Smart retrofitting in maintenance: a systematic literature review

If data analysis is to occur, devices must have the capability of collecting data from sensors (Barksdale et al., 2018; Ge et al., 2019; Liu et al., 2018). This can be done through preexisting sensing capabilities if they are available (Luo, 2020; Chau et al., 2015; Calabrese et al., 2020), but other times the installation of additional sensors is necessary (Huber et al., 2019; Prathima et al., 2020; Kiangala & Wang, 2018). Such sensors are sometimes added in the form of sensor nodes (Christou et al., 2020; Surico et al., 2020; Catenazzo et al., 2018), which are IoT devices that integrate both sensing and communication capabilities. Although it might not always be possible, some authors emphasize the need to collect data in a non-invasive manner. Jónasdóttir et al. (2018) propose a retrofitting solution that does not require machinery to be opened or heavily modified. Wiemer et al. (2019) cite the need of not interrupting a running production while the upgrading process takes place by performing any pilot test in a testbed rather than in the production line. Regardless of the specific details of the process, it becomes clear that any SRM process involves capturing data from the target legacy devices.

Devices also need the ability to communicate with other entities. The OM2M platform (Alaya et al., 2014) defines three actors present in any Machine to Machine (M2M) architecture: M2M devices (which are natively capable of communicating to an M2M network), legacy devices (which are not capable of communicating to a network), and M2M gateways (which enable M2M communication capabilities for legacy devices). Communication in SRM closely resembles this paradigm. According to the reviewed literature, devices must be able to communicate with external entities such as smart mobile devices (Ranjbar et al., 2019; Hussain et al., 2020; Cologni et al., 2015), local central computers or processors (Atluru et al., 2012; Ashjaei & Bengtsson, 2017; Short & Twiddle, 2019), or to a remote location (Nordal & El-Thalji, 2021; Damanik et al., 2020; Bucci et al., 2020) (such as the equipment maintainer or the cloud). Through communication, legacy devices can be integrated into management software, like Maintenance Execution Systems (MES) and Computerized Maintenance Management Systems (CMMS) (Balogh et al., 2018; Barton et al., 2019; Chau et al., 2015). Some authors identified that such communication should be wireless (Ziegelaar et al., 2020; Magadán et al., 2020; Uhlmann et al., 2017), sometimes referencing the use of wireless sensor networks (Talmoudi et al., 2019; Priyanka et al., 2021; Sadiki et al., 2018). Furthermore, enabling interoperability by being compatible with multiple vendors and protocols is also cited as a requirement for all these types of communication (Priller et al., 2014; Chen et al., 2018; Liang et al., 2020). When communication is not initially possible for a device, it is often enabled using a gateway device (Alexandru et al., 2016; Richter et al., 2019; Scholtz et al., 2018). The specific means of communication and the specific entity with which devices communicate change depending on the considered application, but the requirement of data communication remains constant in most of the analyzed works.

Data that are collected and transmitted must then be processed, converting it into information, knowledge and eventually wisdom (Ackoff, 1989). Such processing involves any manipulation of data that requires computational capabilities before the data can lead to decision making and other added value services. Data are generally stored in databases (Bousdekis et al., 2019; Prudenzi et al., 2019a; Arosio et al., 2014), so that current and historical data can be retrieved at any point (Ardila et al., 2020; Romero et al., 2021). Processing then includes data cleaning or preprocessing (Uhlmann et al., 2017; Åkerman et al., 2018; Chen et al., 2018), data visualization (Yu et al., 2014; Bousdekis et al., 2019; Prathima et al., 2020), and in general, data analysis (Wiemer et al., 2019; Ranjbar et al., 2019; Calabrese et al., 2020). Although data processing is often performed entirely in the cloud (Gayathri & Vasudevan, 2018; Balogh et al., 2018; Iqbal et al., 2019), edge and fog computing can be utilized as alternatives or supplements to cloud computing (Strauß et al., 2018; Shapsough et al., 2020; Magadán et al., 2020).

