According to a study by McKinsey, predictive maintenance will save businesses $ 630 billion by 2025. These savings will be made possible by several factors—first, a reduction in maintenance costs of 10 to 40%. Then reduce the number of breakdowns by half. Finally, by decreasing the amount invested in new machines by 3 to 5% by increasing the life of existing machines.
It is, therefore, a bright future promised by the famous consulting firm to the world of industry. But if predictive maintenance breaks from what is done today, it is first necessary to study the most widespread maintenance strategies today.
Corrective maintenance, preventive maintenance, two diametrically opposed models
Corrective maintenance is the most basic approach to maintenance. It very roughly repairs – or replace – a room once the failure noted. This type of maintenance has the advantage of saving all preventive maintenance activities. Corrective maintenance is legitimate, even recommended, in some cases:
- replacement parts are low cost
- the parts can be changed quickly
- the impact of machine failure is low for the end user
On the other hand, preventive maintenance is proactive. It is a question of anticipating the defects and the breakdowns of a machine. Preventive maintenance is in the form of an intervention schedule. The supplier plans frequent visits to the site to ensure that the delivered machines are not damaged. The problem lies in the cost of the vendor’s coming for machines on which there is no problem. So there is a waste.
What is predictive maintenance?
Predictive maintenance can detect anomalies on machines before they become too serious. The strength of predictive maintenance is, therefore, to anticipate breakdowns. This avoids any expensive shutdown of the production line. If the predictive maintenance emerges is that it is now possible to capture the weak signals on the machines. It then remains to trace the data and analyze. These analyses help increase customer satisfaction and save money.
Types of Maintenance
Predictive maintenance, serving customer satisfaction
Predictive maintenance, compared to preventive maintenance, makes it possible to go from a logic of pushed flow to a logic of pulsed flow. The supplier intervenes only when signals emitted by a machine reflect a probable failure in the short term. It is, therefore, the actual state of the asset, and not an academic calendar, that triggers an intervention. The maintenance needs what’s called CMMS.
The anticipation of breakdowns is made possible by:
- The implantation of sensors. They can go back thousands of data each day. It’s the Internet of Things (IoT).
- The modeling of a fault diagram. Based on the operating history of the machines, beyond the symptoms, it is possible to identify the root causes of the failure.
- The development and optimization of predictive algorithms that determine alert thresholds. It’s machine learning or machine learning. Technologies like Apache Mahout or SparkMLlib are appropriate for this case.
That’s why Schindler decided to test this approach by installing sensors in 50 lifts in 2016 in Germany. The data reported concerns the temperature of the elevator, the door openings, or the number of floors delivered. As part of a project with Sicara, a startup specializing in AI, the company predicted of time series on these temperature data. Then they identified when the number of door openings would be reached before failure. So, they could predict when the technician would have to intervene on the machine. By multiplying the sensors of this type the objective is to have a lift that works 100% of the time (finally)!
