This paper reports on our experiences in developing a conceptual architecture based on Intelligent Services within the context of ramp-up systems for aerospace manufacturing: a domain characterised by the con ict between the need for control and rigour and the reality of changes and problems a ecting cross-boundary systems. To alleviate these issues, we have decided to focus on combining agent technology with enterprise service bus, providing basis for real-time automatic negotiation, planning, scheduling and optimisation, intra and across factories. With this we push forward a new de nition of Intelligent Services.
Manufacturing has been characterised as a highly dynamic domain [9].
Information systems, in particular service-based systems, haven been used for
manufacturing support due to their characteristics for dealing with the dynamism
inherent in the domain, e.g. [
Our work focuses on small lot production management which is a type of manufacturing where there is a low-volume and high customisation of nal products. Examples of this are aerospace and shipbuilding. This production type involves a large number of parts to be assembled by a large number of people, potentially in separate geographical locations and dealing with a large number of suppliers. Problems arise when orders arrive requesting large numbers of products in a short period of time, stressing the limits of the production capacity.
The supplier network has to be planned and controlled in an appropriate way. The production of airplanes is often organised in several ow lines, each of them consisting of several stations. The assembly often takes place in geographically distributed manufacturing sites. Therefore, the information system for manufacturing becomes very important to help in planning and making decisions.
Let us exemplify this with a ramp-up scenario of a new airplane production. This is based on an industrial use case within the context of the ARUM project6. The scenario consists of three factories which have to communicate with each other in order to synchronise and optimise a production cycle, avoid delays and stops and so on.
The scenario starts when the factories receive instructions for producing a new aircraft batch. The production plan has to be distributed among the factories and each production stage has a clear deadline. The whole plan is arranged in such a way to meet the last deadline when the customer received the aircraft.
First, an initial scheduling of the production is performed. The operations are allocated to the factories (i.e. to the stations and workers at the shop oor) according to the speci ed production processes. The factories have to con rm the schedules before starting to work. At some point, factory A identi es that a machine is broken thus it cannot complete the scheduled operations in time. Therefore tasks have to be rescheduled and reallocated to workers in order to accommodate this change. The consequence of such a problem when not solved in a short time is a cascade of delays which require replanning of tasks not just days ahead but perhaps months. This demands for a solution that incorporates changes from the physical world at system runtime.
In either case other factories have to be noti ed of the reschedule in order to prepare for delays and reschedule their own operations. Such noti cations should be timely and involve several levels of the organisation: from administrative people to station managers and workers at the workshop stations. This is to accommodate 1) immediate changes to a worker's tasks, 2) changes to the next shifts and perhaps with a few days in advance, and 3) changes to the long term planning. This demands for a solution that takes into account communication between factories which are geographically distributed.
Reschedule across several factories is not an easy nor automated work.
Factories have to negotiate new deadlines, shift tasks and reallocate resources.
Sometimes negotiation is one-to-one between factories. Sometimes it is many-to-many.
And other times short-term decision making is mandatory. This demands for
near real-time automatic negotiation such as that provided by agent technology,
cf. [
Interoperability is yet another issue. Exchanging information in an electronic way between factories can be complicated due to information being stored in disparate formats from factory to factory, and even from engineering discipline to engineering discipline. To alleviate these issues, an integrated solution is necessary where information is lifted from external systems, e.g. ERP and MES (manufacturing execution system), converted to a standard format and incorporated into the decision support. Moreover, such solution should guarantee information delivery to whoever needs it. 6 http://www.arum-project.eu
Aerospace manufacturing is labour-intensive. Issues mentioned above arise yet people manage to overcome them, but at a high cost: an increased di erentiation between original plan versus actual execution which is translated in economic loss. 2
Customer needs have been increasingly diversi ed in small lot production
manufacturing. Thus a large variety of sophisticated products has to be managed in
current and future production sites to meet the needs. For example, aerospace
companies often develop a large number of di erent airplane models to
satisfy customers' requirements. In particular, small lot production results in
lowvolume, high customisation manufacturing where an e cient utilisation of labour
and machinery creates signi cant challenges [
A knowledge-base where this information is described is then needed;
4. These functionalities have to take into account the challenges of the
geographically distributed production. That is, the solution has to be present
at a factory yet be able to communicate seamlessly with another instance of
the solution in another factory;
5. Negotiation-based approaches for short-term, near real-time, decision
making are well known to be bene cial in these type of environments [
Traditional information systems for manufacturing and logistics like ERP and MES tend to be not su cient in this dynamic, highly complex environment with a large number of disturbances, especially for near real-time decision making. Although the MAS (multi-agent system) and SOA (service-oriented architecture) paradigms support most of the domain requirements, ensuring interoperability between MAS and SOA tends to be restricted by their implementation. 3
We propose a conceptual architecture for the support of small lot production management { Intelligent Enterprise Service-based Bus (iESB). The architecture is based on Intelligent Services de ned as independent pieces of software that are expected to provide a particular result; such result is either produced by the Intelligent Service itself or by requesting support from other Intelligent Services. An Intelligent Service is not committed to a single technology which would limit the focus and scope of the solutions to deal with dynamism. Instead it rather focuses on the conceptual features.
An Intelligent Service is di erent from an intelligent Web service [
Figure 1 depicts our architecture and internal components as Intelligent Services, namely an ESB as the central communication channel, gateway, process manager, planner, scheduler, peer-2-peer layer, ontology manager, service registry and node manager. We describe these Services below and indicate how they ful l the requirements in Sect. 2.