By enabling data collection, communication, and processing for a legacy device, the exploitation of service-based architectures can be achieved (Lesjak et al., 2014; McNally et al., 2020; Alonso et al., 2018). Such services can be classified according to the Prognostics and Health Management (PHM) architecture proposed in Li et al. (2020). These services consist in:

Only four of the analyzed works explicitly study SRM in the context of reactive maintenance. Alexandru et al. (2016) generate smartphone notifications when a machine’s programmable logic controller (PLC) triggers an alarm. Ramani et al. (2016) and Deroussi et al. (2018) use SRM to generate notifications that are sent to operators when a machine failure is detected. Priller et al. (2014) mention the potential of using retrofitting to execute predefined reaction patterns when a machine breaks down (such as automatically scheduling a repair action), rather than simply generating a notification. It can be noticed that the objective is the automation of reactive maintenance actions through SRM after the occurrence of functional failures. However, such exploitation of SRM in reactive maintenance policies appears to be relatively uncommon.

It is worthy of mention that this scarcity of reactive maintenance-related papers was expected. The chosen keyword matrix includes terms such as Industry 4.0 and Smart Maintenance, whose technologies are not necessarily required in reactive maintenance. Instead, traditional industrial automation technologies (such as PLC’s) might be sufficient to enable maintenance-related services in reactive maintenance; e.g. fault detection. This low number of papers does not reflect a lack of research on smart retrofitting in reactive maintenance, but instead shows that the focus of this literature review lies elsewhere.

Planned maintenance papers were found to be researched more frequently than reactive and strategic maintenance. Within planned maintenance papers, two variables that were frequently measured are the operational state and operational mode of devices. An operational state refers to the level of activity within a structure. Operational states are commonly described in simple terms of binary ON and OFF values (Wasson, 2006). Whereas, an operational mode specifies an abstract user-selectable set of system activities that focuses on satisfying an objective. They are not limited to ON / OFF values, since operational modes might also include modes like initialization, calibration, configuration, cleaning, production, and even maintenance (Wasson, 2006).

Regarding operational states (ON/OFF), machinery was retrofitted so that start and stop times could be recorded, allowing maintenance to be done according to total work-hours (Damanik et al., 2020; Bhandari et al., 2020; Calabrese et al., 2020; Erazo Navas et al., 2021). When devices do not have the ability to measure their own state, variables such as voltage and current can be used to monitor the operational state (Bhandari et al., 2020; Mourtzis & Vlachou, 2018). The machine operational mode is also recorded in some of the analyzed works (Pistofidis & Emmanouilidis, 2012; Mourtzis & Vlachou, 2018; Cattaneo & Macchi, 2019). These operational measurements allow for an increased awareness of the shop floor condition, which in turn allows better maintenance and workshop planning (Mourtzis & Vlachou, 2018).

Proactive maintenance papers were by far the most frequent. Such works are characterized by the implementation of CBM techniques, which aim to predict potential failures through the analysis of physical variables (Jardine et al., 2006).

The most frequently measured variables were vibration (Surico et al., 2020; Aqueveque et al., 2021; Magadán et al., 2020) and temperature (Ciancio et al., 2020; Iqbal et al., 2019; Short & Twiddle, 2019), followed by electric current (Ge et al., 2019; Strauß et al., 2018; Barksdale et al., 2018). The most analyzed components were motors, induction or otherwise (Rubio et al., 2018; Eiskop et al., 2017; Talmoudi et al., 2019), followed by bearings (Richter et al., 2019; Bernal et al., 2018). Most research on SRM involves rotating machinery. Other variables of interest that were measured were fluid pressure (Schneider et al., 2019; Lalanda et al., 2017) and electric voltage (Al Kindhi & Pratama, 2021; Prudenzi et al., 2019b).

