2023 05 12 Predictive Maintenance
2023 05 12 Predictive Maintenance
2023 05 12 Predictive Maintenance
Imagine one day you receive a phone call saying, “Hi, this is your washing
machine company. Can we send a tech to your home today to replace the pump
on your machine?”
As you stare at the basket of folded clothes next to the couch, you respond
“Thanks, but the machine is fine. I used it today.”
The company counters with, “Yes, we know, but our readings suggest the pump is
on its way out. At this point, it typically fails in one of the next few uses. This could
cause flooding in your home and will damage other internal components, which
means a much bigger repair or even a new machine. Plus, there are clean-up
costs and other inconveniences while the machine is out of order.”
Naturally, with this in mind you schedule a time to service the machine.
This same process of identifying problems before they’re problems is what industrial operators
are trying to do today for their critical equipment. Instead of waiting to fix machinery when it
breaks or hoping a machine makes it to a scheduled maintenance point, companies want to
predict when things will fail. This is very different from ‘preventive maintenance’ where
companies replace some or all parts, regardless of wear, which is very expensive.
Predictive maintenance harnesses historical data to identify patterns and signs that indicate when
equipment failed in the past. Companies then use those models to create alerts so they can
monitor equipment for the same signs and act when they occur – before the equipment fails and
they experience downtime.
To be successful at predictive maintenance, companies need both historical data and real-time
data. Without real-time data from your equipment’s sensors, you can’t react to issues in a timely
manner. When considering equipment like power stations, wind turbines, infrastructure, industrial
machinery, aircraft engines, and others, reacting in real-time is mission critical.
2 / 12
The objective of predictive maintenance is to reduce the amount of unplanned downtime and
increase the lifespan of the equipment being monitored. This results in more efficient operations,
higher productivity, and improved business outcomes for the organizations that utilize predictive
maintenance.
The predictive maintenance approach differs from preventative or reactive maintenance. The
former involves regularly scheduled inspections, repairs, and replacements designed to keep
equipment running at optimal levels. However, preventative maintenance isn’t as efficient
because it occurs around set schedules and not in response to precise data for each machine.
Reactive maintenance is the general default approach. Here, maintenance only happens when a
problem or malfunction occurs and requires immediate attention, typically on an emergency
basis, which also makes it more expensive.
Different analysis tools exist to serve various predictive goals, like artificial intelligence and
machine learning models or even traditional statistical methods for anomaly detection. Once you
collect enough historical data to make reliable predictions, you can use real-time alerting to know
when to shut down equipment or schedule targeted repairs before issues occur.
3 / 12
Predictive maintenance benefits
From the perspective of industrial operators, using predictive maintenance has several key
benefits.
Worker safety
In workplaces where machine failure could cause injury to workers, having the ability to
accurately predict equipment failures helps keep workers safe.
Cost savings
Predictive maintenance saves you money in several different ways. There are cost savings
associated with having machines up and running in an optimal state. This leads to increased
output, enables process optimization (which means even more, reliable output), and reduces the
costs associated with replacing expensive industrial equipment. Predictive maintenance saves on
repair costs because you can do more maintenance with a smaller workforce, especially for
distributed operations, and you don’t end up replacing parts unnecessarily.
4 / 12
Predictive maintenance techniques
The following list isn’t exhaustive, but it provides a solid starting point for understanding what’s
possible.
Condition-based monitoring
One of the most common techniques is condition-based monitoring. Condition-based data
provides information about the current condition or performance of the equipment being
monitored. This approach detects wear, fatigue, and damage in industrial equipment by tracking
factors like temperature, pressure, and others.
Acoustic analysis
Acoustic monitoring uses equipment to record sounds emitted by equipment at the sonic and
ultrasonic level. This includes things like listening for gas leaks and issues with engines.
Ultrasonic monitoring is particularly useful because it can detect issues beyond the range of
human hearing to identify anomalies in equipment performance.
Vibration analysis
Vibration analysis is useful for monitoring high-speed rotating equipment. This technique uses a
baseline reading when the equipment is in ideal condition to determine the normal vibrational
rhythm. If the machinery deviates from the baseline rhythm, it may be a sign that repairs or
adjustments are necessary.
In many cases, historical data helps to pinpoint problems based on vibration changes. For
example, vibration analysis can tell the difference between loose parts and more significant
motor issues. This makes repairs faster because the technician knows exactly what they need to
fix instead of spending time doing exploratory analysis of every equipment component.
Infrared analysis
You can use infrared cameras to find abnormal temperature readings that might indicate a
problem with equipment. This process can be done manually by employees or automated using
new AI-powered image recognition systems.
5 / 12
Fluid analysis
Fluid analysis involves tracking the physical and chemical aspects of fluids used in machinery, like
coolants or lubricants. Tracking temperature, lifespan, contaminants, and other factors allows you
to optimize fluid replacement and prevent waste.
Visual inspection
Visual inspection involves looking at equipment for signs of damage like cracks or corrosion. You
can automate this process with modern AI image recognition technologies, to increase the
frequency of inspections more easily.
In other words, the context of time is critical to predictive maintenance. Sensor data is, by nature,
time series data, which tracks changes of time from a particular source over a given window of
time.
Sensors used in industrial equipment generate a lot of data. This is a good thing because it’s
exactly what these advanced analytics models need. Using a time series database helps you
manage that data and integrate it with ML and AI technologies.
