A Neuro-Fuzzy-Regression Algorithm for Improved Prediction of Manufacturing Lead Time with Machine Breakdowns

Abstract

Manufacturing lead time estimation is an important task in production system with machine breakdown and maintenance. This study presents a flexible algorithm for estimation and forecasting lead time based on artificial neural network (ANN), fuzzy regression (FR), and conventional regression (CR). First, an ANN is illustrated based on supervised multi-layer perceptron network for the lead time forecasting. The selected ANN model is then compared with fuzzy and conventional regression models with respect to Mean Absolute Percentage Error, hence the name neuro-fuzzy regression algorithm. To show the applicability and superiority of the flexible neuro-fuzzy regression, the proposed algorithm is used to estimate the weekly lead times of an actual assembly shop (producer of heavy electric motor). This is the first study that introduces a flexible neuro-fuzzy algorithm for improved estimation of lead time in manufacturing systems with machine breakdown and maintenance. In addition to accuracy, simplicity and short execution time of lead time estimation are desirable features of the presented flexible algorithm.