Specific mentions to strategic maintenance (i.e. fulfillment of asset management or organizational goals) were seldom found. Only two of the analyzed works study SRM in the context of strategic maintenance, both of which belong to the same research group.

Huber et al. (2019); Wiemer et al. (2019) worked on extending the CRoss-Industry Standard Process for Data Mining (CRISP-DM). CRISP-DM is an open standard that describes a standard approach to data mining projects (Wirth & Hipp, 2000). It consists of 6 layers: Business Understanding, Data Understanding, Data Preparation, Model Building, Model Evaluation, and Deployment. The two first layers are of special interest. Business Understanding focuses on understanding the project objectives and requirements from a business perspective, while Data Understanding involves data collection (Wirth & Hipp, 2000). Authors defined intermediate steps between Business Understanding and Data Understanding. These are: i) business objectives are transformed into technical tasks that fulfill said objectives; ii) a selection of the data required to complete these technical tasks is made; iii) proper measurement equipment and methodology are selected. Such intermediate steps allow a direct link between organizational goals and technical implementations of SRM, making these papers examples of the implementation of strategic maintenance in an enterprise.

The introduction of maintenance services through smart retrofitting can increase the performance of manufacturing systems, optimizing their production capabilities (Tedeschi et al., 2017; Ramani et al., 2016; Jung & Jin, 2018). This is possible by improving the availability of physical assets, which is in turn a function of their reliability and maintainability (Parra & Crespo, 2015).

SRM has the potential to improve the reliability of assets (Cattaneo & Macchi, 2019; Deroussi et al., 2018; Hussain et al., 2020), by increasing the Mean Time To Failure (MTTF) and consequently decreasing the Failure Frequency (FF). Improving reliability is often cited as a result of an improved ability to schedule maintenance actions (Priller et al., 2014; Yiu et al., 2019; Catenazzo et al., 2018), and in some cases, as a result of implementing prognostics and health management through retrofitting (Ranjbar et al., 2019; Cattaneo & Macchi, 2019; Vogl et al., 2015). When failures occur, SRM can also reduce the Mean Down Time (MDT), effectively improving maintainability (Sezer et al., 2018; Lesjak et al., 2014; Alves et al., 2020). Besides an improved maintenance schedule, maintainability might be augmented by retrofitting through faster response times (Wang et al., 2020; Liang et al., 2020; Gayathri & Vasudevan, 2018), which enable the use of alarms, earlier detection of faults, and remote notifications.

The collection of these improvements increase the availability of assets (Jónasdóttir et al., 2018; Bernal et al., 2018; Mykoniatis, 2020), which means that assets stay in a productive state for a greater amount of time. Furthermore, the optimization of production through workshop scheduling (rather than just maintenance scheduling) is also enabled by SRM (Zhang et al., 2017; Vogl et al., 2015; Chen et al., 2018). Finally, SRM can improve the quality of produced products and offered services (Prudenzi et al., 2019a; Sezer et al., 2018; Hsu et al., 2019).

Another way in which maintenance adds value is by reducing risk. Risk is the potential impact to an asset or another source of value that may arise from a present process or from a future situation (Márquez, 2007). In maintenance, risk commonly refers to safety, environmental, operational and not operational effects of failures that might lead to monetary losses (Ben-Daya et al., 2016). Maintenance can reduce risk in an organization through: safety, environmental risk, and stakeholder confidence (GFMAM, 2021). SRM was found to address all these elements in varying levels.

SRM has been said to improve operator safety. Safety sometimes comes from the increased autonomy of assets, which reduce the need of operators directly interacting with potentially hazardous machinery (Gayathri & Vasudevan, 2018; Catenazzo et al., 2018). Risk has also been minimized by reducing the chances of potential injuries due to damaged equipment (Hesser & Markert, 2019), preventing unsafe working conditions (Nordal & El-Thalji, 2021), or by providing clear instructions for operators to follow when performing risky tasks, reducing potential human errors (Arosio et al., 2014). Regarding human-asset interactions, some papers also tracked operator location through retrofitted machinery. These authors propose tracking the position of each operator and which tasks they have completed on which assets (Arosio et al., 2014; Pistofidis & Emmanouilidis, 2012). This tracking mechanics can lead to a more human-centered maintenance enabling greater situational awareness of operators (Oliveira et al., 2013), and increasing safety while reducing errors (Gavish et al., 2015).

