Contribution of Artificial Intelligence to Industrial Maintenance in the Field of Mechanics

Abstract

The global industry is in continuous technological evolution, which aims for reliability, efficiency, availability, and safety while reducing maintenance costs. Modern maintenance follows change, which can no longer be limited to being corrective or preventive, but must be proactive involving the continuous monitoring and verification of the root causes of failure; it must also be predictive which makes it possible to anticipate breakdowns and increase equipment usage time based on the Prognosis and Health Management (PHM), which transforms raw data into indicators and makes it possible to define the Residual Life (RUL) and its extrapolation as a decision-making tool. Our chapter consists of presenting the contribution of AI to industrial maintenance in the field of mechanics. It focuses on industrial maintenance through its concepts, technologies, and methods used. So, the presentation of artificial intelligence and its algorithms applied toward maintenance 4.0 are to show the contribution of AI to maintenance.

1. Introduction

In a market that is continuously growing, and to remain competitive, industries are orienting their management policies toward increasing productivity at a lower cost while optimizing resources and setting availability, reliability, maintainability, and safety as objectives for the proper functioning of their production systems.

These objectives can be achieved through the implementation of an adequate and relevant maintenance strategy. This interest is fueled by the fact that an unplanned shutdown can have important and significant economic consequences for the company. In this context, and in a spirit of anticipation, companies are transforming themselves to extract value from traditional industrial sectors and make them more efficient than ever by providing data-rich digital services thanks to technological evolution as well as the development of artificial intelligence and automated production systems. By leveraging ultra-low-cost connectivity, there is rapid explosion of sensors, powerful analytical tools, as well as data storage capacity. So, it becomes necessary and important to involve these new technologies in industrial maintenance and to integrate artificial intelligence (AI), which has become crucial to ensure the proper functioning of industrial systems, and therefore, to achieve operational and financial efficiency.

Using AI, management can always have accurate information about the operating status of any machine or equipment. Thus, AI in predictive maintenance helps businesses save money and resources. This is done by adapting maintenance routines to the needs of each piece of equipment, rather than forcing them to follow a rigid schedule.

Artificial intelligence is one of these structural innovations that has an impact in many areas. If its roots go back to the 1950s, recent technological developments, including machine learning and deep learning, open new possibilities for using artificial intelligence in industrial maintenance. Investing in maintenance helps avoid repair costs that also lead to production downtime. Having an intelligent solution to determine the precise timing of each technical maintenance intervention on industrial assets constitutes a key competitive advantage by significantly reducing the cost of a critical and urgent repair, as well as avoiding possible interruptions of service necessary for the latter.

To make the transition to 4.0 maintenance, it is important to proceed gradually according to your starting level, your internal resources, and your appetite for digital technology. Indeed, each technological development involves profound changes that affect the organization, the teams, and the technical infrastructure. It is therefore necessary to audit and map internal processes as well as data flows.

The outline of this chapter is divided into three parts. In the first part, we will focus on industrial maintenance through its concepts, technologies, and methods used. The second part will be devoted to the presentation of AI and its algorithms applied toward maintenance 4.0. In the third part, we will study the contribution of AI to maintenance based on research into the history and statistics as well as trends in AI.

2. Industrial maintenance in the field of mechanics

2.1 Definition of industrial maintenance

According to the Association Française de NORmalisation (AFNOR) NF-X 60,000 [1] standard, industrial maintenance is defined as follows: “It is all activities intended to maintain or restore an asset in a specified state or in given operational safety conditions, to accomplish a required function.”

Maintaining means preventing machine malfunctions, and restoring to correct following a machine malfunction. A specified state is what is defined by quantifiable characteristics and objectives. The industrial maintenance function revolves around different actions: management, diagnosis, troubleshooting, repair, verification, control, etc.

2.1.1 Maintenance concepts

The different maintenance concepts can be classified into three main categories: corrective maintenance, preventive maintenance, and predictive maintenance (Figure 1).

Figure 1.

Different forms of maintenance [1].

2.1.1.1 Corrective maintenance

The standard [1] defines corrective maintenance as: “Maintenance carried out after detection of a breakdown and intended to restore an asset to a state in which it can perform a required function.”

Corrective maintenance (Reactive) is carried out after the occurrence of a fault in the system. It is generally adopted for equipment for which:

• the consequences of the breakdown are not critical,

• the repair is easy and does not require much time,

• and investment costs are low.

The concept of corrective maintenance aims to reset the system to its normal operating state after the occurrence of its failure.

2.1.1.2 Preventative maintenance

Preventive maintenance aims to reduce the risk of a failure occurring. Standard [1] defines it as follows: “Maintenance carried out at predetermined intervals or according to prescribed criteria and intended to reduce the probability of failure or degradation of the functioning of an asset.” It is prepared and scheduled before the probable date of occurrence of a failure.

2.1.1.2.1 Systematic preventive maintenance

It is maintenance carried out according to a schedule established according to time or the number of units of use [1]. The frequency of replacements is determined using two methods: the first is block type, and the second is age type. The age-type replacement policy suggests replacing equipment at failure or after T units of uptime. The block-type policy suggests replacing the equipment after a predetermined period T, 2T, etc., regardless of the age and condition of the component. Systematic preventive maintenance can lead to over-maintenance, that is, to an excess of unnecessary interventions, and therefore to financial waste for the company. To compensate for this, other forms of preventive maintenance, based on monitoring the actual condition of equipment, have appeared, such as conditional maintenance [2].

2.1.1.2.2 Conditional preventive maintenance

According to standard [1], it is defined as: “Preventive maintenance based on monitoring of the operation of the asset and/or the significant parameters of this operation integrating the resulting actions.”

2.1.1.3 Predictive maintenance

It aims to compensate for the lack of knowledge of condition-based maintenance. It is defined according to standard [1] as: “Conditional maintenance carried out following the predictions extrapolated from the analysis and evaluation of significant parameters of the degradation of the asset.”

It also aims to offset the costs of corrective maintenance by minimizing the downtime of systems and, above all, by being able to plan these downtimes. This anticipatory method, therefore, makes it possible to ensure better continuity of service and thus reduces operating costs in the long term [3, 4].

Figure 2 shows the process of a predictive action.

Figure 2.

Process of a predictive action.

It considers the current conditions of the equipment and attempts to predict the evolution over time of the condition of the property. The expected benefits are indeed numerous:

• Reduction in the number of breakdowns,

• Reliability of production,

• Reduction of equipment downtime (costly),

• Increased business performance.

Predictive maintenance is based on continuous monitoring of system evolution to prevent shutdown before it happens [3, 4]. Predictive maintenance techniques help determine the condition of equipment in service to predict when maintenance should be performed. This approach saves money compared to routine or time-based preventative maintenance because tasks are performed only when warranted. The primary value of predicted maintenance is to enable convenient planning of corrective maintenance and avoid unforeseen equipment failures. The key is “the right information at the right time” (Table 1).

Standard maintenance	Maintenance
	Fix maintenance		Preventative maintenance
	Palliative	Curative	Systematic	Conditional	Predictive
Trigger event	Failure	Failure	Date/deadline	Limit or threshold crossing	Drifts, trends
Action service	Repair	Repair	Systematic replacement	Conditional replacement	Targeted intervention

Table 1.

Type of maintenance according to triggering event.

2.1.2 Evolution of maintenance

The effectiveness of the maintenance of industrial systems is a major economic issue for their commercial operation because the longer the maintenance phase, the more costly it is and generates unavailability of the system and a lack of productivity. The main difficulties and sources of inefficiency lie in the choice of type of maintenance (see Figure 3). The decision for a maintenance action is very complex and must be based on monitoring and intelligent analysis of the state of the system [1, 3].

Figure 3.

Evolution of maintenance.

2.2 Industrial maintenance technologies

Many companies supported by technological capabilities are improving their businesses. We see this in the various equipment maintenance processes, which are increasingly assisted by IT tools, cloud computing, Big Data, and the Internet of Things (IoT). Recently, IT and OT networks have significantly transformed industrial processes. Information technology (IT) and operational technologies (OT) are gradually transforming industrial organizations into digital businesses based on reliable data exchange.

One of the biggest influences on trends is Industry 4.0. Even before the pandemic, Industry 4.0 technologies were transforming the way manufacturers work. Today, the effectiveness of leaders in using technology is essential [5, 6, 7, 8]. With the advent of Industry 4.0 [9, 10] in manufacturing, businesses can leverage new technologies to monitor and better understand their operations in real time, transforming a typical manufacturing facility into a smart factory.

In the context of predictive maintenance [11, 12, 13], AI makes it possible to analyze data, which is already used by other users or installers to anticipate certain operations. However, these risks resulting in a change in intervention processes among field operators.

Figure 4 shows the benefits of predictive maintenance.

Figure 4.

Benefits of predictive maintenance.

2.3 Industrial maintenance methods

Today, technology plays an important role in many industries and helps speed up work processes. However, not all manufacturers are efficient with new technologies. To prepare your manufacturing business for changes and reshape it according to trends, engineers and managers must be open to technological capabilities. They must prepare their employees for the changes and the process of introducing and adapting to technology.

2.3.1 Classic maintenance methods

For many companies, managing industrial machine maintenance represents a major challenge. However, there are proven methods based on impeccable organization, fluid communication, and efficient technological tools. Figure 5 summarizes the most common classic methods in the field of industrial maintenance management [14].

Figure 5.

Classic methods of industrial maintenance.

These methods must be combined with CMMS software for efficient and smooth industrial maintenance of company equipment. The CMMS is considered a data centralization platform, and this maintenance solution makes it possible to manage the activity, edit reports, and monitor indicators in order to make more relevant decisions.

2.3.2 Prognostics and Health Management (PHM) as a new method of maintenance

Global performance requirements are leading manufacturers to strengthen their capacity to anticipate degradation phenomena and breakdowns. Subsequently, Prognosis and Health Management (PHM) solutions are increasingly implemented to complement maintenance activities [15, 16, 17]. Maintaining industrial systems in operational condition at a lower cost has become a critical factor. From the concept of PHM to predictive maintenance, it describes the emergence of this discipline, which complements traditional maintenance activities with a more proactive consideration of failures.

The overall principle of Prognostics and Health Management is to transform a set of raw data collected on the monitored equipment into health indicators, whose extrapolation over time makes it possible to define detailed decision support. There are several architectures in the field of Industry 4.0 among those that are developed and the best known are Open System Architecture for Condition-Based-Maintenance [5, 6, 7, 8, 9]. An overall view of this PHM architecture is given in Figure 6.

Figure 6.

OSA/CBM architecture [15, 16, 17].

This architecture is made up of seven functional layers:

Table 2 describes the seven layers of the OSA/CBM architecture.

Layer	Layer name	Layer description
L1	Data acquisition	This module provides the system with digital data from sensors or transducers. It covers the different areas of measurement, whether mechanical, electrical, destructive, and tribological.
L2	Signal processing	This module receives signals and data from sensors or, transducers or other signal processors, and performs signal processing through transformation and feature extraction.
L3	Monitoring	This module receives data from signal processing modules and other monitoring modules. It compares data with reference values and must also be able to generate alerts based on predefined thresholds.
L4	Diagnostic	This module determines whether the state of the monitored system or component is degraded or not and identifies the fault responsible for this degradation.
L5	Prognosis	This module relies on data from previous modules, and it predicts the future state of the monitored system and its components by projecting the current state of health of the system into the future.
L6	Help with the decision	This module aims to recommend maintenance actions or other alternatives linked to the management of the system to ensure the continuity of its operation.
L7	Human-machine interface (HMI)	This module builds an Human-Machine Interface (HMI) for the system, and it displays information on the health status of the system as well as alerts received from other previous modules.

Table 2.

The 7 layers of the OSA/CBM architecture.

The main objective of prognosis is to estimate the Residual Life (RUL) of a system by projecting the evolution of its state of health in the future at an early stage of degradation. Failure prognosis can be carried out using different methods using different modeling, processing, and analysis tools. These methods can be grouped into different categories or approaches such as physical model-based approach, data-driven approach, and experiment-based approach [18, 19].

3. General information on artificial intelligence

Artificial intelligence (AI) corresponds to a set of technologies that makes it possible to simulate intelligence and automatically accomplish tasks of perception, understanding, and decision-making. These techniques particularly involve the use of computer science, electronics, mathematics (notably statistics), neuroscience, and cognitive sciences.

3.1 Definition of artificial intelligence

Historically, work in AI began in the 1950s with the work of Alan TURING. AI became a field of research in the summer of 1956, during the first conference of the pioneers in this discipline, notably John MCCARTHY, Marvin MINSKY, Allen NEWELL, Herbert SIMON, and Donald MICHIE. Before 2000, the limits imposed by calculation and storage capacities did not allow significant progress to be made in the field of AI. We had to wait until the beginning of the 2000s to see the main factors of technological disruption that enabled current advances appear:

• The Internet network and the shared use of data have made it possible to create technologies such as search engines or decentralized and hyperscalable architectures.

• Exponential growth in the quantity of data: The storage space offered for €1 doubles every 14 months.

• Exponential growth in computing capacity: The total amount of data created each year doubles every 2 years.

• Mobility and the development of connected objects which promote access to real-time data flows: In 2020, there will be 50 billion connected objects, which will produce 10% of the total data created (Figure 7).

Figure 7.

Chronology of digital developments.

As a result, artificial intelligence [2, 20] has developed very strongly for more than 10 years with an acceleration in the last 5 years, to enable uses such as:

• Visual perception: recognition of an object or description of scenes.

• Understanding written or spoken natural language: automatic translation, automatic production of press articles, sentiment analysis.

• Automatic analysis by “understanding” a query and returning relevant results, even if that result does not contain the query words.

• Autonomous decision-making.

AI currently requires considerable data resources and computing power to learn effectively. Research is now developing techniques to reduce energy consumption and limit the need for data, and other techniques to make it possible to generalize a solution to several uses or to make AI robust in the face of an isolated disruptive event.

Apart from technological aspects, AI also poses new questions in terms of ethics and risk: dependence on automation, misuse, or error linked to “contaminated” data, impact on private life, etc. These questions require us to think about defining the trust framework to be put in place as these technologies develop. Even though artificial intelligence is mainly associated with a mathematical discipline and algorithmic techniques, it also includes other bricks to meet full use. The main building blocks of an AI system are shown in Figure 8.

Figure 8.

AI technological building blocks.

3.2 Algorithmic technologies

This section presents the algorithmic techniques of artificial intelligence, particularly the techniques of Machine Learning (see Figure 9). These algorithmic techniques make it possible to develop analysis, prediction, and decision-making capabilities and to intelligently and constantly adapt to situations based on data already acquired and currently being acquired [20].

Figure 9.

Timeline of artificial intelligence algorithmic techniques [21, 22, 23].

The Machine Learning [13, 24, 25, 26] process shown in Figure 10 begins by defining a cognitive task to automate. This can be a perception, comprehension, or decision task (by level of increasing complexity). We identify the flow data likely to respond to the problem posed. The problem to be solved and the type of data available often determine the type of AI algorithm that we will be able to use.

Figure 10.

Machine learning process [23, 27].

Furthermore, the quality of automation invariably relies on the following triptych: data, expertise in Machine Learning to model, and business expertise to define usage and interpret the results.

3.3 AI algorithms applied to maintenance

Artificial intelligence offers tools that are completely decoupled from the structure of the system, not requiring prior modeling of the latter and allowing real-time monitoring of its evolution, and its tools use several algorithms, among others, those applied for maintenance [22].

3.3.1 Methods based on behavioral models

There are two main approaches for building these models: finite state automata and Petri nets.

3.3.1.1 Finite state automata

See Figure 11.

Figure 11.

Very simple example of a finite automaton.

They make it possible to directly model the operation of the system, thanks to a global automaton obtained by composing elementary automata corresponding to local systems (system components). This representation is therefore directly adapted to simulation and detection. However, there are systems for which this representation is also used for diagnosis. A method presented in Refs. [14, 15] is characterized by two steps to perform the diagnosis.

3.3.1.2 Petri nets

See Figure 12.

Figure 12.

Example of a petri net.

The Petri net is a mathematical and graphic tool suitable for many applications where the notions of events and simultaneous evolutions are important. They are one of the most used models when it comes to discrete event systems. However, they have been enriched in several aspects (timed, stochastic, fuzzy RdP), so as to better account for the dynamics of discrete event systems. Used initially as generating models, they allow simulation to be carried out as well as detection with a view to use in system diagnostics. In this context, Petri nets can be qualified as models of good functioning. In Ref. [15], backward chaining type reasoning on Petri nets is defined.

3.3.1.3 Other formalisms

There are also other formalisms to be linked to methods based on behavioral models such as qualitative physics models which make it possible to obtain a model by abstraction from the numerical model [12] or approaches in classical or linear logic (also used with Petri nets) [2].

Finite state automata and Petri nets therefore constitute relatively well-suited tools for constructing detection mechanisms when the normal operation of the system is described by these formalisms. On the other hand, their uses in diagnosis are still limited. For automata, the main difficulties are linked to the large size of the state space, which therefore lead to problems with memory and speed of diagnostic execution. As highlighted in Ref. [2], Petri nets constitute a powerful modeling tool and can be considered as a tool for describing the knowledge necessary for diagnosis.

3.3.2 Pattern recognition methods for surveillance

These methods assume that no model is available to describe cause-and-effect relationships. The only knowledge is based on human expertise supported by solid feedback [24]. Most of these methods are based on artificial intelligence with, in particular, tools such as expert systems, statistical tools (pattern recognition), case-based reasoning (CAR), neural networks, fuzzy logic, and neuro-fuzzy networks.

3.3.2.1 Expert systems

See Figure 13.

Figure 13.

Expert system.

An expert system [23] is software that reproduces the behavior of a human expert performing an intellectual task in a specific domain. It is composed of two independent parts:

• a knowledge base, itself composed of a rule base which models the knowledge of the domain considered and a fact base containing the information concerning the case that we are currently processing.

• an inference engine capable of reasoning from the information contained in the knowledge base, of making deductions, etc.

3.3.2.2 Statistical tools for pattern recognition

The first technique presented is a classic probability-based discrimination technique. This technique may prove insufficient because it assumes a priori knowledge of all operating states and does not consider the evolution of the system [24].

3.3.2.3 Case-Based Reasoning—CBR

Case-Based Reasoning (CBR) is a recent approach to solving and learning problems. It corresponds to solving a new problem by remembering a previous similar situation and reusing information and knowledge from that situation [25]. The principle of operation of the method consists of storing previous experiences (cases) in memory to solve a new problem [21, 25, 26]:

• find the experience similar to the new problem in memory,

• reuse this experience in the context of the new situation (completely, partially, or by adapting it according to differences),

• memorize the new experience in memory (learning).

3.3.2.4 Case structure

The structure of the cases will depend on the areas of use and the tasks to be accomplished. Adapted to the diagnosis, the structure of the cases is therefore as follows:

• Problem ↔ symptoms (description of the diagnostic situation)

• Solution ↔ origins (several possible origins)

• Conclusion ↔ actions (maintenance strategy)

Some more recent work has been developed on the use of CBR for diagnosis [28]. In conclusion, CBR constitutes a technique for solving problems based on experience, and therefore relatively well-suited to diagnostic problems for which the notion of experience is relatively important.

3.3.3 Pattern recognition using neural networks

Neural networks are tools capable of performing perception, classification, and prediction operations. Their operation is based on the operating principles of biological neurons. Their main advantage over other tools is their ability to learn and generalize their knowledge to unknown inputs. One of the qualities of this type of tool is its suitability for the development of modern monitoring systems, capable of adapting to a complex system with multiple reconfigurations. Neural networks can also be implemented in electronic circuits, thus offering the possibility of real-time processing.

Their use is mainly guided by their following properties:

• learning ability,

• generalization ability,

• parallelism in processing (speed of processing),

• adapted to the nonlinearities of systems.

Each neuron performs a simple function (linear function, piecewise linear function, threshold function, sigmoid, Gaussian), with the global properties of the tool emerging from its structure. All the characteristics of neural networks are exploited through the main property of neural networks, which is learning. Indeed, learning mechanisms are at the origin of the problem-solving capabilities of neural networks. This learning makes it possible to configure the synaptic weights as well as the activation functions to adopt a desired behavior. Two types of learning are used: supervised learning and unsupervised learning [23, 27].

3.3.4 Pattern recognition using fuzzy logic

Fuzzy logic makes it possible to formalize the representation and processing of imprecise or approximate knowledge. It offers the possibility of dealing with highly complex systems in which, for example, human factors are present. It intervenes in the manipulation of imperfect knowledge. In these various applications, the use of fuzzy logic is quite natural, insofar as it makes it possible to deal with imprecision, uncertainty, and incompleteness linked to domain knowledge. In addition, fuzzy logic gives them the ability to be used in prognosis [28].

