Year: 2026
Type: article
Abstract: Neural time-series methods hold increasing promise for improving how manufacturing systems operate, enabling more accurate monitoring, earlier faultdetection, and better-informed operational decisions. Human-in-the-loop architectures extend this further, pairing machine learning with human expertise torefine and guide how analytical pipelines perform in practice.Building such pipelines, however, surfaces a range of challenges. Industrialsystems record machine states, buffer levels, and event logs for operationalsupervision, not to produce clean training sequences. The data entering neuraltime-series pipelines are often irregular in time and incomplete across variables,while most neural methods assume regularly sampled and densely observedinputs.The aim is to relate how observation conditions arise in manufacturingsystems to where neural methods encode assumptions about time and absence,and thereby ground method selection in the observation process for human-inthe-loop (HITL) decision-support settings.A simulation study of a discrete manufacturing system shows how ordinaryproduction-structure mechanisms generate temporally uneven, asynchronous,and dependent traces under controlled assumptions. A structured surveyexamines how neural methods for irregular and incomplete time series differ intheir treatment of time and absence within the modeling pipeline. The maincontribution is methodological: comparison should begin from the observationprocess, treating irregular sampling and missingness as distinct properties withdifferent implications for representation, evaluation, and method selection inHITL decision-support settings.
Author: Marti, Silvan
Institution: Chalmers University of Technology
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7205298520
