On the optimization of material usage in power transformer manufacturing

The manufacturing process of a power transformer core constitutes a highly complex optimization problem. It involves the optimal slitting of a set of available metal coils into bands of desirable width. The optimization is a generalization of the so-called cutting-stock problem, originally encountered in the paper industry. It usually addresses the minimization of the metal scrap or the minimization of the slit coils. Recently, though, stricter noise control regulations and client specifications have complicated further the optimization problem, increasing the number of constraints and the potential objectives. Due to the increased number of nonlinear constraints and objectives, traditional approaches for addressing cutting-stock problems (such as linear-programming relaxations) are no longer appropriate. Furthermore, constraints reflecting properties of the final product (such as noise constraints) may only be estimated from the properties of the metal bands used. To this end, this paper intends on providing a framework for addressing such generalized cutting-stock problems within the scope of stochastic-local search algorithms. The proposed optimization framework provides flexibility in the number and nature of the constraints and/or objectives. An additional learning system provides the necessary predictions for constraints that depend on properties of the final product. Comparison is performed with existing software in an industrial site.