Concerning environmental risks, some authors in the studied papers concerned themselves with environmental sustainability (Scholtz et al., 2018; Zhang et al., 2017). Although sustainability was seldom mentioned directly in SMR-related works, one of the main functions of maintenance is maintaining plant and environmental safety (Muchiri et al., 2011). This indicates that the improvement of the maintenance function has the potential to improve environmental sustainability. Conversely, an inadequate execution of maintenance can lead to hazardous emissions, production wastes due to untimely breakdowns, and inefficient use of energy and resources (Liyanage & Badurdeen, 2010). Proper maintenance (and SRM as an extension) then has a significant sustainability impact in the organizations (Franciosi et al., 2020).

Finally, the literature indicates that SRM has the potential to increase stakeholder satisfaction (Jung & Jin, 2018; Gayathri & Vasudevan, 2018), by enabling working-culture-oriented and reliable business models (Nordal & El-Thalji, 2021). The studied works then indicate that SMR has the potential to bring a positive impact on risk management through safety, environmental risk, and stakeholder confidence.

The third main added value that maintenance offers is cost reduction. The costs associated with asset utilization can be divided into capital expenditures and operational expenditures. Maintenance can support the reduction of these costs (GFMAM, 2021), and SRM can aid this process.

Capital expenditures (CAPEX) involve money spent on acquiring or upgrading productive assets. This includes buying or replacing machinery to increase overall productivity or augment redundancy. One way SRM enables the reduction of this cost is by increasing the remaining useful life (RUL) of assets (Hesser & Markert, 2019; Aqueveque et al., 2021; Strauß et al., 2018). When machines can perform their functions for longer, less refurbishments and replacements are needed. Another driver that reduces CAPEX is enabling interoperability between machinery from different vendors (Mourtzis & Vlachou, 2018; Barton et al., 2019; Iqbal et al., 2019). Interoperability eliminates the need to replace functioning machinery simply because it uses a different or outdated communication interface, and removes the need to purchase separate software tools that are compatible with specific vendors only. Finally, SRM solutions usually aim to have a low-cost implementation, meaning that transitioning from legacy to retrofitted assets does not involve high CAPEX (Chen et al., 2016; Eiskop et al., 2017; Sezer et al., 2018).

Operational expenditures (OPEX) are related to money spent on operating costs such as raw materials, utilities (electricity, water, etc.) and labor. It also includes maintenance itself, both labor-wise and parts-wise. Numerous examples can be found in which SRM is demonstrated to save resources, such as maintenance costs (Ramani et al., 2016; Surico et al., 2020), materials (including spare parts) (Bernal et al., 2018; Priller et al., 2014), and energy (Alonso et al., 2018; Pignatelli et al., 2015). Of special interest are ways in which SRM can simplify the maintenance process, potentially saving money and time. Retrofitting has enabled the use of remote services (Khademi et al., 2019; Liang et al., 2020) and outsourcing maintenance through after-sales services (Chen et al., 2016; Zhang et al., 2017), which reduce the need of having expert personnel on-site. Meanwhile, it has enabled an increased asset autonomy (Chau et al., 2015; Prathima et al., 2020) and ease of use (Lin et al., 2014; McNally et al., 2020), meaning that human resources may be reduced through retrofitting.

