IFAC PapersOnLine 58-8 (2024) 133–138
ScienceDirect ScienceDirect
Available online at www.sciencedirect.com
2405-8963 Copyright © 2024 The Authors. This is an open access article under the CC BY-NC-ND license.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2024.08.062
analysis of historical data increases the effectiveness of
decisions made and thus reduces losses caused by machine
downtime. However, most often this analysis is based on the
experience of the decision-maker, and a formal approach based
on mathematical methods and models is used to a very limited
extent. Although experience is important and influences the
effectiveness of decisions, using data analysis methods we can
significantly increase this effectiveness.
The large amount of historical data contains information about
events that occur along an industrial production line. Having a
set of historical data about emergency events and their causes
makes it possible to automate decision-making processes
based on a data-driven approach.
Data-driven approaches, particularly machine learning (ML),
are attracting attention. According to Quatrini et al. (2020) “by
using ML tools it is possible to discover the relationship
between different factors and analyze the degree of influence
of related variables”. By training models on historical data,
these approaches enable more targeted maintenance activities
(Çınar, et al., 2020; Cline, et al., 2017).
The novelty of the study lies in the use of relatively simple,
fast, and computationally inexpensive ML methods (from the
group of decision trees and random forests) to analyze
complex maintenance problems. This approach will facilitate
both the development of a family of scalable ML systems for
use in predictive maintenance in companies of all sizes,
including those without large maintenance department
structures and their implementation on mobile devices
(smartphones, tablets), which will be handier to use in
industrial settings (including the Industrial Internet of Things).
1. INTRODUCTION
In today's competitive landscape, maintenance has become one
of the most important business functions, and having an
effective maintenance system is important to ensure
production continuity, quality, and timely delivery to the
customer, reduce losses, and ensure operational safety
Over the years, maintenance management professionals have
implemented many diverse types of maintenance strategies to
make their maintenance programs as effective, efficient, and
responsive as possible. The goal of the latest evolution of this
task, predictive maintenance (PdM), is to monitor the
condition of machines to decide when repair/replacement is
necessary, taking into account the degree of degradation
(Wang, et al., 2023; Saihi, et al., 2023). This approach is
becoming more and more popular among both scientists and
practitioners. However, in many enterprises, maintenance
activities are still planned and implemented based on
manufacturer recommendations and/or historical data
regarding failure events, maintenance activities, and their
effectiveness. Despite increasingly effective methods of
planning maintenance activities, machine and equipment
failures still occur, and the maintenance staff, based on the
failure report, must decide as quickly as possible what actions
to take and what human and material resources to allocate to
ensure that the downtime caused by the failure is as short as
possible.
To determine the scope of work and the resources needed in
many enterprises, maintenance employees use data on similar
events that occurred on this machine or similar machines. The
Keywords: data-driven maintenance, decision-making, machine learning, decision tree, boosted trees
Abstract: The large amount of historical data contains information about events that occur along an
industrial production line. Having a set of historical data about emergency events and their causes makes it
possible to automate decision-making processes based on a data-driven approach. Data-driven approaches,
particularly machine learning (ML), are attracting attention. Due to its visualization and interpretability
characteristics, the decision tree (DT) model is an important ML tool for decision analysis. This paper aims
to present the possibility of using DT to increase the efficiency and effectiveness of maintenance activities
by identifying the probable cause of failure based on historical data. Based on the research conducted, we
have shown that the use of machine learning techniques can improve the accuracy of decisions regarding
the type of maintenance work that should be carried out to efficiently and effectively remove failures and
reduce losses caused by machine downtime.
Keywords: data-driven maintenance, decision-making, machine learning, decision tree, boosted trees
*Faculty of Computer Science, Kazimierz Wielki University, Bydgoszcz, 85-064
Poland(e-mail: izabela.rojek@ukw.edu.pl, dariusz.mikolajewski@ukw.edu.pl)
** Faculty of Engineering Management, Poznan University of Technology in Poznan, 60-965
Poland (e-mail:malgorzata.jasiulewicz-kaczmarek@put.poznan.pl)
*** Faculty of Engineering Management, WSB Merito Univeristy of Poznań, 61-895
Poland (e-mail: mariusz.piechowski@wsb.poznan.pl)
Izabela Rojek*Małgorzata Jasiulewicz-Kaczmarek** Mariusz Piechowski*** Dariusz
Mikołajewski*
The use of decision trees to identify the causes of failures
in a medical enterprise - a case study
Copyright © 2024 The Authors. This is an open access article under the CC BY-NC-ND license
(https://creativecommons.org/licenses/by-nc-nd/4.0/)