Equipment maintenance is essential for avoiding unplanned downtime and preserving quality standards in multiple industries such as power generation, oil & gas and manufacturing. In the past and in most industrial facilities today, preventive scheduled maintenance processes are in place, using a schedule of maintenance tasks to ensure equipment issues are addressed before problems occur.
In a perfect world, the preventive maintenance schedule lines up perfectly with the actual wear and tear on the machine. For example, when a millwright opens up a machine to change a seal, they find the seal is at the edge of failure. Of course, preventive maintenance sacrifices some of this perfect timing for reliability. Changing a seal, belt or lube that is still in good condition is an acceptable cost compared to unplanned downtime caused by an equipment failure.
However, the concept of predictive maintenance takes the idea of right-on-time preventive maintenance and optimizes it. With digital systems that use sensors to measure the machine’s condition and software to track and analyze this data, it is possible to book maintenance when and where it’s needed, without a fixed schedule. In short, predictive maintenance systems act as the “check engine light” for your equipment, whether that is large rotating equipment, an industrial robot or a CNC mill.
While both approaches to maintenance avoid unplanned downtime by choosing to take machines offline at scheduled times, predictive maintenance saves costs by allowing maintenance to be scheduled less often, and only when needed.
What Is Condition Monitoring?
Condition monitoring enables predictive maintenance. In common use, the terms are used interchangeably, but in short, predictive maintenance is the overarching maintenance strategy, and condition monitoring is the means to achieve it.
Condition monitoring uses machine sensor data to measure parameters that can indicate machine health, and analyzes that data using software. Vibration and temperature sensors are most common, but many parameters can be tracked, including motor run time and even audio capture in some cases. This data is captured and analyzed using machine monitoring software running on a PLC or controller, or in the cloud.
Condition monitoring is typically considered an industrial IoT solution, because vendors such as Allied Electronics specialize in on-premises solutions on the machine, and well-known cloud solutions include Siemens Mindsphere and IBM Maximo. With either approach, condition monitoring software raises an alert when machine health changes, and some systems automatically put in a work order for technicians. Alerts can be as simple as the traditional red light, or as advanced as smartphone notifications.
Machine Learning and Predictive Maintenance
Machine learning algorithms play a key role in the function of predictive maintenance systems. It is possible to set up predictive maintenance using manually set thresholds for sensor data, such as triggering maintenance on a pump when vibration reaches a certain threshold. However, with a learning algorithm, predictive maintenance systems can learn from the massive data pool collected by equipment sensors and begin to make advanced predictions further in advance and with greater accuracy.
One fascinating example of predictive maintenance using artificial intelligence is IBM Acoustic Insights. This system uses audio recordings of running equipment to assess machine health. It is designed to be used for quality inspection of finished products, such as white goods. In a nutshell, Acoustic Insights works like telling your mechanic that your car is making a “squeak, then a thud” and the mechanic uses their knowledge and experience with cars to make an accurate guess at the problem. With the IBM system, the AI is trained using a training data set of captured sounds of both go and no-go products. Each sound is manually annotated by subject matter experts. With this training set ingested, the AI is able to begin assessing new sounds, and is provided continuous feedback on its accuracy. This allows the system to continuously learn and reach high levels of accuracy.
One real-world application of this technology is in robotic spot welding of automotive chassis. A bad weld makes a different sound than a good weld, and by raising an alert when it detects this sound, workers can immediately address the welder issue, decreasing the number of poor welds.
IoT or Edge for Condition Monitoring?
While some pieces of equipment have built-in sensors that can be leveraged for predictive maintenance, additional sensors can also be applied to measure data. The question is, how should this data be collected?
The answer today is via network connectivity. IoT-enabled sensors can pass data via a local network to a central controller, or pass data directly to the cloud. There are two prevailing strategies for where data should be processed: in the cloud and at the edge.
Cloud Processing of Machine Health Data
Sending raw, high-volume data directly from the machines and sensors to the cloud has a few advantages. First, it avoids the cost and deployment of edge computing devices. Like all cloud computing applications, the cloud provides turnkey access to processing power and storage. IT does not need to procure, maintain or expand these capabilities to support a predictive maintenance strategy on an ongoing basis. In addition, all the data is gathered and stored in the cloud, so it can be mined for analytics in future if needed.
Disadvantages of cloud processing center around the IT infrastructure needs. Sending high-volume data from many pieces of equipment requires a robust network and internet connection, and an organization’s IT and Information Security policies may have issues with sending vast quantities of internal data to a third party. Data residency, cybersecurity and connectivity can be concerns. In addition, the cloud approach requires a reliable internet connection 24/7, or at least as long as the equipment is running.
Edge Processing of Machine Health Data
Edge processing refers to the approach in which data is processed locally, and the result is output to a central cloud-based dashboard. The key advantage of edge processing is flexibility. It does not require as much IT infrastructure, because the raw, high-volume data does not upload to the internet. In addition, this reduced reliance on internet connectivity reduces all the associated concerns around connectivity and security.
When starting small, edge devices can be more cost-effective than a cloud solution because the software cost of an entire cloud solution such as Siemens Mindsphere or PTC Thingworx is going to be roughly the same no matter how many machines are connected to it at the start. With edge devices, it is possible to start slow and build up.
Extending Equipment Life with Predictive Maintenance
No matter how it is deployed, a predictive maintenance strategy can extend the life of equipment. In manufacturing, issues of equipment reliability and performance may raise questions of whether it’s best to run to failure, or continue to maintain. Bringing in new machines creates a learning curve for operators, comes at a high cost and may disrupt production, in addition to the high up-front cost of the new equipment. Condition monitoring allows manufacturers to keep a close eye on machine health and address issues before they impact production.
With an effective predictive maintenance strategy in place, industrial equipment can benefit from longer life, greater efficiency and greatly reduced unplanned downtime. There can be significant costs to deploy condition monitoring solutions when no infrastructure or hardware already exists, but edge devices can enable a smaller start which can be scaled to meet future goals.
With condition monitoring linked to a high-level AI analytics and reporting solution, predictive maintenance can be realized—and meaningful cost reduction too.