Neuro-fuzzy networks were born from the association of neural networks with fuzzy logic, so as to take advantage of the benefits of each of these two techniques. The main property of neuro-fuzzy networks is their ability to process digital and symbolic knowledge of a system in the same tool [28]. They, therefore, make it possible to exploit the learning capabilities of neural networks on the one hand and the reasoning capabilities of fuzzy logic on the other hand. Different combinations of these two artificial intelligence techniques exist and highlight different properties. The following combinations can be identified [28]: Neural fuzzy network, Neural/blur system simultaneously, Cooperative neuro-fuzzy models, and Hybrid neuro-fuzzy models.

Neuro-fuzzy structures for modeling, prediction, control, or diagnosis can be produced by a wide variety of architectures for the same type of given combination. For example, in Ref. [23], a use of a neuro-fuzzy system Recurrent Self-Adaptive Neuro-Fuzzy Inference System (R-SANFIS) for controlling an autonomous underwater vehicle. Another use of neuro-fuzzy networks is presented in Refs. [23, 27], where the NEuro Fuzzy function apPROXimator (NEFPROX) architecture is used for function approximation.

In diagnostic applications, we mainly find hybrid neuro-fuzzy models, for which neural network and fuzzy system are combined homogeneously.

3.3.5 Methods based on explanatory models

These methods are mainly based on representing the relationships between different fault states and their (possibly observable) effects. They are therefore based on a deep analysis of the system, so as to have sufficient knowledge to express its cause-and-effect relationships. The models thus obtained allow—for some—an abductive approach, which consists of going back to the causes of breakdowns from observations corresponding to the symptoms. Several artificial intelligence tools allow such formalization of the knowledge available on a system. These include causal graphs and contextual graphs, techniques which are also joined by approaches based on fuzzy logic or Petri nets.

3.3.5.1 Causal graphs

The exploitation of causal knowledge is quite natural for diagnosis. Indeed, a “dysfunction” can be quite simply described by the relationships associating its causes with its observable manifestations. Causal graphs constitute a formalism well-suited to the representation of these causal links. In diagnostic use, they make it possible to express the causal sequences governing the operation of the system to be monitored in the event of a breakdown [29].

The causal graph represents a particularly interesting tool for diagnosis in the sense that it can provide justification for the diagnosis proposed by the system through the causal path followed in the graph. Additionally, abductive diagnosis algorithms make it possible, from the observation of symptoms, to search for a set of possible causes that explain the observations through causal relationships. Finally, the introduction of temporal constraints, contradictory effects, and the consideration of interactions between failures more accurately reflect the physical reality of the system to be diagnosed.

3.3.5.2 Contextual graphs

Contextual graphs therefore appear to be a tool suitable for modeling activities involving a procedure/practice duality. They are therefore applicable in areas where an interpretation or adaptation of general rules is necessary to consider the richness of the real context of application. For diagnosis, they will be applied in areas where the causes of system failures are strongly linked to the context in which the failure occurred. In the context of a supervision application, they could be applied in cases where the context takes an important place in the link between fault diagnosis and recovery actions [30, 31].

4. Contribution of AI to maintenance

Thanks to artificial intelligence, all data collected by the sensors are analyzed in real time to find relationships between historical data and current readings, but also to alert technicians in the event of a risk of failure. Companies are taking a more proactive approach to their maintenance strategy. To do this, they rely on data from IoT sensors linked to CMMS software to monitor anomalies and use predictive models to request human intervention. This strategy notably avoids shutting down a production line when it is not useful and reduces corrective maintenance which is expensive and disrupts the activity [32, 33].

To make the transition to 4.0 maintenance, it is important to proceed gradually according to the starting level, the internal resources, and the appetite for digital technology. Indeed, each technological development involves profound changes that affect organization, teams, and technical infrastructure [33].

4.1 Toward maintenance 4.0

Maintenance 4.0 makes it possible to move from classic corrective maintenance to more intelligent predictive maintenance. Thanks to digital technologies, it is no longer possible to react to breakdowns but to anticipate them and deal with them before they occur. It allows maintenance costs to be optimized and product quality to be improved (see Figure 14) [34, 35].

Figure 14.

Toward maintenance 4.0.

4.1.1 Technological bases

The technological bases of predictive maintenance are based on advanced technologies and data processing capabilities [31]. Below are the essential elements that constitute predictive maintenance:

4.1.1.1 Sensors and data collection

Predictive maintenance relies on sensors integrated into machines and installations. The sensors continuously collect data on the condition of the machines, such as vibrations, temperature, pressure, flow rates, and much more. They collect data in real time and transmit them to data processing platforms (see Figure 15).

Figure 15.

Current communication levels.

The form of communication used until now, between sensors, control units and the process control, production, and business level, constitutes a closed system. Data are transmitted from field devices, namely sensors and actuators, to the programmable logic controller (PLC) (see Figure 16).

Figure 16.

Communication levels in the age of Industry 4.0.

Decentralized computing power converts data into information directly in the sensor. Decisions will be decentralized. Process, production, and business-relevant information will be transmitted directly to the Ethernet and the cloud (see Figure 17).

Figure 17.

Connected information.

In the future, the cloud will gain ground in general process management. But core computing power will increasingly shift to the edge. The sensors convert the collected data into information which is then processed in the Ethernet or cloud for the next process.

4.1.1.2 IoT (Internet of Things)

The Internet of Things [23] plays a crucial role in enabling the networking of sensors, machines, and installations and the transparent transmission of data to central platforms [12, 18, 36, 37] and processing of data or cloud-based systems is thus guaranteed (see Figure 18).

Figure 18.

Connected objects and the highlight [34].

