Maintenance optimisation and planning represent a major challenge for industry. Armelle Le Gall, Head of the Design Department and an expert in optimisation in system design and maintenance, explains how EURODECISION’s expertise can assist maintenance professionals in their work and shares her thoughts on the sector’s forthcoming challenges.
Why is equipment maintenance planning important?
Corrective maintenance, which involves correcting a failure on a piece of equipment, is costly, particularly because it requires equipment shutdown, which can degrade service. To minimise the impact on operations and reduce overall costs, maintenance professionals plan maintenance operations optimally by following preventive maintenance cycles while respecting regulatory constraints and maintenance standards for different equipment. Effective organisation also involves accounting for maintenance resource constraints, whether human or material (e.g., workshop track occupancy, facility accommodation capacity).
How can decision mathematics contribute to optimising maintenance planning?
The optimisation techniques that are EURODECISION’s area of expertise are well-suited to this highly constrained planning problem. Based on the planning horizon, the number of equipment items, and the specific constraints, we can select the appropriate algorithmic solution from the available options. Our mathematical models enable optimised maintenance planning by reducing operational impacts, dimensioning maintenance resources, evaluating different scenarios, and proposing operational guides for various equipment items.
For example, in the case of a fleet of uniform equipment whose commissioning was simultaneous (e.g., the opening of a new metro line), avoiding similar usage across all trains should be avoided to smooth, over the long term, renovation and maintenance operations that entail immobilisation for several weeks. This results in greater complexity for maintenance planning, and EURODECISION’s models can help optimise short- and medium/long-term planning.
EURODECISION is also consulted on “predictive maintenance” subjects. What exactly does this involve?
Predictive maintenance is dynamic preventive maintenance: it is based on wear measurement histories, obtained, for example, via connected sensors (IoT), to estimate the remaining useful life (RUL) of equipment. The objective is to plan maintenance operations as precisely as possible based on detected wear signs or anomalies.
Compared with preventive maintenance based on average wear, predictive maintenance enables equipment to be examined individually. Thus, equipment can sometimes be used longer before a maintenance intervention is necessary, and at other times be repaired or replaced earlier than average, thereby avoiding a breakdown and service interruption.
Manufacturers are clearly interested because this represents significant potential gains. However, the greater the predictive element, the more difficult it is to have a long-term planning overview because equipment shutdowns must be managed on a case-by-case basis. There is, therefore, a real issue here that requires a complete rethink of the maintenance organisation.
Could you tell us more about the techniques EURODECISION uses for predictive maintenance?
Machine learning approaches are very well suited to predictive maintenance problems. Indeed, this artificial intelligence technology is particularly appropriate for time-series forecasting, and we therefore seek to integrate it into our models to predict the remaining time before breakage.
However, in predictive maintenance, one must not overestimate equipment’s remaining useful life, because predicting a breakdown immediately after the actual breakdown serves no purpose, even if the error is very small. Yet, in machine learning, error functions are generally symmetrical. That is to say that an overestimation error has the same importance as an underestimation error. We therefore use modelling of asymmetrical error functions in neural networks (using Scikit-learn and TensorFlow libraries). The initial results we have obtained are promising!
What do you find appealing about this type of problem?
Maintenance planning is a complex subject that poses a major challenge for industry, and I am convinced that our algorithmic expertise can provide significant added value. When I began taking an interest in the subject, I was surprised by the power dynamic that can exist between the operator and the maintenance provider, the latter having to carry out their mission under very demanding constraints, or face penalties. I therefore find it particularly interesting to support them in resolving this very business-specific problem.
To find out more about artificial intelligence and decision mathematics in the service of maintenance: