Use of Intelligent Techniques for Throughput Estimation of Unreliable two-machine Production Lines with Random Processing Times: Preliminary Results.

In this paper, an application of intelligent techniques for modeling production systems is proposed. Such problems are usually highly complex so for most types of production systems, no accurate general formulas exist in the literature for the calculation of throughput. The specific problem studied in this paper is that of an unreliable two-machine production line with random processing times. Training data corresponding to specific configurations of the production line are produced from solving the system analytically with the use of stationary probabilities using the Markovian model. These training data are used as input for training (a) a neural network and (b) a genetic programming approach, trying to induce a generalized model of high accuracy (i.e., a network structure), or high comprehensibility (i.e., closed-form formula), respectively. Results show that the applied intelligent techniques succeed to obtain both, highly accurate approximation, and highly comprehensible outcomes, for out-of-sample test data, taken from the same range of production parameters as the training data range. The paper contains only encouraging preliminary results. Experiments are still in progress, regarding either results concerning out-of-range data for testing the generalization ability of the methods used, or results regarding the performance of intelligent approaches for larger production lines of similar type.