The Internet of Things (IoT) is radically transforming the way businesses approach managing their data and IT systems. With each passing day, users generate a massive amount of data, reaching up to 2.5 quintillion bytes, creating a significant challenge to stay relevant in an ever-changing digital environment. However, IoT is emerging as a key solution to making sense of this abundance of data by introducing automation where it is needed.

The Workflow Management Software is specifically designed to aggregate data from these connected devices, bringing significant improvements in both the professional and personal spheres:

• Data processing and analysis

• Algorithms and models

• Predictive analytics

• Integration into business systems

• Continuous learning

• Data entry and processing

• Data analysis and modeling

Data analysis and modeling allow businesses to identify patterns and anomalies in their data and respond quickly to potential issues before unplanned downtime occurs. This helps to improve plant availability and reduce maintenance costs.

4.1.2 Implementation of predictive maintenance

Implementing predictive maintenance requires a well-thought-out strategy and approach structured. Here are the steps businesses should take to implement maintenance [31, 38, 39] predictive (see Table 3).

Steps	Description of steps
Define goals and requirements	Define clear objectives for the implementation of predictive maintenance. Identify the specific machines or installations for which predictive maintenance must be implemented.
Identify data sources	Identify relevant data sources and sensors needed for machine health monitoring. Ensure that data can be collected and transmitted in real time to a central platform or system.
Set up a data infrastructure	Establish a robust data infrastructure that enables data collection, storage, and processing. Take security and data protection regulations into account when processing data.
Ensuring data quality	Monitor and maintain the quality of collected data to ensure it is suitable for analysis. This may include data cleaning and noise removal.
Choosing analysis and modeling techniques	Decide which analysis and modeling techniques best suit your needs. This could be machine learning, statistical models, or a combination of both.
Model development and training	Develop and train models based on historical data. Use these models to monitor machine health in real-time.
Set thresholds and alarms	Set thresholds and criteria that determine when alarms or notifications trigger. This helps identify problems proactively.
Integration into existing processes	Integrate predictive maintenance into your existing maintenance and operations processes. Ensure that maintenance personnel can effectively use information and alarms.
Training and awareness	Train your team in the use of predictive maintenance tools and systems. Raise your employees’ awareness of the importance of the new strategy and how it helps improve efficiency.
Monitoring and optimization	Establish a continuous monitoring system to ensure that the predictive maintenance strategy is effective. Continuously optimize models and algorithms to improve prediction accuracy.
Measuring success	Define clear key performance indicators (KPIs) to measure implementation success. These can include indicators such as reduced downtime, reduced maintenance costs, and increased plant availability.
Data ethics and data protection	Data ethics and data protection must be respected throughout the process, especially when it comes to data collection and storage.

Table 3.

Steps for implementing predictive maintenance.

4.2 AI contribution statistics for maintenance

Global industry, particularly the aeronautics sector and the rail transport sector, is booming thanks to Maintenance 4.0 linked to technological changes, notably the integration of IoT equipment and associated data [33, 40].

For many industries, predictive maintenance remains a competitive and profitable solution in terms of investment. They have started to reap the benefits thanks to the integration of new AI technologies, which have changed and continue to change the business landscape and industrial system technology. Figure 19 shows the contribution shares of predictive maintenance by sector of activity:

Figure 19.

Contribution of predictive maintenance by sector of activity. Source: JDN Journal Du Net/Predictive maintenance market, Global forecast to 2021.

Companies that have fully adopted AI-powered software and are committed to adopting AI highlight several benefits that show how AI has changed the game. According to “Grand View Research,” the global artificial intelligence market is valued at $136 billion.

• Growth of 1400% is forecast for the next 7 years.

• In 2030, this market is expected to be valued at more than $1.81 trillion.

• In 2018, the market generated only $10 billion annually.

• In 2025, 97 million people will work in the field of artificial intelligence.

Figure 20 presents the result of company statistics carried out by “MIT Sloan Management” and “Sales Forces” in answering the question on the use of artificial intelligence and its applications, and their answers were as follows:

Figure 20.

Statistics result.

4.2.1 Impact of AI on the industrial sector

AI technology is slowly penetrating all spheres of our lives and is being incorporated into all kinds of devices and software. In 2023, AI is everywhere around us: in the industrial sector thanks to maintenance 4.0, autonomous cars are already much more than a simple element of a science fiction film, virtual assistants are efficient and credible, and countless software uses the technique of machine learning [41, 42]. As technology advances, AI capabilities increase and become not only more practical for the industry but also a new standard.

According to statistics carried out by “Passport-Photo.online” shows that 78% of IT managers are already using AI or planning to do so to automate workflow. After all, its abilities, at the very least, make life significantly easier in handling tasks that would otherwise be time-consuming and arduous for the human mind. Data management alone is one piece that confirms the AI is truly a game-changer in how businesses operate and take advantage of new possibilities, not to mention machine learning [43, 44, 45].

4.2.2 Benefits of artificial intelligence

In recent years, considering the rise of AI and its omnipresence in daily life, we can affirm that this development would not have taken place if there had been only a handful of advantages of its implementation. The latest AI statistics show that the path to success as a business owner is to immerse yourself in AI’s capabilities and exploit them before the competition does. It is particularly interesting to note that the first benefit mentioned is improving the customer experience, something that many observers believe should be at the top of a company’s priorities. The same study shows that most people (63%) believe that AI will help solve the problems of modern society and help us live more fulfilling lives. This is another very positive and remarkable result that counters the fear that AI will become a problem in the future [46]. Integrating AI into predictive maintenance has many benefits:

• Optimized equipment or system availability

• Getting closer to operational excellence through “zero breakdown”

• Reduced maintenance costs

• Optimization of the production chain by planning machine maintenance operations at the right time

• Reduction in breakdowns

• Improvement in the supply of spare parts to limit excess stock.

