An error, a failure, a sudden malfunction. Imagining the unexpected is the goal of the predictive maintenance system that provides manufacturing companies with a prediction of a future failure of physical assets — generally machinery — in order to significantly reduce downtime periods during the production process. "A new maintenance philosophy," as Davide Chiaroni, professor at the Polytechnic Institute of Milan, defines it, that can take advantage of new tools such as IoT sensors, able to collect large volumes of data on which to conduct studies with machine learning and predictive analysis tools.
Monitoring, measurement and predictive maintenance
"Predictive maintenance is measurement-based," explained Chiaroni, MIP project leader of the Schmidt MacArthur Fellowship focused on the circular economy since 2013. "Instead of focusing on a certain assumption about the operation of a plant or asset, it puts in place a series of practical, continuous measures to monitor the state of the asset, allowing for intervention." In the past, standard maintenance, carried out by individuals, was the methodology used to check the condition of a given piece of machinery. A concept similar to the overhauling of automobiles. "Condition-based maintenance and predictive maintenance are two substantially recent methods currently adopted to perform maintenance," explained Emanuele Fabbiani, chief data scientist at xtream. "Both are based on the same basic concept but in the CBM case, the idea is to try to do maintenance when the sensors tell me it’s necessary. Instead, predictive maintenance goes a step further: I have models that tell me when the machinery will fail, so I’m able to predict the evolution of the state of the asset."
There are several benefits to predictive maintenance and they all positively affect primarily from an economic standpoint, significantly reducing the costs associated with machine failures and the time spent maintaining an asset. In addition, by monitoring the health of machinery in real time, its useful lifespan can be lengthened. According to Statista Research Department, the global predictive maintenance market is expected to reach around $23.5 billion by 2024 with an annual growth rate of nearly 40 percent between 2018 and 2024.
The importance of machine learning able to improve from mistakes
The progressive diffusion of IoT devices and applications, together with the availability of advanced analytical tools and the emergence of artificial intelligence and machine learning technologies, make it finally possible to integrate different types of sensors into industrial machinery, and to network such equipment. "The internet of things allows to link any type of device connected to the network," added Fabbiani. "Based on the data collected, the central system processes predictive models that allow optimizing the maintenance strategy."
The global predictive maintenance market is expected to reach around $23.5 billion by 2024 with an annual growth rate of nearly 40% between 2018 and 2024.
In case of predictive maintenance, IoT sensors aggregate data, and then machine learning will cross-reference this data to let us know what needs to be fixed. "There are various ways of working," said Chiaroni. "You don't necessarily need machine learning to do maintenance, but certainly if I want to perform a self-adaptive maintenance method that benchmarks — meaning it takes into account that the same machine is installed in different places with different usage characteristics — machine learning becomes fundamental." Mathematical and engineering models, however, are still very valuable: "Everything is still evolving," Fabbiani told Renewable Matter, "but the hybrid way that combines physical and engineering knowledge with machine learning modeling seems to be the most promising." Cybersecurity is another issue: "Operational costs are a thing, but if these practices are done correctly and placed in a proper operational context, the cost of all this is much lower than the risks of economically burdensome downtime or failures."
Upsides in the automotive sector
Currently, predictive maintenance is a methodology applied mainly in the B2B (business to business) market. Considering the cost-benefit ratio, it’s rare for this type of maintenance to be applied to consumer products. "The mass market today is still experiencing some difficulties," explained Chiaroni. "On one hand, it's not easy to charge for maintenance services. On the other hand, where operations are more fragmented, like they would be for a washing machine, it’s economically complicated to carry out predictive maintenance." The theme of maintenance on components, however, is also very present in the automotive sector, where the components of the final product are evaluated according to predictive models.
"Today we associate predictive maintenance with the automotive sector in the context of circular economy," said Chiaroni. "It’s done with engines, for example: Rolls-Royce does it by applying a service of engine rental, even on aircrafts, where maintenance played out in a predictive perspective is included. The logic of reuse, reutilization and remanufacturing makes it possible to intervene selectively on a machine. And perhaps even reusing it after an appropriate configuration." A methodology that brings undoubted upsides, both on the user side, where you have the possibility of reducing machine downtime and obtaining better performance from the car, and on the manufacturer side, by being able to intervene in a predictive manner and monitor the life of the machine and its components.
The maintenance frontier for offshore turbines
The interesting game of predictive maintenance for now is being played mainly between large infrastructures, particularly in the energy field. Scientific research has set its sights on the offshore wind sector to solve the huge practical problems that wind turbines present when it comes to maintenance. Franz Langmayr knows it well, as managing director of UPTIME Engineering, currently working on the Romeo consortium, a project supported by the EU through its Horizon 2020 program. Through the development of advanced technology solutions, the Romeo team is working to reduce the cost of operating and maintaining offshore wind farms. "In the offshore sector, predictive maintenance is particularly important," explained Langmayr, "because unlike onshore conditions, there is no easy access to the turbines. When a failure occurs in the cold season, often enough the turbines are not accessible for months at a time." Turbine downtime, often caused by easily solvable malfunctions, causes significant economic damage.
"When an error occurs, trips that last half a day on the sea end up being necessary — a huge amount of energy is wasted to fix a problem. A lot of field research is needed because currently the energy capacity of offshore turbines is far greater than onshore turbines, but the fault lies in the amount of energy wasted."
Only in the last 4 to 5 years people have been starting to apply machine learning to predictive maintenance.
The basic idea of the project is to collect all relevant data about the state of turbines, send it to a central ecosystem and, through analytical tools and machine learning, figure out what they mean. "The challenge is to collect them in a way that they can be understood enough to generate results," added Langmayr, "doing predictive maintenance has become essential in this industry. If, for example, a gust of wind were to suddenly deform a wind blade, we could control the energy in order not to overheat the gear ratio of the turbine motor." This type of maintenance also allows us to understand if certain components are nearing the end of their life or if and how they might fail.
Oil and gas
Predictive maintenance has been a focus in the Oil & Gas industry for a few years. According to an assessment of the Predictive Maintenance in Oil & Gas report published by GlobalData Thematic Research, oil price volatility is driving increased adoption of predictive maintenance technologies to help companies reduce operating expenses by optimizing maintenance planning and increasing productivity.
"Only in the last four to five years people have been starting to apply machine learning to predictive maintenance," explained Massimiliano Conti, Energy Sme senior digital adviser. "The oil and gas industry is 100 years old, so it will take a while to actually measure the benefits of predictive maintenance." The equipment and infrastructure currently in use for oil and gas operations was built several decades ago. These machines were based on designs, materials and technologies available at that time. Such outdated equipment is bound to fail and therefore requires constant inspection and monitoring. "From a technical point of view," Conti continued, "maintenance concerns rotary machines (compressors, pumps, etc.), which serve the process of crude oil extraction, transportation and refining. When fluids are moved, the machines are subject to failures that lead to slowdowns or even stoppages. Obviously, these have economic and environmental impacts. Halting a chemical process that doesn't stop according to an established protocol can cause out-of-control CO2 emissions into the atmosphere.”
A proactive approach such as predictive maintenance, even in an industry likely to be replaced by renewable energies in the future, is a more reliable and safer way to manage large plants.