AI-Driven Predictive maintenance is when data is used to predict when and where an issue may occur in the system. It is meant to minimize the risk of failure and find existing relationships in data. Previously, we relied on preventive maintenance. Preventive maintenance is performed regularly on a system to prevent unexpected issues.
The shift from preventive maintenance to predictive maintenance is a product of the rise of automation, data, and IoT--also known as Industry 4.0. According to the International Society of Automation, $647 billion is lost to maintenance every year. Hence, it is in the best interest of businesses worldwide to decrease downtime and improve effectiveness.
Industry 4.0 is also known as the fourth industrial revolution. It refers to the rise of automation technology in manufacturing. The first industrial revolution saw mechanization through water and steam power, the second industrial revolution saw mass production and assembly lines using electricity, and the third industrial revolution saw the emergence of nuclear energy, electronics, telecommunications, and computers.
The whole premise of the fourth industrial revolution revolves around smart machines that get even smarter with the advent of big data. Factories and maintenance systems will become more efficient, productive, and less wasteful.
One concrete example is that of an African gold mine that was able to identify a problem with oxygen levels during leaching which led to an increase in yield by 3.7%. The AI-driven predictive maintenance system effectively saved them $20 million annually.
Predictive maintenance will also help usher in Industry 4.0. This transformation enables companies to produce higher-quality goods at lower costs. This shift stresses the importance of prevention then optimization.
According to McKinsey & Company, predictive maintenance can save up to 40% of maintenance costs over the long term and reduce capital expenses on new machinery and equipment by up to 5%.
Preventive maintenance is largely reactive. Repairs are driven by time and events. Because maintenance is scheduled regardless of whether it is working or not--this method runs the risk of over-maintenance and under-maintenance.
Preventive maintenance relies on known best practices to guide maintenance procedures. This is done regardless of whether or not an issue is actually present. The whole point of the ordeal is to keep the system in top shape according to a tried and tested maintenance guide.
Preventive maintenance is ideal for systems that have a critical operational function, have failure modes that can be prevented with regular maintenance, and have a likelihood of failure that increases with time or use.
On the other hand, it is not suitable when a system has random failures that are unrelated to maintenance and do not serve a critical function.
Here are key characteristics of preventive maintenance:
This refers to maintenance that is performed over a fixed period of time. For example, equipment inspection performed regularly to assess if it is still fit to operate or not.
This refers to maintenance that is done after a certain metric has been fulfilled. For example, a vehicle would be scheduled for service every 10,000 kilometers of travel since its last check-up.
While preventive maintenance is primarily reactive, AI-driven predictive maintenance is proactive. Preventive maintenance ignores contextual information whereas AI-driven predictive maintenance thrives on it.
It uses data from various sources such as historical maintenance records, sensor data, weather data, and others to figure out when a machine needs maintenance and if any problem areas that need attention arise.
PwC states that AI-driven predictive maintenance reduces costs by 12%, improves uptime by 9%, reduces safety, environmental, and quality risks by 14%, and extends the lifetime of an aging asset by 20%.
These are some examples of data types that can be harvested for predictive maintenance:
This mostly refers to monitoring device temperatures and ambient air temperature
This can apply to check the function of heating, air vents, and air-conditioning systems.
Water pressure can be monitored to look out for unexpected fluctuations
Vibrations can be monitored to check for faults before the piece of machinery breaks down
As Machine Learning is applied to the harvested data, the following effects can be observed:
A logistics company wants to identify problem areas in their trucks by collecting and analyzing data. The trucks are outfitted with IoT hardware that monitors the trucks’ health. From there, AI methods can be applied to the data to find patterns that point to problems very early on.
Their product marketing manager state that “the latest AI capabilities have enabled VTNA to uncover hidden insights in our collected data and merge it with our engineering knowledge of our trucks”
There is a public CAN (controller area network) and an OEM private CAN. The public CAN is fed information from the vehicle’s electronic control units whereas the OEM private CAN contains more detailed information. These monitoring systems can be installed on a company’s trucks to monitor their health.
The logistics company found through observation that a lot of the faults in the trucks actually start in the after-treatment in the exhaust side of the vehicle. From this, they have identified the exhaust system to be the likely origin of truck faults.
The nitrogen oxide sensors pick up different O2 levels, moisture readings, and heat readings during after-treatment. The variances in the data signal problems early on and allow the proactive reparation of the problem area.
AI is able to sniff out meaningful insights based on the data which allows the company to use its engineering knowledge to identify the root cause of the problem and its corresponding solution.
The usage of AI-driven predictive maintenance has led to a 20% reduction in roadside breakdowns based on the monitored systems through the predictive maintenance application. There is also an 8%-9% improvement in technical efficiency and 2%-3% improvement in fuel efficiency.
A software supplier sells software that enables effective production process management. In particular, it had to minimize the risk of server failure by detecting anomalies in the server functionality. They wanted to create an algorithm that achieves this goal.
The data first had to be normalized to prevent skewed data from interfering with the results. To do this, wrong values such as negative electric power had to be removed because faulty data are usually products of measurement errors and hardware failures.
Then, missing values were interpolated based on the time-series data. Finally, the algorithm was trained over continuous sequences of over 24 hours. The quality of the algorithm was measured via RMSE (Root Mean Square Error).
Since the data is a time-series, there were 3 different models: Moving Averages (MA), Autoregressive Integrated Moving Average (ARIMA), and Long Short Term Memory (LSTM). It was found that each model worked well with certain signals. As such, it was decided to implement the module as a microservice with separate models for different signals.
This module was used by the client for anomaly detection and ultimately led to fewer server downtimes and minimized server failure.
Here at Codvo.ai, we believe that AI-Driven Predictive Maintenance represents a future where automation reduces business costs and operational failures. Outfitting your systems with AI-Driven Predictive Maintenance can reduce costs by 12%, improve uptime by 9%, reduce risk by 14%, and extend the lifetime of an aging asset by 20%. In addition, getting a stake in Industry 4.0 early on will allow you to be a step ahead of your competitors.
If you are interested in AI-Driven Predictive Maintenance, contact us today at firstname.lastname@example.org!
Read this article to know why the construction industry needs e-commerce:
Do you think your company’s data warehouse is efficient, scalable, and consistent? Read this article to learn more!