“If we can save a single major component, the predictive maintenance solution has paid off the licence fees for its lifetime.”
This quote is from a customer to whom our team was developing a software solution for predictive maintenance. It reveals the size of the business potential around preventive service actions.
This blog post is an intro to predictive maintenance from business and management perspective in industrial environment. We will go through an example business case.
Here is the more technical sequel: predictive maintenance tutorial for data scientists and developers.
Predictive maintenance is a decision making framework
First, let’s extract predictive maintenance into pieces.
Maintenance most often refers to a physical machine that needs service. It could be a car, an escalator or quite often a production unit in a factory or plant. Same principles could be applied also to non-industrial domains such as human disease prediction and non-physical events like customer churn prevention in an online shop.
Predictive means that the target events can be detected before the costly failure will occur. The prediction part implies that the failure detection happens either fully automatically or by human aided by meaningful analytics. Either case requires a software solution.
“In general, predictive maintenance could be defined as tools and processes to detect and prevent failures in advance in order to improve business unit’s performance in terms of finance, safety and legal regulations.”
Even though the need for the maintenance can be predicted and prioritized by an algorithm, most often the maintenance itself will be done manually. Predictive maintenance framework answers to the question “what should be done“ instead of “how it should be done”.
Building a business case around predictive maintenance
Safety, legal requirements, reducing monotonous work, spare parts inventory planning, increasing productivity. Being able to monitor real time and systematically the equipment fleet. Those are some benefits that predictive maintenance provides.
One thing is certain: Money is always involved when kicking off these initiatives.
Cost savings is the most straightforward measure for predictive maintenance benefits. If the equation shows profit, you have a strong signal to take actions.
A simplified predictive maintenance cost saving formula looks like this:
[Number of failures that can be predicted] x [Saving per correct prediction] - [Cost of the solution]
Number of failures that can be predicted. This comes from the hit rate of the predictive maintenance algorithms. Machine learning and data science play an important role. It is also about what and how reliable data can be collected. If the cost of incorrect prediction is costly, that can be added to the equation as well.
Saving per correct prediction. Purchasing a new component may be expensive. Depending on the domain, the maintenance work can accumulate costs. In industrial environment, it is usually the downtime and lost production that becomes by far the most expensive part. It is noteworthy, that either multiple predictions with low value or few predictions with high value can provide the same outcome.
Cost of the solution. Predictive maintenance solutions can be custom developed in-house or purchased as a software product if the problem can be generalized.
As stated in the first quote, even a moderate success can be enough to make predictive maintenance worth the effort. As everything is software by nature, it is possible to create a real time dashboards to monitor achieved value.
As a business manager you can run the numbers for your own product or service.
Organizing predictive maintenance operations
Even the best machine learning algorithms are not enough to solve critical business problems alone. All pieces of the chain need to be involved to create continuous practices around predictive maintenance.
In industrial supply chain equipment manufacturing and operation are often done by different companies. For example truck drivers do not produce the vehicles they operate. There are a few common setups how (predictive) maintenance is arranged:
- Manufacturer provides a predictive maintenance software to operators for an additional fee
- Manufacturer takes care all of the maintenance operations on behalf of operator
- Big operators can afford developing their own predictive maintenancesolutions
My personal experience is mostly from the first case as a data scientist. Creating analysis ideas to sell as well as developing the algorithms to detect failures. Based on what I have seen, everything is more or less communication between the stakeholders.
Predictive maintenance as a systematic process
From the very first days in predictive maintenance I felt it difficult to connect the raw data on my computer screen to event in the real world. Not that it would be difficult to plot the data time-wise, but data was often contradictory.
Maintenance records might show that a vehicle has been under engine replacement but the distance sensor shows it has been driving all the time. What is the truth, and how can you make analysis from that?
It is essential to keep ongoing discussion between the maintenance crew, machine operators, data engineers and data scientists. Picking specific cases and examining them deeper is crucial. Only so it is possible to combine the best of both worlds: Perceptions from the site and the power of data analysis.
The debate between the stakeholders makes it possible to develop better processes. Maybe the maintenance crew can log the information more accurately. Data scientist can ignore the discrepancies in data for specific cases and discuss about the options with the data engineers.
The discussions are learning opportunities for everyone. Operators and their managers always do not understand how to read certain information from the predictive maintenance applications. Once explained, the operational side have much better chances to redeem the benefits of the predictive maintenance solution.
Future steps in predictive maintenance
24/7 monitoring services, product X as a service and predictive maintenance applications are still in the early stages for most companies. The awakening to the data driven solutions has happened only during the recent years. Nevertheless, some future trends can be seen already.
Once robotics and intelligent machines such as autonomous vehicles get more foothold in the market, we will see more automation also on the maintenance side. After a predictive maintenance algorithm has detected that a car needs to replace its battery, the self driving car could drive to the workshop. There a robot can independently do the replacement.
Instead of making decision on an individual unit, the maintenances of the whole fleet could be prioritized more efficiently. The same way that we can optimize ship logistics on harbours, it would be possible to automatically schedule the maintenances in the businesses where the time slots are narrow due to high service frequency.
Predictive maintenance in industrial environment rely heavily on sensor data gathered from the equipment. In smaller closed area data can be transferred through WLAN, but many application require wireless network like 5G. The factory parking video below demonstrates 5G network in the context of autonomous car parking on factory site. The same network could easily handle the requirements of predictive maintenance data transfers.
Mikael Ahonen
Data Scientist at Unikie
