Predictive maintenance: how it works and what is at stake

In recent years, industry has been experiencing a new revolution, the 4th or 4.0, with the appearance of the Internet of Things (IoT) and artificial intelligence. After having addressed the issues of improving efficiency and productivity of industrial equipment, new technologies are now addressing the issue of their maintenance, which is called predictive maintenance.  

Definition of predictive maintenance

Unlike corrective maintenance (otherwise known as curative or reactive maintenance), which consists of having technicians intervene when a breakdown occurs, predictive maintenance will make it possible to anticipate the breakdown and therefore to intervene upstream and thus optimise human and financial resources.   

 Predictive maintenance consists of anticipating failures in equipment, machines or systems using sensors (IoT) which monitor the state of the machine in real time and send back data via the Internet of Things. This data is then analysed and used to develop algorithms that will predict and identify the signs of a breakdown so that technicians can intervene proactively, before the breakdown occurs. Parts are therefore not changed unnecessarily and the risks of production stoppages are reduced.

The benefits of predictive maintenance

Predictive maintenance is a real innovation for the industry, and if properly programmed, presents numerous benefits:

• Reduced maintenance costs (10-40% depending on a Mckinsey study)
• Reduction in the number of breakdowns: the number of breakdowns is on average halved (Mckinsey study)
• Reducing equipment downtime rates
• Reducing production stoppages (in the automotive sector the average hourly cost of machine stoppages is estimated at €1,127,000 and at €184,700 for the oil and gas sector in 2019 according to a Senseye study).  
• The extension of the life of equipment
• Improving the reliability of equipment and thus its productivity
• Better management of spare parts stocks

Predictive maintenance 4.0 cannot be reduced to simple maintenance, but aims to reduce the unavailability of equipment (by maintaining it preventively). It is therefore a step further than the productivity improvement mentioned before, but in the same direction: improving production at constant installation.

In the Covid 19 period, where travel had to be avoided as much as possible, predictive maintenance projects also had the advantage of not sending technicians to the site when they were not needed. 

Predictive maintenance operation 

The prerequisites for integrating a predictive maintenance system are the following: 

• Implementing precise sensors that will provide several thousand pieces of data every day in real time on the operating status of the machine (thanks to the Internet of Things: IoT).

To detect a failure or malfunction, sensors can be based on different parameters: vibration analysis, ultrasonic fault detection, infrared thermographic analysis, oil and fluid analysis, spectral analysis (frequency analysis) or visual analysis by cameras. 

• Modeling a failure pattern: based on the history of machine operation; this is the learning data that will subsequently feed the machine learning algorithms.
• Develop predictive algorithms based on machine learning algorithms and current data, which will determine alert thresholds.

‍

What are the challenges of predictive maintenance?

First of all, the deployment and integration of a predictive maintenance programme represents a significant cost for a company: the cost of the data sensors that will monitor the state of the machine in real time as well as the cost of the analytical tools dedicated to the exploitation of this data. In addition, the cost and training time required to master these tools for the maintenance service teams must also be taken into account. Finally, the development and deployment of a predictive maintenance programme also requires a time investment.  

Dedicated solutions are available to ensure that the data is properly retrieved and used, and that machine learning is implemented. 

IBM Maximo Predict, for example, which, when integrated with IBM Maximo, will be able to "look for patterns in asset, usage and environmental data, and correlate these patterns with known problems, to help reliability engineers and maintenance managers predict failures and share data and results".. 

Discover in our dedicated use case with IBM how we helped them develop their solution by training their algorithms with our image annotation solutions.