• Increased lifespan of operating assets

• Increased customer satisfaction through quality service on time

4.3 Artificial intelligence trends

By following a structured approach to implementation and leveraging the right technology, processes, and expertise, organizations can realize significant benefits in terms of equipment reliability, maintenance efficiency, and operational performance [47, 48, 49]. Since AI has the capacity to analyze considerable quantities of data that would be impossible to analyze manually, its integration into industrial systems allows companies to have advanced tools to optimize the reliability and efficiency of different manufacturing operations maintenance, such as planning interventions, purchasing spare parts in advance, or wasting raw materials and energy. The primary objective is always to avoid costly production interruptions. Data analysis is truly a key element for real-time detection of technical problems. By using AI to analyze data from sensors and machines, anomalies and potential issues are quickly identified. This allows you to react as quickly as possible and reduce operational interruptions [48].

4.3.1 Success stories of the contribution of AI to predictive maintenance

AI is a powerful tool that can significantly increase the effectiveness of predictive maintenance by allowing companies to monitor the health of their equipment, detect possible future failures, and quickly notify maintenance teams of a check to be carried out so that they can plan interventions and avoid possible costly shutdowns.

The reduction in corrective and curative maintenance generates real financial gain for companies. By using information from sensors and equipment performance, AI can therefore help identify equipment that requires more frequent or intensive intervention [48, 49, 50].

As follows, some case studies and success stories are given [46, 48, 51].

4.3.1.1 General Electric (GE) Aviation

GE Aviation implemented a predictive maintenance program called “Prognostic Health Management” (PHM) for aircraft engines. By equipping engines with sensors to collect real-time performance data, GE Aviation can monitor engine health, predict potential failures, and schedule maintenance proactively. As a result, airlines can avoid unscheduled downtime, reduce maintenance costs, and optimize fleet operations. For example, GE Aviation’s PHM system helped Cathay Pacific Airways reduce engine related delays by 35% and decrease maintenance costs by 10%.

4.3.1.2 Schneider Electric

Schneider Electric, a global leader in energy management and automation solutions, implemented predictive maintenance for its electrical distribution equipment. By analyzing data from sensors embedded in switchgear, circuit breakers, and transformers, Schneider Electric can detect early signs of equipment degradation or malfunction and schedule maintenance proactively. As a result, customers experience increased equipment reliability, reduced downtime, and improved safety. For example, a manufacturing facility in Europe reduced equipment downtime by 20% and achieved a return on investment (ROI) within 6 months of implementing Schneider Electric’s predictive maintenance solution.

4.3.1.3 Rio Tinto

Rio Tinto, a multinational mining corporation, implemented predictive maintenance for its fleet of autonomous haul trucks used in mining operations. By deploying sensors and predictive analytics software, Rio Tinto can monitor truck performance, identify potential issues, and schedule maintenance proactively to avoid costly breakdowns. As a result, Rio Tinto has increased equipment availability, improved productivity, and reduced maintenance costs. For example, predictive maintenance helped Rio Tinto achieve a 10% increase in truck availability and a 15% reduction in maintenance costs at its Pilbara iron ore mine in Australia.

4.3.1.4 Siemens Gamesa Renewable Energy

Siemens Gamesa Renewable Energy [46, 51], a leading manufacturer of wind turbines, implemented predictive maintenance for its wind turbine fleet to optimize performance and reduce maintenance costs. By analyzing data from sensors embedded in turbines, Siemens Gamesa can detect early signs of component wear, identify potential failures, and schedule maintenance proactively. As a result, wind farm operators experience increased turbine availability, higher energy production, and reduced operational costs. For example, predictive maintenance helped a wind farm operator in Europe achieve a 20% reduction in maintenance costs and a 10% increase in energy production.

4.3.1.5 Pacific Gas and Electric Company (PG&E)

PG&E, a utility company serving millions of customers in California, implemented predictive maintenance for its electrical distribution network to improve reliability and reduce outage duration. By analyzing data from smart meters, sensors, and weather forecasts, PG&E can identify potential equipment failures, prioritize maintenance activities, and deploy crews proactively to restore service quickly in the event of an outage. As a result, PG&E has achieved significant improvements in grid reliability, customer satisfaction, and operational efficiency. For example, PG&E reduced outage frequency by 15% and outage duration by 20% within the first year of implementing predictive maintenance.

These case studies demonstrate the diverse applications and benefits of predictive maintenance across different industries, including aviation, manufacturing, mining, renewable energy, and utilities. By leveraging advanced technologies and data analytics capabilities, organizations can optimize maintenance practices, enhance asset reliability, and drive operational excellence, ultimately delivering value to customers and stakeholders [46, 51].

4.3.2 Trends in industrial maintenance with the exploitation of AI

Artificial intelligence technologies will be increasingly used in the field of maintenance and Industry 4.0. One of the major axes for the application of AI in industry is the quality and quantity of available information. For AI to work effectively, it is necessary to have high-quality and sufficient information to train and power the models and algorithms.

