Multi-Agent Model of a Sample Transport System for Modular In-Vitro Diagnostics Laboratories Lluís Ribas-Xirgo*, Antonio Miró-Vicenteº, Ismael F. Chaile*, and A. Josep Velasco-González* *Microelectronics and Electronic Syst. Dept. Universitat Autònoma de Barcelona (UAB) Campus UAB, 08193 Bellaterra, Spain {Lluis.Ribas, IsmaelFabricio.Chaile, Josep.Velasco}@uab.cat º BioSystems, SA Costa Brava 30, 08030 Barcelona, Spain AMiro@BioSystems.es Abstract Thousands of sample measurements for clinical analyses in large hospitals are made every day, which requires laboratories with modular analyzer systems that can operate with reliability and adaptability. In this paper we propose to transform a conventional, centrally controlled laboratory facility to a multi-agent system in order to distribute the control and make it more adaptable to priority samples and other unexpected events. The conversion requires replacing passive sample transport systems of laboratories by colonies of robots, which become the transportation agents. These multi-agent systems are completed by agents for the analyzers and the ones related to management software of laboratories. Result systems are compatible with the conventional ones so laboratories could be transformed in a straightforward way. 1. Introduction Analysis laboratories [1] in major hospitals operate with a continuous flow of samples, and can make thousands of daily measurements. The change of sample priorities because of the urgency of the arriving ones requires a dynamic planning module and a flexible transport system that can change the order of presentation of samples to the analyzers accordingly, thus optimizing performance. A typical organization of this type of labs includes several analyzers in chain, each performing a different class of tests on selected samples, which are placed in racks that running through the network of analyzers. Usually, samples must be processed several times in order to determine a number of parameters that conforms their full analyses. Some tests, as the hematocrit, use whole blood, i.e. before separating the cells in it, while others need the serum, urine, et cetera. This paper is focused on multi-analyzer systems that work with blood serum because most routine tests use it. Typically, for each sample corresponding to a patient a certain number of determinations must be made, each to be categorized according to the type of analyzer which is required for measurement. For example, the ions are measured with an ISE (Ion Selective Electrodes) unit, biochemical determinations or turbidimetry, as glucose or cholesterol, are determined with a photometric absorption analyzer, and clotting factors are determined by a coagulometry unit. These laboratory systems are built by combining a number of units of each type of analyzers, in accordance to the percentage of demand for each type of test and the processing speed of each analyzer. Often, biochemical parameters are routinely extracted from virtually 100% of the samples while coagulometry is only required in 10 to 15% of total samples. Consequently, each lab must be configured upon the needs of its owner. These facilities are managed by a central computer through LIMS (Laboratory Information Management System) software to which analysts introduce the lists of samples and the types of tests to be performed on each. These programs schedule the plant work and manage both the transport mechanisms as well as the analyzers, which report the results back to them. If any result is outside the range of linearity, the corresponding LIMS orders to repeat the test, after dilution. Only when all the results are correct, the LIMS generates output reports. These systems are very dependent on the computer that runs the LIMSs, and are quite hard to change or update once configured for a given application’s needs and requirements. Furthermore, as laboratories grow in size, their centralized control becomes more complex, particularly when considering the cycle time of high level control loops of large system and the trade-off between control speed and the number of events (faults, re-entrant samples, priority samples, et cetera) to handle. A relief to the complexity of a central controller is to distribute part of its functions to the systems’ components. In this paper we explore this approach to an extreme so to reduce to the minimum the central controller, which is only in charge of distributing data about the laboratory workload. With the advent of agent technologies, it is possible to devise a multi-agent system (MAS) that implements a biochemical analysis laboratory with little global control IEEE 17th Conference on Emerging Technologies and Factory Automation (ETFA) September 17-21, 2012, Krakow, Poland