Cybersecurity is the computer science field that devotes itself to protecting the privacy, confidentiality and integrity of data that is stored and transmitted. The importance of this field is ever-increasing, as cyber-attacks become more frequent and sophisticated (Babiceanu & Seker, 2016). In this context, multiple considerations are made regarding security of retrofitted machinery. Intellectual property becomes a concern, as poor security measures in data transmission might lead to third parties capturing production data (Eiskop et al., 2017; Mourtzis & Vlachou, 2018). Such considerations become a greater concern if a device is connected permanently to the internet (Lesjak et al., 2014), so alternatives like edge and fog computing become of interest (Ashjaei & Bengtsson, 2017).

Collecting data for maintenance purposes is not a new concept since many plants already do this. However, these datasets are usually fragmented across systems because of their heterogeneous semantics and formats (Christou et al., 2020; Barbieri & Gutierrez, 2021). The development of common standards for device interoperability were found to be some of the main challenges of health management in smart manufacturing (Weiss et al., 2015). Even if various machine to machine (M2M) communication standards have been developed to this day (such as MQTT¹, MTConnect², and OPC-UA³), heterogeneity and the lack of interoperability are still challenges today (Barton et al., 2019; Prathima et al., 2020). Interoperability issues are not limited to machine communication. Existing data analysis and visualization dashboards (such as Power BI, Tableau, and Google Data Studio) have been found to be incompatible with sensor data streams and streaming technologies, but instead operate properly only on classical data storage (Moens et al., 2020).

Cloud services are commonly used in retrofitting scenarios. However, cloud services usually do not guarantee the reactiveness required from time-critical operations (Khelifi et al., 2018; Wang et al., 2020). Even for applications that are not time-critical, network reliability has been identified as an issue in using the cloud (Wang et al., 2020). Some authors have also cited real-time data collection and processing as challenges (Mourtzis & Vlachou, 2018; Cologni et al., 2015). Consequently, edge computing has been proposed as a solution to these issues (Ashjaei & Bengtsson, 2017; Strauß et al., 2018). However, edge computing has several constraints that the cloud does not, such as computational power and available data storage (Scholtz et al., 2018). The aforementioned issues make the challenge of latency in data transfer still relevant.

Volume is one of the main dimensions of Big Data, which is increasingly relevant in the manufacturing domain (Babiceanu & Seker, 2016). Big Data refers to amounts of data that are too large to be processed efficiently through traditional methods (Kaisler et al., 2013). The communication, storage and real-time processing of such large data volumes have become a challenge for IoT devices (and by extension, retrofitted devices) (Lee et al., 2017). Regarding communication, the bandwidth of existing networks in industrial scenarios can be insufficient for these data volumes (Catenazzo et al., 2018). Edge computing might be considered as a solution to this, but storage and processing of large data volumes are an even bigger challenge in edge computing than in cloud computing (Scholtz et al., 2018). Furthermore enabling this communication, storage and processing of large data volumes might require large start-up (CAPEX) and maintenance (OPEX) investments in plants and businesses (Lee et al., 2017). Note that this challenge should be more relevant in proactive maintenance than planned maintenance.

When performing CBM, sensor data must be collected before being able to perform analyses. Run-to-failure data is necessary to train CBM models so they can detect the current state of a device (Calabrese et al., 2020; Alves et al., 2020). However, such data is not always available to plants that perform CBM on a device for the first time (Lei et al., 2018). The collection of sensor data through retrofitting well in advance of obtaining condition estimation capabilities must be a long-term commitment for enterprises that implement CBM. Much like with data volume, this challenge should be more relevant in proactive maintenance than planned maintenance.

To properly conduct a retrofitting project, an interdisciplinary team of domain experts is needed (Huber et al., 2019); e.g. data scientists, process engineers, control engineers, maintenance engineers, etc. After the termination of the retrofitting activity, professional and technical persons that are familiar with Information and Communication Technologies (ICT) are required to operate and maintain the retrofitted equipment (Liang et al., 2020; Bucci et al., 2020). Acquiring expert knowledge and/or personnel that has such knowledge can be especially challenging for Small and Medium Enterprises (SMEs) whose budget might be more restricted than larger enterprises (Jung & Jin, 2018; Surico et al., 2020).