A trend towards more accurate and real-time decision-making algorithms has been revealed by this article in the context of the rise of intelligent management technological platforms. The increasing availability of industrial sensors and big data technologies has led to more responsive information systems capable of supporting efficient decisions with an automated system approach (even ahead of time). To this end, predictive maintenance algorithms can significantly benefit from the further exploitation of information systems engineering by implementing big data technologies and algorithms. Regarding our future work, we aim to extend the literature review in the following directions,
- Further analyze the methods and algorithms of the reviewed papers
- Determine the role of operational research, data analytics, and artificial intelligence in predictive maintenance decision making in smart factories
- Investigate information systems engineering-related aspects e.g. architectures, interoperability, etc. and,
- Assess the choices-making algorithms in prototype and industrial predictive maintenance platforms
The Scope of Predictive Maintenance
The predictive maintenance indicates the phase triggered by sensor-driven (near) real-time predictions (e.g., about a future failure mode) to get proactive recommendations about maintenance actions and plans that eliminate or mitigate the impact of the anticipated failure. It has led to extensive use of sensors with automated systems for condition monitoring, which facilitates decision-making under time constraints has shown much efficacy to that of human-centered design. The P-F interval is, therefore, the time between the point where a potential failure occurs and the point at which it deteriorates into a functional failure, which can be seen as an opportunity window during which decision-making algorithms can recommend actions to eliminate the anticipated functional failure or mitigate its effect.
Limitations of Predictive Maintenance
We often refer to limitations as the inherent inflexibility of methodologies, regardless of whether they are processes or strategies. Following are some limitations with Predictive Maintenance,
- Dynamic and static models, such as offline and real-time, are not separated
- There is no focus on predictive maintenance in their various maintenance strategies
- The Decision Making is not necessarily triggered by predictions
- They focus on specific categories of decision methods, such as optimization, and/or maintenance aspects, such as maintenance policy
Analysis and Synthesis
The increasing complexity and uncertainty of the manufacturing environment have leveraged the emergence of several methods and algorithms aiming to better support decision-making. Smart decision making is a core aspect that can be enabled with an automated system processing sensor data. However, the uncertainty existing in predictive analytics but also in the degradation process itself and the time constraints under which a decision should be taken pose challenges in the applicability of the decision-making algorithms. During recent years, with the emergence of predictive maintenance has remained a keystone for maintenance management, there is an increasing interest in algorithms aiming to better support maintenance decisions that improve equipment operating life. In order to facilitate the comprehension and the investigation of the existing decision-making algorithms, it has been categorized into five areas decision making these areas are,
- Maintenance Planning and Scheduling
- Reliability and Degradation-based Decision Making
- Joint Optimization
- Maintenance Cost, Risk Estimation and Optimization
- Multi-State and Multi-Component Systems Optimization
Several decision-making algorithms for predictive maintenance are based upon model-based automated systems or human-centered design prognostic algorithms instead of data-driven ones. Therefore, the associated deciding methods and algorithms are mainly knowledge-based with insufficient knowledge analytics capabilities. The probabilistic nature of the degradation process makes decision-making for predictive maintenance highly uncertain and complex. For this reason, a large amount of existing decision-making algorithms utilize simulation models or iterative solution procedures. Only simple models are usually involved with exact solutions. Moreover, sometimes simulation is combined with more advanced optimization techniques in order to decrease the computational effort and provide more reliable results.
Effective Research Ideas for the Future
Development of scalable and efficient (near) real-time decision-making methods and algorithms instead of processing batches of data
This facet has both a technological use of appropriate technologies (automated systems) e.g., distributed computing for streaming and heterogeneous data, and functional use of appropriate decision models, e.g., recursive and computationally efficient, probabilistic methods in an exceedingly streaming context with the aim to tackle uncertainty perspective.
Development of generic decision models representing the decision-making process instead of the physical process
They will be applicable to any physical model, manufacturing equipment, and production process by utilizing prescriptive analytics, artificial intelligence, and machine learning algorithms on the basis of the large availability of real-time sensor-generated and historical logs maintenance-related data. They will be ready to recommend perfect and imperfect maintenance actions, taking into consideration other operations so as to enhance the general business performance.
Development of data-driven techniques for building the decision models
There is the capability of automated data-driven model building and finding appropriate patterns, instead of manually building the decision model by the expert according to their knowledge, the physical model, or the industry. There is a clear trend in literature, currently mainly at a conceptual level, towards less human intervention (e.g., information based on expert judgment) in decision making by conducting advanced big data analytics.
Development of feedback mechanisms for improving the decision-making algorithms
Frequently, in the dynamic, sensor-driven manufacturing environment, the problem sets will change rapidly, for example when new constraints need to be added, and the model may need to be re-built, or when there is an unobservable change and the model is no longer valid. Despite widespread awareness, the mechanisms for tracking the recommendations are underexplored for diagnosis and prognostic algorithms. Another aspect of the recommendations not taken into account is feedback from humans, such as engineers and operators about their suitability.
The use cases that illustrate practical inferences from Predictive Maintenance
Use Case of Connected Car
If we take the automobile industry into consideration. Today’s connected cars create and cascade vast amounts of performance data from sensors spread throughout the vehicle. This information directly goes to the manufacturers or car dealerships, who can then alert drivers of any issues that require servicing before they experience the inconvenience of their car breaking down.
Use Case of Utility Suppliers
In other sectors, companies are looking ahead to predictive maintenance to work smarter internally. Utility suppliers are applying predictive analytics to the big data generated by smart meters to be proactive in detecting early warning signs of supply and demand issues on the grid and address them before they lead to service interruption. This not only saves them from costly repairs but also helps them avoid customer dissatisfaction.
Use Case of Insurance
Let us assume that the insurance industry will benefit from more robust predictive analytics around the likelihood and impact of extreme weather conditions. Supermarkets and their associated suppliers could improve crop yield and production operations by relying on accurate predictions. The opportunities are virtually endless.
Innovations in IoT have resulted in significant disruptions in manufacturing strategies due to emerging technological advancements. Based on this, the concepts such as “smart manufacturing” and “digital factory” have arisen. In these contexts, predictive maintenance is becoming increasingly important to reducing costs and raising business performance because its predictive model either an automated system or human-centered design, is able to detect abnormal behaviors of equipment and predict future failure modes to support decision making in advance.