Hybrid machine learning/constraint programming method for proactive and reactive scheduling applied to PCB assembly
Year: 2026
Type: conference-paper
Abstract: Printed Circuit Board (PCB) assembly is a demanding production environment where due-date compliance must be maintained under variable operating conditions. This paper addresses scheduling under uncertainty for uniform parallel machines and introduces a proactive-reactive rescheduling policy learned as a decision tree. Starting from an initial schedule, the policy evaluates the ongoing scenario at data-driven checkpoints—whose timing is learned by a classifier—and classifies it as Good or Bad with respect to due-date performance. If Good, the schedule is continued; if Bad, a rescheduling action is triggered. The decision tree is trained from simulation labels and complemented by a post-processing step that encourages early decisions to support timely reactions. The evaluation protocol examines whether the learned policy increases the proportion of on-time jobs on scenarios predicted as adverse for the initial schedule. Experiments are conducted with stochastically generated speed scenarios to emulate variability; future work will extend the study with realistic data from a Schaeffler manufacturing plant. Results highlight the feasibility of learning proactive rescheduling rules that balance responsiveness and planning stability in PCB assembly.
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7201873379