feedback controllers shall ensure the satisfaction of parts of spatio-temporal constraints such as coupling state constraints or short term time-constraints. In contrast, planning algorithms account for those parts that require computationally intense planning operations. Powerful low-level controllers can simplify the planning task significantly. For this reason, the focus of this thesis is on the development of low level feedback controllers.  In the first part, we focus on handling coupling state constraints using a model predictive control (MPC) approach. Commonly, the distributed handling of coupling state constraints requires a sequential or iterative MPC scheme which however is computationally time-intense. We address this issue by employing consistency constraints to develop a parallelized distributed model predictive controller (DMPC). By using consistency constraints, each subsystem guarantees to its neighbors that its states stay within a particular neighborhood around a reference trajectory. Furthermore, we propose extensions to robust and iterative schemes. Building up on this, also systems with bounded dynamic couplings can be controlled. In the second part, we focus on methods for ensuring set-invariance. In particular, we focus on control barrier functions (CBF). We show how spatio-temporal constraints that comprise disjunctions (logic OR) can be encoded in non-smooth time-varying control barrier functions and how subgradients can be used to synthesize an efficient gradient-based controller. For these results, controllability assumptions must be invoked. To extend the results to systems with weaker controllability properties, we investigate the connection between controllability properties and the construction of CBFs. As a result, we propose a construction method for CBFs based on finite horizon predictions. The constructed CBF exhibits favorable properties for the extension of the previous results on encoding spatio-temporal constraints in CBFs to systems with weaker controllability properties. At last, we investigate with a case study how set-invariance methods can be used to implicitly coordinate systems subject to coupled state constraints. Our proposed method is fully decentralized and subsystems coordinate themselves purely via their actions and the adjustment of their individual constraints. In the end, we draw a conclusion and outline how the presented results contribute to the development of a distributed control framework for spatio-temporally constrained systems. / I den här avhandlingen utvecklar vi metoder som leder till distribuerad styrning av tillstånds-temporalt begränsade system. Vi följer två olika tillvägagångssätt: å ena sidan en modellprediktiv styrning och å andra sidan ett tillvägagångssätt som bygger på att säkerställa invarians i mängden via kontrollbarriärfunktioner. Det är en utmaning att utveckla ett ramverk för distribuerad styrning för tillstånds-temporalt begränsade system, eftersom flera delsystem är sammankopplade via sina tillståndsbegränsningar som varierar över tiden. Ofta ges sådana begränsningar endast implicit som logiska formler, till exempel i Signal Temporal Logic (STL).  Vår metod för att hantera tillstånds- och tidsmässiga begränsningar är följande. Vi strävar efter att kombinera beräkningseffektiviteten hos återkopplingsregulatorer på låg nivå med planeringsalgoritmer. Återkopplingsregulatorer på låg nivå skall säkerställa att delar av de tillstånds- och tidsmässiga begränsningarna uppfylls, t.ex. sammankopplande tillståndsbegränsningar eller kortsiktiga tidsbegränsningar, medan planeringsalgoritmerna tar hänsyn till de delar som kräver beräkningsintensiva planeringsoperationer. Kraftfulla styrsystem på låg nivå kan förenkla planeringsuppgiften avsevärt. Därför fokuserar vi i denna avhandlingen på utvecklingen av återkopplingsregulatorer på låg nivå.  I den första delen fokuserar vi på att hantera sammankopplande tillståndsbegränsningar för distribuerade system med hjälp av en modell prediktiv styrning (MPC). Vanligtvis kräver den distribuerade hanteringen av kopplingsbegränsningar ett sekventiellt eller iterativt MPC-system som dock är tidskrävande. Därför utvecklar vi en parallelliserad distribuerad modell prediktiv styrning (DMPC) baserad på konsistensbegränsningar. Därigenom garanterar ett delsystem till sina grannar att det håller sig inom ett visst område runt en referensbana. Den generiska formuleringen av vårt DMPC-system möjliggör flera realiseringar. En särskild realisering föreslås. Dessutom utvecklas utvidgningar till ett robust och iterativt system samt ett DMPC-system för system med begränsade dynamiska kopplingar. I den andra delen fokuserar vi på metoder för att säkerställa invariansen av mängder. Vi fokuserar särskilt på kontrollbarriärfunktioner (CBF). Vi visar hur tillstånds- och tidsmässiga begränsningar kan inkodas i icke-glatta tidsvarierande kontrollbarriärfunktioner och hur subgradienter kan användas för att konstruera en effektiv gradientbaserad styrning. För dessa resultat måste antaganden om kontrollerbarhet åberopas. För att utvidga detta resultat till system med svagare kontrollerbarhetsegenskaper undersöker vi kopplingen mellan dynamiska systems kontrollerbarhetsegenskaper och konstruktionen av en CBF. Som ett resultat av detta föreslår vi en konstruktionsmetod för CBF:er som bygger på förutsägelser för ändliga horisonter. Den konstruerade CBF:n uppvisar gynnsamma egenskaper för att utvidga det tidigare resultatet om kodning av rums-temporala begränsningar i CBF:er till system med svagare kontrollerbarhetsegenskaper. Slutligen undersöker vi med hjälp av en fallstudie hur metoder för att säkerställa invariansen av mängder kan användas för att implicit samordna system som är kopplade via tillståndsbegränsningar. Vår föreslagna metod är helt decentraliserad och delsystemen samordnar sig själva endast via sina handlingar och justeringen av sina individuella begränsningar. Slutligen drar vi en slutsats och beskriver hur de presenterade resultaten bidrar till utvecklingen av ett ramverk för distribuerad styrning av tillstånds- och tidsmässigt begränsade system. / <p>QC 20230520</p>

