Towards a smart workflow in CMMS/EAM systems: An approach based on ML and MCDM

Abstract

In the context of Industry 4.0, process automation and predictive maintenance play an essential role. There is a need to provide more effective and faster maintenance through the integration of industrial tools and processes, to support manufacturing operations, in the perspective of integration standards and architectures. In a typical maintenance system, registration and maintenance requests are made through maintenance orders, which consist of a standard form and usually are created and filled manually. However, predictive maintenance requires a higher level of automatization, from data acquisition to maintenance order generation in a Computerized Maintenance Management System (CMMS) / Enterprise Asset Management (EAM). The proposal is to automate the process of generating maintenance orders, providing for automated form completion. At the physical level, assets are monitored by sensors, and, based on a set of rules, the respective predictive maintenance order will be issued in CMMS/EAS. Maintenance orders may contain variable fields according to each asset, so Machine Learning (ML) and Multicriteria Decision Making (MCDM) will be applied to fill in these fields, as well as the allocation of the maintenance orders to the maintainer that best fit for the maintenance specification. This automatic process will assist the maintenance workflow, leading to a Smart Workflow concept. A serial and parallel framework are presented, the former consists in applying TOPSIS (MCDM method) to extract features for the use in ML classification, in order to automatically fill in the appropriate form fields. The latter consists in a classification using both methods (ML and MCDM), where TOPSIS performs the initial classification, and if the alternatives ranking scores are close to each other, ML is used for more accurate classification. A case study was carried out in a Brazilian company that develops a CMMS/EAM system, distributed worldwide, and the results demonstrate that the concept of Smart Workflows is valuable, simplifying and enhancing maintenance processes.

Introduction

A maintenance strategy outlines an organization's plan on how to preserve the health and safety of its assets [1] and consists in procedures for survey, inspection, repair, upkeep, and renewal of systems, subsystems, and components [2]. Moreover, this requires the evaluation and selection of efficient maintenance approaches that provide and deploy information regarding aspects that affect equipment or component lifecycles [3]. In this context, predictive maintenance is a strategy that delivers the ability to predict imminent breakdowns and faults in applications, based on real-time data. Consequently, actions can be taken proactively to ensure a given component's availability and to minimize downtime needed for repair [4]. Nevertheless, due to the increasing amount of data in organizations, adopting tools to assist in managing maintenance activities as well as making the information available in real-time to support the decision-making process has become essential [5].

In this sense, Computerized Maintenance Management Systems (CMMS) and Enterprise Asset Management (EAM) play a vital role in maintenance and repair activities. CMMS aims to assist in the planning, management, and administrative functions required for effective and efficient maintenance [6], whereas EAM aims to manage the asset's lifecycle by enhancing its efficiency and reliability, reducing maintenance costs and risks for the organization [7]. Deploying a CMMS/EAM may promote many benefits for organizations, due to their capability in storing large volumes of data (e.g., machine and/or sensor data, event logs), manage work orders, record previous maintenance cases and best practices, track services, parts inventory, purchase orders, machine downtime, among others [8,9]. Therefore, CMMS/EAM have been designed to trigger maintenance actions, offering users a more realistic activity needs scenario in responding to on-demand work orders or just performing routine inspections [10].

Despite being a fundamental element in maintenance management, a typical CMMS/EAM workflow depends on manual resources to prepare maintenance orders and allocate maintainers to perform activities. The time between the occurrence and the arrival of a technician is known as the response phase [11], which consists in the detection of an occurrence (i.e., machine malfunction), creation of a work order, which can be triggered manually or automatically, and action requirements to be evaluated (i.e., maintenance analyst), planned and scheduled for delivery by a technician. However, this process might generate bottlenecks caused by high numbers of maintenance requests that may arise, resulting in additional costs due to the lack of timely maintenance.

From the integration perspective, according to ISA-95/IEC62264 [12], CMMS/EAM systems belong to level 3 in the automation pyramid, i.e., the same level as Manufacturing Execution Systems (MES). In this sense, data integration is essential in order to exchange assets’ real-time data (e.g., sensors, machine status) to CMMS/EAM systems. So, data integration associated with automatic techniques to support decision-making would be able to reduce maintenance lead-times (from detection to execution) and turn traditional CMMS/EAM workflow into a Smart Workflow [13].

Recent improvements resulted in higher availability of sensors, data acquisition systems, industrial controllers, and computer networks in the manufacturing area [14]. These technologies allow vertical, horizontal, and end-to-end data integration (e.g., machine-to-server, server-to-server, and machine-to-machine) by adopting standardized communication protocols such as OPC UA, MQTT, RESTful HTTP [15], among others. However, the maintenance analysis and planning stages depend on several criteria that must be taken into account when making a decision, such as safety aspects, failure costs, prioritization, product quality, manpower experience, and availability, among others. Additionally, some of these criteria cannot be expressed in financial terms making quantifying them even more complex [16]. So, expert knowledge may be used to aggregate qualitative and quantitative information to support analysis and improve decisions in these stages [17].

From this perspective, Multiple Criteria Decision Making (MCDM) methods aim to apply a set of criteria to assist decision-makers in evaluating complex scenarios. This approach results in a finite set of alternatives that decision-makers must evaluate, select, or rank [1]. Based on their individual importance, the finite set of criteria is weighted, and by using suitable metrics, each alternative is evaluated with respect to each criterion. The evaluation ratings are then pooled, and the alternatives prioritized, ranging from best to worst [16]. As stated by [18], MCDM methods can be linked to CMMS/EAM in order to provide a decision support capability, adding value to data collected in a predictive maintenance context. On the other hand, prediction techniques are a highly debated research area and a trending topic in manufacturing industries [19]. To this end, Machine Learning (ML) techniques can be applied to discover relations among features [20], and, in terms of maintenance management, this information can be used to predict a system's behavior, predict abnormal events, generate warnings, and notify systems and/or operators, assisting in diagnosis and maintenance tasks [21,22].

The proposal is to turn a typical maintenance management workflow into a Smart Workflow, in the areas of preparing maintenance orders and allocating activities to maintainers, in order to promote faster delivery, thereby reducing bottlenecks caused by human activities. In this article, two frameworks are proposed that integrate ML and MCDM to provide Smart Workflow capabilities, with one approach being parallel and the other, series.

The parallel framework is a proposal for integrating ML and MCDM where both are used in the classification stage in order to reduce computational costs and apply a hierarchy of methods to provide lowest cost classification (MCDM) and only use the more expensive (ML) as real proof, in case MCDM is unable to achieve an accurate classification. The parallel framework is applied in selecting the most suitable maintainer to perform the maintenance activity. In the serial framework, MCDM is responsible for choosing the best features for ML, driving higher accuracy in results. Thus, ML operates as a method of classification. In this case, MCDM serves as decision-making support for the ML operation. The serial framework is applied when filling in maintenance order forms.

This paper is structured as follows. Section 2 presents the main concepts and related works. Section 3 introduces the Smart Workflow within a CMMS/EAM system. Section 4 presents the serial and parallel frameworks, used to fill out form fields, and select the most appropriate maintainer to allocate each maintenance order. Section 5 presents the case study in which the research frameworks were applied and their results. Section 6 is responsible for the research conclusions and final considerations.