Using advanced analytics algorithms and machine learning, predictive maintenance searches for relations and correlations between source behaviour, its environment, subsequent failures or downtime in order to determine an optimal maintenance plan.
The rise of digital technologies in production and the phenomenon of the Internet Of Things (IoT) pushes the boundaries of what is possible and presents maintenance specialists with the challenge of predictive maintenance. Our project experience tells us that predictive maintenance may reduce the time spent on maintenance planning by 20-50%, increase the efficient and effective utilisation of machinery and equipment by 10-20%, and reduce total maintenance costs by 5-10%.
Predictive maintenance simply refers to the fact that service shutdowns come neither too late (when a breakdown occurs) nor too soon (the machine is all right and there is no reason for a shutdown) but at the right moment (when the risk of a breakdown exceeds the accepted threshold). As the shutdown of a single machine has repercussions for the entire line and production chain, service planning is optimised with a view to the whole production process, eg by combining several shutdowns in a single time frame outside the busy hours. The concept of predictive maintenance is decades old, but it can be effectively and cheaply implemented only now when the physical and digital worlds form a functional whole.
An 80% decrease in unplanned shutdowns during the first three months of operation. A press for plastic products.
- Early warning against an imminent breakdown;
- Increase in the effective and efficient utilisation of machinery and equipment (reduction of shutdowns);
- 24/7 availability;
- Better utilisation of service time;
- Guaranteed availability of spare parts;
- Less work with the planning of service and shutdowns; and
- Decrease in total maintenance costs.
The introduction of predictive maintenance has four basic stages:
1. Data Generation and Collection
Physical, topographical and operating data, data about breakdowns and shutdowns, construction data. Data sources include databases, sensors, SCADA systems and ERP systems. Data needs to be combined in one pile, sorted out, interconnected, cleansed and prepared for machine learning.
2. Modelling and Seeking Interdependencies
Using advanced analytics and machine learning algorithms, we seek relations and correlations between the behaviour of the machine and its surroundings and the subsequent breakdown or shutdown. We seek time sequences and examine for which machines the same model may be used and for which a different one is needed.
3. Introduction in Production
The finding of models and interdependencies gives rise to an algorithm for determining the optimal maintenance plan, which will be programmed and integrated into the maintenance system. The improvement of OEE and other KPIs will be assessed in comparison to the previous state.
4. Model Updates based on Data Feedback
Following the implementation of predictive maintenance, the production process keeps generating new data. It is advisable to regularly re-train models using new data and refresh their performance. In this way, we elevate the maintenance planning model to the status of best in class.