Negotiation and cooperation in multi-agent systems

Abstract

The Hoist Scheduling Problem (HSP) is directly related to the management of chemical processing lines, also known as electro-plating lines, such as those that can be found in the PCB industry for instance. Such lines are composed of several tanks containing the chemicals, a rail that runs above the tanks and multiple hoists, moving on the mentioned rail, that must carry jobs from tank to tank. Jobs must go through the line according to a given production sequence (also called a routing). All jobs do not necessarily have the same production sequence. A production sequence gives not only the tanks sequence, but also the treatment times (as an interval). The problem is to appropriately schedule the tanks and hoists to minimize the number of defective jobs while maximizing the throughput. This thesis proposes a layered system for solving the HSP as three sub-problems, namely the Input Date Decision Problem (IDDP), the Hoists Assignment Problem (HAP) and the Hoists Conflict Resolution Problem (HCRP) . Each layer uses multiple agents with mixed cooperative and competitive behaviors. Agents represent resources such as the tanks or the hoists. The first layer is concerned with the resolution of the IDDP for new jobs. The second layer is concerned with the assignment of transfer operations to hoists. The third and last layer is designed to ensure the integrity of the line by providing mechanisms to resolve remaining conflicts between physical entities of the production line. The first layer is based on a democratic algorithm (DA), which is a contribution of this thesis, that is used as a method for cooperative decision-making. This democratic algorithm is based on a repetitive process “proposal-making → vote ”. These two steps are described and a few instances of the DA are suggested. Since no predictive scheduling of the allocations of hoists to transfer operation, or tanks to processing operations, is performed, the representation of the tanks occupation as well as t...