Collecting all the data needed to train and feed predictive models is possible with Telegraf. This
is an open source data collection agent. It has a library of 300+ plugins to connect to any data
source and, if a plugin doesn’t exist for a source, you can easily write a custom one. You can even
use Telegraf to clean your data before writing it to InfluxDB, saving time and resources for your
most critical operations. Telegraf is extremely lightweight and can sit on edge devices and
gateways to collect data directly from its source.
6 / 12
While Telegraf makes data collection easier, InfluxDB has everything you need to store, manage,
and transform your time series data. InfluxDB is a purpose-built time series database, designed to
handle the volume and velocity of sensor data. Built on open source technologies, InfluxDB
integrates with any system or software, including the ML and AI tools used in predictive
maintenance processes. InfluxDB’s data compression ratio means that industrial operators can
store high fidelity time series data forever at minimal cost. This historical data provides the
backbone for predictive models.
However, InfluxDB can also query data in real-time. This means that you can run queries on
real-time data against predictive models and trigger alerts based on the output. Having real-time
and historical data in a single solution, and equally accessible, streamlines the entire process and
gives you greater flexibility when training and implementing predictive models.
InfluxDB in action
ADLINK
ADLINK is a company that designs and manufactures products for embedded computing, test
and measurement, and automation applications. Its products include everything from
computer-on-modules, to industrial motherboards, to data acquisition modules, and more.
ADLINK uses InfluxDB as a part of their ADLINK Edge™ IoT solution to store IoT data. The platform
connects IoT data to advanced analytics systems to visualize it for monitoring and analysis and to
extract actionable insights for operational efficiency.
One particular customer, a defense contractor, wanted to shift from reactive to predictive
maintenance on its Thermotron test chambers and turned to ADLINK Edge™ to connect existing
software, equipment, and systems securely, seamlessly, and cost effectively. These machines
perform testing at extreme temperatures. The problem is that when malfunctions occur, they take
a long time to identify because the machines can take several hours to start up. The cost to repair
the machines can exceed $100,000, plus the loss of efficiency due to machines being offline and
interrupting scheduled testing and production.
Using ADLINK Edge™ allows the company to collect data from external sensors that provide
information about what resources go into the machines, and to connect directly to PLCs to
understand what’s happening inside the machines. Visualizing this data helped the company to
understand the factors that caused machine failures. Armed with these models, the company
uses real-time data from the sensors and PLCs to predict machine failure and take proactive
steps to repair and maintain them. Overall, the company reduced downtime related to planned
and unplanned maintenance shutdowns, improved machine performance, and reduced costs.
7 / 12
AnalyticsPlus
AnalyticsPlus is a leading Chicago-based advanced analytics and predictive modeling company
and InfluxData partner. The company has strong domain knowledge in multiple industries. In the
manufacturing sector, it creates unique, IP-driven streaming analytics, anomaly detection, and
edge computing solutions.
The AnalyticsPlus team worked with Nebbiolo, a company that specializes in edge computing, to
build a platform that uses InfluxDB to generate real-time insights. One example of this platform in
action is performing quality control of engine heads produced at a factory in Alabama. Workers at
the factory performed manual inspections on engine heads at random. This was a labor-intensive
process and only accounted for about 1% of production.
The plant wanted to automate the inspection process to increase the inspection rate and more
accurately identify problematic parts. So, the factory instrumented its shop floor, installing sensors
across five manufacturing lines and twenty-five CNC machines to inspect various aspects of the
engine heads. The data from these sensors feed various predictive models to determine whether
a part is likely to fail. After implementing this predictive modeling approach, the manufacturer
increased its inspection rate of engine parts from 1% manually to 100% automated, in-line testing
and was able to predict failure across all production lines in real-time.
Industrial machinery
In industrial settings, predictive maintenance models monitor the performance of equipment such
as pumps, motors, and generators. This helps reduce the amount of unplanned downtime and
keep operations running smoothly. Predictive maintenance in this setting is very useful for
increasing overall equipment effectiveness (OEE). Predictive maintenance practitioners can see
OEE surpass 90%.
8 / 12
Power generation
Monitoring power generation equipment, like turbines and generators, helps operators
understand when machines need repair or are in danger of failing. This data helps electrical
companies understand and prioritize what infrastructure investments are necessary and provide
customers with more reliable service.
Data quality
For predictive maintenance to work, above all else you need accurate data. That means
calibrating sensors properly, ensuring that you’re not missing data, and making sure your data
represents all hardware being monitored.
9 / 12
Company culture
Implementing predictive maintenance can require major changes to existing processes and
workflows. For predictive maintenance to be effective, leaders need to make sure their workers
buy into the new system.
False positives
False positives from inaccurate predictive maintenance algorithms can result in wasted
maintenance costs, and cause workers to lose confidence in the system.
Conclusion
Technological advancements make predictive maintenance easier to implement and manage
than ever before. For industrial operations, predictive maintenance allows organizations to
extend the life of expensive equipment, optimize equipment efficiency, and reduce maintenance
time and costs, all of which combine to provide a competitive advantage.
The advanced analytics tools needed to create predictive models require large amounts of time
series data, and InfluxDB is the ideal solution for collecting, managing, and storing that data in a
single datastore. InfluxDB also provides interoperability with virtually any system, ensuring that
10 / 12
you can collect and access the data you need, even as you add new equipment and scale out
your operations.
11 / 12
12 / 12