[pt] O uso de sistemas baseados em agentes é adequado à construção de software complexo. Para garantir uma ordem social desejável é preciso lidar com a autonomia e a diversidade de interesses dos agentes e as normas são mecanismos eficientes de controle usados para regular o comportamento dos agentes. O teste de software continua sendo amplamente aplicado para garantir a qualidade do software. No contexto de sistemas multiagentes normativos, os testes devem lidar com a necessidade dos agentes atuarem de forma robusta sob condições normativas dinâmicas as quais os desenvolvedores não consideraram. Neste contexto, propomos uma abordagem para testar agentes normativos que seguem o modelo belief-desire-intention. Como contribuições, esta tese apresenta: um framework para desenvolvimento de agentes BDI normativos, o NBDI4JADE; um modelo modelo de faltas para apoiar a identificação dos diferentes tipos de falhas em agentes normativos; um framework para testar agentes BDI normativos, o N-JAT4BDI e, um método para geração de casos de teste a partir de modelos ANA-ML dos agentes. Avaliamos o framework de teste através de um estudo experimental no qual discutimos a eficiência e a eficácia do framework. Avaliamos, também, a eficácia do método de geração de casos de teste,aplicando-o a dois cenários de uso: (i) um sistema para gerenciar a submissão e o processo de revisão de artigos em conferências, e (ii) um sistema de venda de pacotes turísticos de uma agência de viagens. Os resultados obtidos nas avaliações de ambos os frameworks apresentam indícios positivos da eficiência e eficácia na detecção e identificação de falhas em agentes normativos e eficiência na geração dos casos de teste. / [en] The use of agent-based systems is suitable for the construction of complex software. To guarantee a desirable social order one we must deal with the autonomy and diversity of interests of the agents and norms are efficient mechanisms of control used to regulate the behavior of the agents. Software testing still have been widely applied to ensure the software quality. In the context of normative multiagent systems, the test must deal with the need for the agents to act robustly under dynamic normative conditions in which the developers did not consider. In this context, we propose an approach to test normative agents that follow the belief-desire-intention model. Among the contributions, this thesis presents: a framework for development of normative BDI agents, the NBDI4JADE; a fault model to support the identification of different types of failures in normative agents; a framework for testing normative BDI agents, the N-JAT4BDI, and a method for generating test cases from ANA-ML models.We evaluated the test framework through an experimental study in which we discussed the efficiency and effectiveness of the framework. We also evaluate the effectiveness of the test case generation method, applying it to two usage scenarios: (i) a system to the manage submission and the review process of articles in conferences, and (ii) a system of sale of tour packages from a travel agency. The results obtained in the evaluations of both frameworks presented positive indications of efficiency and effectiveness in the detection and identification of failures in normative agents and efficiency in the generation of test cases.

<p dir="ltr">Learning, planning, and representing knowledge at multiple levels of temporal abstractions provide an agent with the ability to predict consequences of different courses of actions, which is essential for improving the performance of sequential decision making. However, discovering effective temporal abstractions, which the agent can use as skills, and adopting the constructed temporal abstractions for efficient policy learning can be challenging. Despite significant advancements in single-agent settings, temporal abstractions in multi-agent systems remains underexplored. This thesis addresses this research gap by introducing novel algorithms for discovering and employing temporal abstractions in both cooperative and competitive multi-agent environments. We first develop an unsupervised spectral-analysis-based discovery algorithm, aiming at finding temporal abstractions that can enhance the joint exploration of agents in complex, unknown environments for goal-achieving tasks. Subsequently, we propose a variational method that is applicable for a broader range of collaborative multi-agent tasks. This method unifies dynamic grouping and automatic multi-agent temporal abstraction discovery, and can be seamlessly integrated into the commonly-used multi-agent reinforcement learning algorithms. Further, for competitive multi-agent zero-sum games, we develop an algorithm based on Counterfactual Regret Minimization, which enables agents to form and utilize strategic abstractions akin to routine moves in chess during strategy learning, supported by solid theoretical and empirical analyses. Collectively, these contributions not only advance the understanding of multi-agent temporal abstractions but also present practical algorithms for intricate multi-agent challenges, including control, planning, and decision-making in complex scenarios.</p>

