Daniele Scrimieri

An Agent Based Framework to Support Plug And Produce

The new market trends are very different, so it is crucial to the companies improve the tools and capabilities that allow themselves readjust rapidly and effectively to the news market changes and to the new requirements. In order to facilitate this process, it is proposed in this paper an agent based implementation that can provide to the existent systems the capacity to quickly adapt and reconfigure using standard technology. The proposed framework provides an intelligent tool to autonomously help the configuration when a production operator pretends to introduce a new variant of product in runtime and consult important information provided by the system to monitor execution.

A multi-agent architecture for plug and produce on an industrial assembly platform

Modern manufacturing companies face increased pressures to adapt to shorter product life cycles and the need to reconfigure more frequently their production systems to offer new product variants. This paper proposes a new multi-agent architecture utilising “plug and produce” principles for configuration and reconfiguration of production systems with minimum human intervention. A new decision-making approach for system reconfiguration based on tasks re-allocation is presented using goal driven methods. The application of the proposed architecture is described with a number of architectural views and its deployment is illustrated using a validation scenario implemented on an industrial assembly platform. The proposed methodology provides an innovative application of a multi-agent control environment and architecture with the objective of significantly reducing the time for deployment and ramp-up of small footprint assembly systems.

Accelerated Ramp-Up of Assembly Systems through Self-learning

The ramp-up process of assembly systems has a huge impact on both the productivity of those systems and the quality of the output. In this work we present a new technique for accelerating the ramp-up process by automatically capturing knowledge about a machine and subsequently reusing it to inform an engineer performing ramp-up. This technique relies on a novel process called the Knowledge Object Algorithm. The technique is explained and demonstrated using synthetic data, designed to emulate a typical use case of such a system. The future direction for this work is also outlined and further experiments detailed.

Learning and reuse of engineering ramp-up strategies for modular assembly systems

We present a decision-support framework for speeding up the ramp-up of modular assembly systems by learning from past experience. Bringing an assembly system to the expected level of productivity requires engineers performing mechanical adjustments and changes to the assembly process to improve the performance. This activity is time-consuming, knowledge-intensive and highly dependent on the skills of the engineers. Learning the ramp-up process has shown to be effective for making progress faster. Our approach consists of automatically capturing information about the changes made by an operator dealing with disturbances, relating them to the modular structure of the machine and evaluating the resulting system state by analysing sensor data. The feedback thus obtained on applied adaptations is used to derive recommendations in similar contexts. Recommendations are generated with a variant of the k-nearest neighbour algorithm through searching in a multidimensional space containing previous system states. Applications of the framework include knowledge transfer among operators and machines with overlapping structure and functionality. The application of our method in a case study is discussed.

An in-core grid index for transferring finite element data across dissimilar meshes

An indexing technique for mapping finite element data is presented.The underlying space of an indexed mesh is decomposed into variable-sized cells.Creation of an index and search for nodes and elements are fast.An experimental evaluation of mapping techniques using the index is conducted.The algorithms have been implemented in FEDES (Finite Element Data Exchange System). The simulation of a manufacturing process chain with the finite element method requires the selection of an appropriate finite element solver, element type and mesh density for each process of the chain. When the simulation results of one step are needed in a subsequent one, they have to be interpolated and transferred to another model. This paper presents an in-core grid index that can be created on a mesh represented by a list of nodes/elements. Finite element data can thus be transferred across different models in a process chain by mapping nodes or elements in indexed meshes. For each nodal or integration point of the target mesh, the index on the source mesh is searched for a specific node or element satisfying certain conditions, based on the mapping method. The underlying space of an indexed mesh is decomposed into a grid of variable-sized cells. The index allows local searches to be performed in a small subset of the cells, instead of linear searches in the entire mesh which are computationally expensive. This work focuses on the implementation and computational efficiency of indexing, searching and mapping. An experimental evaluation on medium-sized meshes suggests that the combination of index creation and mapping using the index is much faster than mapping through sequential searches.

Fast mapping of finite element field variables between meshes with different densities and element types

In the simulation of a chain of manufacturing processes, several finite element packages can be employed and for each process or package a different mesh density or element type may be the most suitable. Therefore, there is a need for transferring finite element analysis (FEA) data among packages and mapping it between meshes. This paper presents efficient algorithms for mapping FEA data between meshes with different densities and element types. An in-core spatial index is created on the mesh from which FEA data is transferred. The index is represented by a dynamic grid partitioning the underlying space from which nodes and elements are drawn into equal-sized cells. Buckets containing references to the nodes indexed are associated with the cells in a many-to-one correspondence. Such an index makes nearest neighbour searches of nodes and elements much faster than sequential scans. An experimental evaluation of the mapping techniques using the index is conducted. The algorithms have been implemented in the open source finite element data exchange system FEDES.

Capture and Application of Adaptation Knowledge on Assembly Stations

Dealing with disruptions affecting an assembly station is a complex process which requires deep knowledge of the assembly process, the product being assembled and the technologies adopted. We present a method for automatically acquiring and reusing operational knowledge on how to react to disruptions and restore productivity. Shop floor operators typically perform a series of adjustments by trial and error until the performance targets are achieved. These adaptations are captured and their effect on the station is measured. Context-dependent adaptation knowledge is then derived to inform the operator about potential adaptations whenever the station re-enters a state similar to one in which adaptation information was captured.

A Multi-Agent System Architecture for Self-configuration

Due to constant globalisation new trends on the market are coming up. One of the trends is the customisation of products for the customer and shorter product life cycles. To overcome the trends industries identified as key element self-reconfigurable production systems. A change to a running system means loss of time, money and manpower. A reconfigurable production system can automatically adapt to changes in terms of changing a machine or a product. The methodology behind is adapted from the office world and is called plug and produce. However, a production system has different requirements which need to be met. Due to a lack of homogeneity of industrial controllers in terms of communication and reconfigurability, as well as the interaction with the end user, the multi-agent technology was identified as a superior communicator. We present a new multilayered multi-agent architecture where the necessary agent types are introduced to fulfil the requirements for plug and produce. One scenario is shown where the architecture is employed to enable plug and produce capabilities and allow the system to adapt itself.

Automated experience-based learning for plug and produce assembly systems

This paper presents a self-learning technique for adapting modular automated assembly systems. The technique consists of automatically analysing sensor data and acquiring experience on the changes made on an assembly system to cope with new production requirements or to recover from disruptions. Experience is generalised into operational knowledge that is used to aid engineers in future adaptations by guiding them throughout the process. At each step, applicable changes are presented and ranked based on: (1) similarity between the current context and those in the experience base; (2) estimate of the impact on system performance. The experience model and the self-learning technique reflect the modular structure of the assembly machine and are particularly suitable for plug and produce systems, which are designed to offer high levels of self-organisation and adaptability. Adaptations can be performed and evaluated at different levels: from the smallest pluggable unit to the whole assembly system. Knowledge on individual modules can be reused when modules are plugged into other systems. An experimental evaluation has been conducted on an industrial case study and the results show that, with experience-based learning, adaptations of plug and produce systems can be performed in a shorter time.