MACHINE LEARNING-BASED POSITIONING USING MULTIVARIATE TIME SERIES CLASSIFICATION FOR FACTORY ENVIRONMENTS - A PREPRINT

Aug 3, 2026

Upload by Quinn

Indoor Positioning Systems are essential for industrial tracking and navigation, yet conventional solutions often depend on external infrastructure that creates privacy risks, additional operational requirements, and assumptions that limit long-term usability. This work develops a machine learning-based indoor positioning approach designed for factory environments where supplementary infrastructure is infeasible. The method fuses motion and ambient sensor signals and formulates localization as multivariate time series classification. A custom dataset emulates factory assembly lines to compare models by accuracy, memory footprint, and inference speed. Results show all evaluated models exceed 80% accuracy, with CNN-1D offering balanced performance, MLP following, and DT achieving the lowest memory and latency for potential real-world deployment.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Abstract

1 Introduction

arXiv :2308 . 11670v1 [ ee ss . SP] 22 Aug 2023 MACHINE LEARNING-BASED POSITIONING USING MULTIVARIATE TIME SERIES CLASSIFICATION FOR FACTORY ENVIRONMENTS A PREPRINT Nisal Hemadasa Manikku Badu Institute of Telematics Hamburg University of Technology 21073 Hamburg, Germany [nisal.hemadasa@tuhh.de](nisal.hemadasa@tuhh.de) Marcus Venzke Institute of Telematics Hamburg University of Technology 21073 Hamburg, Germany [venzke@tuhh.de](venzke@tuhh.de) Volker Turau Institute of Telematics Hamburg University of Technology 21073 Hamburg, Germany [turau@tuhh.de](turau@tuhh.de) Yanqiu Huang Faculty of Electrical Engineering University of Twente 7522NH Enschede, The Netherlands [yanqiu.huang@utwente.nl](yanqiu.huang@utwente.nl) August 24, 2023 ABSTRACT Indoor Positioning Systems (IPS) gained importance in many industrial applications. State-of-theart solutions heavily rely on external infrastructures and are subject to potential privacy compromises, external information requirements, and assumptions, that make it unfavorable for environments demanding privacy and prolonged functionality. In certain environments deploying supplementary infrastructures for indoor positioning could be infeasible and expensive. Recent developments in machine learning (ML) offer solutions to address these limitations relying only on the data from onboard sensors ofIoT devices. However, it is unclear which model fits best considering the resource constraints of IoT devices. This paper presents a machine learning-based indoor positioning system, using motion and ambient sensors, to localize a moving entity in privacy concerned factory environments. The problem is formulated as a multivariate time series classification (MTSC) and a comparative analysis of different machine learning models is conducted in order to address it. We introduce a novel time series dataset emulating the assembly lines of a factory. This dataset is utilized to assess and compare the selected models in terms of accuracy, memory footprint and inference speed. The results illustrate that all evaluated models can achieve accuracies above 80% . CNN-1D shows the most balanced performance, followed by MLP. DT was found to have the lowest memory footprint and inference latency, indicating its potential for a deployment in real-world scenarios. Keywords Indoor positioning · Machine learning · Sensor fusion · Multivariate time series classification 1 Introduction Indoor Positioning is a technology widely adopted in many industries, including medical, sales, manufacturing, logistics and construction [5, 19] . It is also among the foremost in technological fronts such as Smart Cities, Industrial Internet of Things (IIoT) [8] . In each of these fields, IPS’s play important roles in tracking, navigation, proximity, and inertial measurements [5], thereby injecting more efficiency, accuracy, and safety to processes. Our work is motivated by insights from animal behavioral scientists. Many animal species possess a natural ability to navigate and recognize their location by utilizing various cues such as geomagnetic fields, celestial bodies, wind direction, temperature, scent, and visual landmarks. They develop mental maps through learning and memory, enabling them to find routes, recognize environments, and differentiate between different locations. This concept can be applied to the localization of entities following a predetermined path. By processing sensory inputs acquired at a given moment or over a specific time period, an estimation of the current position can be derived. This could be perceived asa sub-problem of Indoor Positioning. However, unlike the conventional indoor localization approaches on determining precise x-y coordinates, we reframe the problem to ascertain a relative segment on a pre-determined path. A plethora of research efforts addresses the indoor positioning problem from a broad range of approaches. These approaches provide precis

FAQs

Why are conventional indoor positioning solutions problematic in factory environments?

They often rely on external infrastructure, which can introduce privacy compromises, require additional external information, and depend on assumptions that reduce feasibility for long-term operation.

How does the proposed method localize an entity in the factory?

It uses motion and ambient sensors, fuses the recorded signals into multivariate time series, and formulates localization along a predefined path as a multivariate time series classification task.

How were different machine learning models evaluated and compared?

A novel dataset emulating factory assembly lines was introduced, and models were compared using accuracy, memory footprint, and inference speed, showing all models above 80% accuracy.

Aug 3, 2026

Upload by Quinn