<p dir="ltr">We present a framework for Trustworthy Artificial Intelligence (TAI) that dynamically assesses trust and scrutinizes past decision-making, aiming to identify both individual and community behavior. The modeling of behavior incorporates proposed concepts, namely trust pressure and trust sensitivity, laying the foundation for predicting future decision-making regarding community behavior, consensus level, and decision-making duration. Our framework involves the development and mathematical modeling of trust pressure and trust sensitivity, drawing on social validation theory within the context of environmental decision-making. To substantiate our approach, we conduct experiments encompassing (i) dynamic trust sensitivity to reveal the impact of learning actors between decision-making, (ii) multi-level trust measurements to capture disruptive ratings, and (iii) different distributions of trust sensitivity to emphasize the significance of individual progress as well as overall progress.</p><p dir="ltr">Additionally, we introduce TAI metrics, trustworthy acceptance, and trustworthy fairness, designed to evaluate the acceptance of decisions proposed by AI or humans and the fairness of such proposed decisions. The dynamic trust management within the framework allows these TAI metrics to discern support for decisions among individuals with varying levels of trust. We propose both the metrics and their measurement methodology as contributions to the standardization of trustworthy AI.</p><p dir="ltr">Furthermore, our trustability metric incorporates reliability, resilience, and trust to evaluate systems with multiple components. We illustrate experiments showcasing the effects of different trust declines on the overall trustability of the system. Notably, we depict the trade-off between trustability and cost, resulting in net utility, which facilitates decision-making in systems and cloud security. This represents a pivotal step toward an artificial control model involving multiple agents engaged in negotiation.</p><p dir="ltr">Lastly, the dynamic management of trust and trustworthy acceptance, particularly in varying criteria, serves as a foundation for causal AI by providing inference methods. We outline a mechanism and present an experiment on human-driven causal inference, where participant discussions act as interventions, enabling counterfactual evaluations once actor and community behavior are modeled.</p>

<p dir="ltr">This dissertation presents decentralized and agent-based solutions for organizing machine learning resources, such as datasets and learning models. It aims to democratize the analysis of these resources through a simple yet flexible query structure, automate common ML tasks such as training, testing, model selection, and hyperparameter tuning, and enable privacy-centric building of ML models over distributed datasets. Based on networked multi-agent systems, the proposed approach represents ML resources as autonomous and self-reliant entities. This representation makes the resources easily movable, scalable, and independent of geographical locations, alleviating the need for centralized control and management units. Additionally, as all machine learning and data mining tasks are conducted near their resources, providers can apply customized rules independently of other parts of the system. </p><p><br></p>

Thesis (M. Tech.) -- Central University of Technology, Free State, 2010 / Traditional corrective maintenance is both costly and ineffective. In some situations it is more cost effective to replace a device than to maintain it; however it is far more likely that the cost of the device far outweighs the cost of performing routine maintenance. These device related costs coupled with the profit loss due to reduced production levels, makes this reactive maintenance approach unacceptably inefficient in many situations. Blind predictive maintenance without considering the actual physical state of the hardware is an improvement, but is still far from ideal. Simply maintaining devices on a schedule without taking into account the operational hours and workload can be a costly mistake. The inefficiencies associated with these approaches have contributed to the development of proactive maintenance strategies. These approaches take the device health state into account. For this reason, proactive maintenance strategies are inherently more efficient compared to the aforementioned traditional approaches. Predicting the health degradation of devices allows for easier anticipation of the required maintenance resources and costs. Maintenance can also be scheduled to accommodate production needs. This work represents the design and simulation of an intelligent maintenance management system that incorporates device health prognosis with maintenance schedule generation. The simulation scenario provided prognostic data to be used to schedule devices for maintenance. A production rule engine was provided with a feasible starting schedule. This schedule was then improved and the process was determined by adhering to a set of criteria. Benchmarks were conducted to show the benefit of optimising the starting schedule and the results were presented as proof. Improving on existing maintenance approaches will result in several benefits for an organisation. Eliminating the need to address unexpected failures or perform maintenance prematurely will ensure that the relevant resources are available when they are required. This will in turn reduce the expenditure related to wasted maintenance resources without compromising the health of devices or systems in the organisation.