From preventive to predictive maintenance
When we think of high-volume manufacturing, we often imagine a smoothly running factory with every machine functioning perfectly to turn out finished products efficiently and in volume. For the people running such a factory, their objective is to keep the factory running at its optimal speed with minimal downtime.
This objective recognises every machine with moving parts suffers some wear and tear and will inevitably need to be serviced or have some parts replaced. The question is, when is the best time to do this? Do you do it according to a fixed schedule or do you wait for the machine to start showing signs of failure?
Approaches to maintenance
A first approach is to schedule maintenance tasks based on a fixed predetermined schedule, which ignores the actual condition of the equipment. Think of a regular car checkup at fixed intervals or mileage. This approach has the advantages of being simple to plan but also has significant drawbacks in that maintenance may happen too late, resulting in equipment damage and danger for workers, or it may be carried out when it isn’t necessary.
A smarter approach is condition-based maintenance. This approach drives maintenance actions based on the estimated condition of the machine that is typically monitored through inspection or using data from embedded sensors. This has the benefits that maintenance will happen before failure and only takes place when necessary, but the drawback is that maintenance only begins after the machine begins to show signs of failure and the necessary maintenance intervention may not be optimal for production scheduling.
A third approach is predictive maintenance. Here the aim is to predict, at the earliest point in time possible, the maintenance actions that will be required at some point in the future. It is an approach based on condition monitoring combined with a dynamic predictive model for failure modes. This has the advantages of optimising maintenance to both the machine’s life and the factory’s production efficiency, although it does require a more complex overall system.
The primary promise of predictive maintenance is that it enables corrective maintenance to be scheduled at a convenient time while maximising the equipment’s useful life by preventing equipment failures. With the knowledge of when machines need to be serviced and what needs to be done, maintenance work can be planned optimally with the right people and parts ready. This approach aims to eliminate unplanned line stops and reduce stoppage time overall. In addition to increasing the factory uptime, there are other advantages including a reduction in accidents associated with equipment failure and increased equipment lifetimes.
Building a predictive-maintenance system
To build a predictive-maintenance system, a number of elements are needed.
First, automated condition monitoring must be installed on the target machines. This monitoring can involve visual inspection with cameras, measurement of vibration with accelerometers, measurement of noise levels or ultrasonic sound with microphones, and heat or humidity measurements, for example.
Next, some embedded processing is required to handle the first analysis of the raw data, turning it into useful information that can be shared with supervising systems. For example, the embedded software can continuously run a comparison of the vibration characteristics of the machine over time to determine when changes are occurring. By embedding the processing capabilities in the sensor unit, the amount of data that needs to be communicated is vastly reduced. This is particularly important for visual inspection where the amount of data can quickly become huge.
Next, the information must be communicated to local and remote supervising systems. This communication must be done securely and efficiently, taking into account the infrastructure of the plant to determine what kind of connectivity is best suited to the task. For example, an existing plant lacking wiring for sensors would best use wireless communication as a cost-effective and fast way to implement a connected sensor network.
Lastly, a predictive model for the equipment failure mode(s) must be created. Engineers can build this model on theoretical failure models combined with the data collected from actual field installations. When a large and reliable dataset is available correlating the sensor data and actual failure mechanisms, you can use machine-learning techniques to create a more refined predictive-maintenance model.
The conditions for the widespread adoption of predictive maintenance are now in place with the availability of all the key components, combined with cloud services and artificial intelligence.
ST offers a range of solutions for predictive maintenance, such as:
- motion, vibration and environmental sensors including products with a 10-year longevity guarantee
- MEMS microphones capable of ultrasonic noise detection
- microcontrollers with a range of processing power and embedded peripherals
- wireless connectivity solutions including Bluetooth and Sub-1 GHz
- wired connectivity including IO-Link and power-line communications.
Originally published here.
What data scientists and engineers need to know when working with big data as they move from...
Fujitsu Laboratories has developed a technology that offers both high-speed data processing and...
Researchers have used a customised, low-cost 3D printer to print electronics on a real hand for...