Enterprise Service Bus (ESB): An ESB acts as a key Service to alleviate communication and interoperability issues [10] among Intelligent Services, which ful ls requirement 6) from Sect. 2. Via the ESB, messages are sent by Services to each other. Services are also able to register an interest in message types in the ESB to get noti ed when events of their interest within the system occur.
Gateway: It functions as the interface for exchanging data with external data sources and systems such as ERP, MES, SCADA, etc, which satis es requirement 2). The Gateway not only pulls information but also pushes data into external sources as needed. The availability of information from external sources and systems is noti ed to registered Services via the ESB for potential consumption.
Process Manager: It satis es requirement 1), as it deals with the de nition and management of manufacturing processes within the factory as well as de nition and general allocation of resources such as sta , tools, machines, etc. Process instances are initiated by this Service according to how the user de nes the processes themselves.
Planner and Scheduler: These Services take process instances as per the Process Manager and optimise their execution according to available resources, timing, constraints, and unexpected events, as indicated by requirement 3). The di erence between the Planner and the Scheduler is the time scope: the Scheduler deals with imminent tasks and next shift tasks, whereas the Planner deals with tasks weeks and months in advance.
In both cases, unexpected events such as sudden increase of process instances,
o -sick sta , broken machines, etc., are taken into account for rescheduling and
adaptation. There may exist more than one of these Services depending on the
factory con guration and the di erent types of processes to handle. Further
details about the internal operation of these Services can be found in [
Peer-2-peer Layer: In order to deal with near real-time decision making in requirement 5), the Planners and Schedulers can bene t from a dedicated Peer2-peer Layer Service for handling negotiation communications without passing through the ESB. The advantage of this is the minimisation of message delivery time and negotiation turnaround. Notice that this Service only supports the Scheduler and Planner Services, however this does not limit them from using other available Services in the architecture.
Ontology Manager: An important characteristic of the proposed architecture is that it relies on ontologies for modelling a variety of concepts. A uni ed ontological approach for modelling domain knowledge, real-world entities, policies and operational work ows, provides the proposed architecture with several advantages. Two of the main ones, rstly it bene ts from formal and machineinterpretable semantics, semantic interoperability and consistency, as well as inference of knowledge not explicitly contained in the models. And secondly, concepts are treated in an integrated manner so that the model can e ectively capture their dependencies, while allowing for the enforcement of interdependencies at all phases of their lifecycle. With this Service we contribute to the ful lment of requirements 3) mainly, but 1) and 6) are also bene ted.
Node Manager: It establishes initial connectivity between any two architecture instances, which satis es requirement 4). That is, Fig. 1 can be instantiated in di erent geographical locations, e.g. di erent factories. Then the Node Manager allows di erent running instances to communicate and share Services from one instance to the other. This procedure e ectively converts each instance into a node of a system geographically distributed.
Service Registry: This Service keeps track of available Services at a node in the distributed system. It coordinates with the Node Manager for indicating availability of Services at a node on di erent, connected nodes. This Service contributes to requirement 6). 4
Our architecture supports of small lot production manufacturing, addressing the requirements described in Sect. 2. It deals with the unexpected events in the manufacturing process. To deal with the issues described in Sect. 1, three instances of the conceptual architecture can be used one at each factory. Message delivery is guaranteed thanks to the ESB. Near real-time negotiation can be achieved thanks to the dedicated peer-2-peer communication channel. In addition, the introduction of Intelligent Services without committing to a speci c technology allows us to see architecture components in a more abstract way, focusing on functionality rather than implementation constraints.
Committing a technology at design time, for instance Web services, the scenario would have been more reactive in the sense that Web services are stateless and thus they just respond to calls, lacking proactiveness. On the other hand, using MASs only renders the ESB pointless because agents could use their own communication channel. However, interoperability issues arise when agents do not follow a common communication protocol. Combining agents with Web services as an instantiation of the iESB then provides some exibility and minimises interoperability issues. Nevertheless, it all depends on how such combination is carried out, which might impose a combination ` avour' on the overall system.
We argue that all the components in the architecture are the minimum necessary to support and improve small lot production such as aerospace manufacturing. The main advantage the ESB brings to the approach is the homogenisation of communication between all components, thus reducing interoperability issues. Adding the rest of the components as Intelligent Services would aid software developers to maintain the focus on functionality as long as the ESB is used.
Currently our approach is conceptual, thus at this stage is di cult to
analyse any implementation issues. Within our project ARUM, we are starting the
development of the iESB and we are planning test for future evaluations, in
particular, performance comparisons with implementations where all components
are agents, Web services and their combinations. Thanks to the
conceptualisation of our approach we can apply it to other domains with similar requirements
(Sect. 2) such as business ecosystems [
This work is supported by the European Commission funded project ARUM: Adaptive Production Management. Grant agreement No. 314056. 8. Storch, R.L., Lim, S.: Improving Flow to Achieve Lean Manufacturing in Shipbuilding. Production Planning and Control 10(2), 127{137 (1999) 9. Verstraete, P., Germain, B.S., Valckenaers, P., Brussel, H.V., Belle, J.V., Hadeli, K.: Engineering manufacturing control systems using PROSA and delegate MAS.
International Journal of Agent-Oriented Software Engineering 2(1), 62{89 (2008) 10. Ziyaeva, G., Choi, E., Min, D.: Content-Based Intelligent Routing and Message Processing in Enterprise Service Bus. In: International Conference on Convergence and Hybrid Information Technology (ICHIT '08). pp. 245{249. IEEE (2008)