By Philipp H. F. Wallner, industry manager, Industrial Automation & Machinery, MathWorks
Predictive maintenance offers the ability to anticipate equipment malfunctions before they occur, arrange repairs proactively, avoid stalling factory floor operations and most importantly, prevent the failure of machinery to keep businesses running efficiently. Figures from Deloitte show predictive maintenance can reduce the costs of maintenance by 5 to 10 per cent overall.
It is clear predictive maintenance has huge value in the engineering and manufacturing industries, but this is no real surprise. In the era of Industry 4.0, real-time data analytics, automation, machine learning and other technologies are enabling teams to speed up operations and reduce costs.
However, very few businesses have actually implemented predictive maintenance so far, as the process of doing so with existing equipment isn’t without its difficulties. There are four major challenges engineers must overcome in order to work with data scientists and realise predictive maintenance capabilities amid Industry 4.0.
Challenge one: Cultivating a collaborative environment
In order to make the most of the benefits of predictive maintenance, it is necessary to create a collaborative environment in which domain experts in engineering and data scientists work together. If predictive maintenance is approached with a singular data analytics mindset, not all of the insights from the engineering teams that built the equipment and maintain it on an ongoing basis will be captured, and vice versa.
Powerful algorithms based on statistics methods that integrate the expertise and domain knowledge of engineers as well as data scientists are needed to ensure the key elements of each effective application are fully leveraged.
In a competitive marketplace, engineers are under pressure to perform, but they are also resilient problem-solvers with a great deal of ingenuity by nature. With the right approach, it is possible for engineers to work together with data scientists effectively and realise the best predictive maintenance applications they can – ones that include both statistics-based data analytics methods such as machine learning in addition to engineering domain expertise.
Challenge two: Training algorithms with not enough failure data
An important challenge for engineers implementing predictive maintenance to solve is how to train algorithms properly with failure data. Often engineering teams are easily able to include ‘success’ data from everyday production, but a fundamental part of training an algorithm with machine learning is teaching the AI about the numerous error situations that could occur while the equipment is in operation. If the aim is to avoid it malfunctioning in the first place though, how can teams obtain failure data to train the algorithms?
The answer lies in simulation models, which can be used to produce artificial failure data. This data is irrespective of use cases and can range from wind turbines to air compressors. Using simulation to create failure data is a more efficient way to train AI than relying on the results of the factory floor which may not provide enough, or any insight into failed mechanics at all.
Challenge three: Implementing algorithms in the real-world
Once the algorithms have been fully trained on the desktop, the next challenge is deploying them into the industrial system equipment. How easy or difficult this task is depends on the condition of the existing IT and OT infrastructure. Some algorithms are applied onto real-time hardware platforms such as industrial PCs, embedded controllers or PLCs, while others are in the cloud or merged with current non-real-time infrastructure, for example an edge device running on Linux.
More and more, organisations are using toolchains to implement predictive maintenance in the real-world efficiently. These toolchains facilitate automatic generation of code, components or standalone executables. For example, international packaging and paper product manufacturer Mondi installed predictive maintenance software into its manufacturing line to reduce waste and machine downtime in its plastic web production.
Challenge four: Creating a business case for predictive maintenance based on data evidence
All the aforementioned challenges have available solutions, leaving one key problem – how to build a business case for predictive maintenance in the first place. Senior management will need to understand the return on investment that would be achieved before approving it, so detailing a comprehensive, data-driven plan is imperative.
To do this, engineers must develop an approach for how they will monetise predictive maintenance and calculate estimates on savings, such as on the reduction in equipment failure during operation.
There are several other good suggestions proposed by some of our clients for creating a business case for engineers to consider, such as linking service fees to predictive maintenance of the equipment used by the operators (equipment builders’ customers).
Another idea is taking advantage of intellectual property protection to sell the deployed predictive maintenance algorithms themselves. An additional area for consideration is moving to a new business model based on system usage, for example, selling cubic metres of compressed air rather than compressors, or lift usage hours rather than whole lift systems.
Predictive maintenance is a vital part of engineering and manufacturing in Industry 4.0. By combining data science with engineering domain expertise, using simulation to create failure data, toolchains to deploy algorithms and a variety of techniques to build a solid business case, more engineers can implement this vital technology and start realising its value.
From reducing equipment downtime, to generating significant cost-savings, to boosting efficiency throughout the production line, the benefits of investing in predictive maintenance are too great to ignore.