This is particularly one of the challenges with CMMS software, which must allow maintenance services to easily collect and store important information. AI is of great importance to industries, and the maintenance and field intervention sectors are obviously no exception. Indeed, AI makes it possible to optimize processes to make more efficient decisions based on clear and precise data. The proof, according to a report recently released by Bitkom statistics, is that half of the companies specializing in the sector believe that the gain in productivity is the main advantage associated with the use of artificial intelligence. It is a fact: AI can optimize the management and maintenance of industrial systems.

1. Advanced machine learning techniques: future advancements in machine learning techniques are expected to drive improvements in predictive maintenance capabilities. Deep learning architectures, such as deep neural networks (DNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs), will continue to be explored for their potential to extract complex patterns and relationships from sensor data, enabling more accurate fault detection and diagnosis.

2. Edge computing and IoT integration: the integration of edge computing and Internet of Things (IoT) technologies will enable real-time processing and analysis of sensor data at the network edge, reducing latency and bandwidth requirements. Edge-based predictive maintenance solutions will allow organizations to perform data analytics and decision-making closer to the source of data generation, enabling faster response to equipment anomalies and minimizing reliance on centralized cloud infrastructure.

3. Digital twins and simulation modeling: digital twin technology, which creates virtual replicas or models of physical assets, will play a key role in predictive maintenance by enabling simulation-based modeling and predictive analytics. By coupling real-time data from sensors with virtual representations of equipment, organizations can simulate different maintenance scenarios, predict the impact of interventions, and optimize maintenance strategies to maximize asset performance and reliability.

4. Explainable AI and interpretability: as AI-based predictive maintenance solutions become more pervasive, there will be a growing emphasis on explainability and interpretability of predictive models. Explainable AI techniques, such as model-agnostic methods, feature importance analysis, and visualization tools, will enable stakeholders to understand the rationale behind model predictions, build trust in AI systems, and facilitate human-in-the-loop decision-making.

5. Predictive Maintenance as a Service (PMaaS): the rise of cloud computing and subscription-based models will pave the way for Predictive Maintenance as a Service (PMaaS) offerings. PMaaS providers will offer scalable, cost-effective predictive maintenance solutions hosted on cloud platforms, allowing organizations to access advanced analytics capabilities, predictive models, and expertise without the need for extensive infrastructure investment or in-house data science resources.

6. Autonomous maintenance systems: advancements in artificial intelligence, robotics, and autonomous systems will enable the development of autonomous maintenance systems capable of self-diagnosis, self-repair, and self-optimization. These systems will leverage AI-based algorithms to continuously monitor equipment health, detect anomalies, and perform maintenance tasks autonomously, reducing the need for human intervention and improving overall operational efficiency.

7. Integration with Industry 4.0 initiatives: predictive maintenance will become an integral component of Industry 4.0 initiatives aimed at digitizing and optimizing manufacturing processes. Integration with technologies such as cyber-physical systems, additive manufacturing, and augmented reality will enable seamless data exchange, real-time monitoring, and adaptive maintenance strategies, leading to increased productivity, reduced downtime, and enhanced competitiveness.

8. Ethical and responsible AI practices: with the proliferation of AI-based predictive maintenance solutions, there will be a growing emphasis on ethical and responsible AI practices. Organizations will need to address concerns related to data privacy, bias, fairness, and accountability in AI systems to ensure that predictive maintenance initiatives are deployed in a socially responsible manner and aligned with ethical principles and regulatory requirements.

In summary, the future of AI-based predictive maintenance will be characterized by advancements in machine learning techniques, edge computing, digital twin technology, and autonomous systems, enabling organizations to achieve higher levels of asset reliability, operational efficiency, and sustainability. By embracing emerging trends and leveraging innovative technologies, organizations can stay ahead of the curve and unlock new opportunities for predictive maintenance optimization and innovation.

5. Conclusion

In this chapter, we have covered the growing integration of artificial intelligence in the field of mechanical maintenance, which offers significant benefits in terms of operational efficiency, cost reduction, and improved equipment reliability. Using techniques such as machine learning and predictive analytics, AI-based systems can anticipate failures, optimize maintenance plans, and extend the useful life of machines.

We explored how AI can be used to optimize the operating parameters of industrial systems equipment. This approach plays an increasingly important role in improving the management of intervention methods and increasing the efficiency of operations starting from fault detection to operations planning, including in-depth data analysis to propose controls and plan maintenance interventions at the best time.

We discussed the contribution of AI to industrial maintenance, thanks to connected devices and the exploitation of new generations of CMMS, moving from classic reactive maintenance to more intelligent and proactive maintenance which will offer real benefits to industrial companies in terms of the performance of their production systems.

We also covered the different basic notions of industrial maintenance through its concepts, technologies, and methods used. However, to take full advantage of these methods and technologies, it is essential to develop specialized skills, ensure data quality, and maintain a balance between automation and human intervention.

In our future work and to strengthen the decision-making reliability of our predictive maintenance system, we will focus on a case study by proposing a method of diagnosis and prognosis of the imbalance fault of a rotating machine using the techniques and algorithms of artificial intelligence and machine learning by highlighting their advantages over traditional techniques of vibration analysis and diagnosis, mainly temporal analysis and frequency analysis of signals.

Ultimately, AI represents a powerful tool to modernize and improve maintenance processes in the mechanical sector, thus helping to increase the productivity and competitiveness of companies.

Conflict of interest

The authors declare no conflict of interest.