Interpretability-Driven Multiple-Steps Optimization for Production Process Decision-Support System

Abstract

In an industrial manufacturing process engineers deal with decision-making problems, facing multiple conflicting operational and business objectives. We create a multiple-step optimization process to resolve the uncertainty of selecting proper objectives and their preferences, and address shortcomings of a typical linear weighted approach. Each step has one objective with a KPI with a clear semantic meaning, and its solution then transformed into a business constraint. The process allows to incorporate a feedback from a user and improve interpretability.