Industry 4.0 observers tend to analyze the impact of Industrial Analytics for Predictive Asset Maintenance on yield rates, asset downtime and O&M expenditures. One of the most overlooked financial benefits of Industrial Analytics is the potential for delaying and even reducing CapEx expenditures by extending the life of an industrial asset.
There are several different approaches to Machine Learning for Asset Maintenance. This article will explain which type of Machine Learning methodology is optimal for extending asset lifetime.
The Problem Today: Aging Machine Assets
Many industrial facilities are plagued by aging infrastructure. This problem is particularly acute in the oil and gas industry. According to a report by Accenture, more than half of offshore platforms are beyond their original design life.
In the US, the average asset age of industrial equipment is about 10 years, the same level as that recorded during World War II.Source: U.S. Bureau of Economic Analysis
When assets age, O&M costs rise. For instance, with the annual maintenance cost of a new wind turbine is 1.5% – 2% of the original turbine cost. However, as the wind turbine ages, maintenance cost rise to 3% of the original cost.
At a macro level, to upgrade aging industrial infrastructure will require a significant increase in CapEx. This article will explore whether Machine Learning for Asset Maintenance can extend the life of an industrial asset and delay the need for incremental CapEx.
Three Approaches to Industrial Analytics
Although there is almost a universal consensus about the potential economic opportunity from Industry 4.0, there is no single roadmap for implementation. Below is a synopsis of the three approaches to Industrial Analytics for Predictive Asset Maintenance.
1) Manual Statistical Modelling
The most rudimentary approach to Industrial Analytics for Predictive Asset Maintenanceis to create an in-house group that can generate predictive models.
Not all organizations have the expertise for Big Data or Machine Learning Centers of Excellence (CoE). The alternative is to rely on the “citizen” data scientist concept. The term was coined by Gartner to describe an individual that “generates models that use advanced diagnostic analytics or predictive and prescriptive capabilities, but whose primary job function is outside the field of statistics and analytics.” Citizen data scientists are armed with off-the-shelf analytical packages that are both powerful and intuitive enough for an intermediary user.
Can Manual Statistical Modelling extend the lifecycle of heavy machinery and industrial equipment? Probably not. Extending the life of an asset is not simple and is the outcome of a well-structured strategic asset management plan. Manual statistical modeling has several limitations. It is based on using sample data that is taken from machines and then data scientists build statistical models. The data scientists extrapolate from the sample data to the specific machines. Manual statistical models become obsolete relatively quickly as they are not based on dynamic Artificial Intelligence.
2) Digital Twin for Predictive Analytics
The Digital Twin concept has been gaining much attention in the Industry 4.0 arena.
A virtual clone of a machine asset is created using the blueprints of the physical machine. Data scientists, design engineers, and other subject matter experts are required to in the design and development stage of each customized Digital Twin.
The Digital Twin “learns” the underlying asset behavior using the Supervised Machine Learning methodology. (Please see our article on Supervised Machine Learning here.)
Can the Digital Twin extend the lifecycle of heavy machinery and industrial equipment? Perhaps in theory. The chart below depicts the Bathtub Curve which compares the failure rate of an asset versus time. During the Break-In or Infancy Period, the failure rate is relatively high because of Manufacturing Defects. However, once these are resolved, failure rate stabilizes during the Steady State period. Finally, at the end of the asset life is the Wear Out Period when there is a substantial increase in breakdowns due to Wear Out Failures.
If the Digital Twin is paired with a new asset or at the “Break-In” stage, one can assume that Manufacturing Defects and Wear Out Failures will be detected, thereby extending the life of the asset.
Can the Digital Twin be deployed at the “Wear Out” phase?
The answer here is more complex. First, to create the virtual clone, the solution provider will require access to accurate machine blueprints. During the life of an industrial asset, repairs and upgrades are made. Often the original blueprints are not updated accurately. Variations between the underlying asset and the machine blueprint will need to be resolved manually to create the Digital Twin.
Second, the Digital Twin is expensive to implement from both an infrastructure and internal resource perspective. The economics of creating a customized Digital Twin towards the end of the asset lifecycle is difficult to justify from an ROI perspective.
In summary, for equipment or machinery approaching end-of-life, the Digital Twin is not a practical or cost-effective solution for extending the asset life.
3) Unsupervised Machine Learning for Predictive Analytics
An alternative approach to Simulated Modelling is Predictive Industrial Analytics based on Unsupervised Machine Learning. What is Unsupervised Machine Learning? Vast amounts of machine sensor data are analyzed using a cloud-based platform. The algorithm is looking for anomalous sensor data. Once the Machine Learning algorithm detects anomalies in the data, correlations and pattern detections between signals are performed. This is done to later present the operators with the exact sequence of abnormal events detected. Once an evolving failure has been detected, a failure alert is generated. This alert includes information on correlated sensor abnormalities. This valuable information significantly helps in tracking the failure origin.
As industrial producers embrace machine learning for predictive asset maintenance as a crucial discipline, it needs to cope with the skill shortage of Big Data Analytics professionals. Starting salaries of data scientists often exceeding the $200,000 mark. According to IBM, the demand for data scientists will soar 28% by 2020.
Why is this important? Because when laying the foundations for industrial analytics for predictive analytics, internal resources are a critical consideration. One of the advantages of Unsupervised Machine Learning is that the algorithm is trained without the need for human input. With Presenso, the industrial plant is not required to hire additional Big Data scientists or engineers.
Can the Unsupervised model extend the lifecycle of heavy machinery and industrial equipment?
Yes. Whereas the Digital Twin is created as a custom virtual clone of the underlying (aging) asset, with Unsupervised Machine Learning the algorithm is completely agnostic with respect to asset age. The algorithm seeks anomalous data patterns – whether they are generated during the assets infancy, steady state or wear out period.
Furthermore, from an ROI perspective, a cloud-based Unsupervised Machine Learning Solution such as Presenso does not require an investment in infrastructure. In other words, asset age does not impact cost because the Machine Learning algorithm analyzes all sensor data in the same way.
Conclusion: It’s all about the ROI
Industrial Analytics for Predictive Asset Maintenance is not a theoretical academic exercise. Deploying a solution for using big data for predictive asset maintenance requires an in investment that needs to be justified from based on ROI. There are many alternative approaches to extending the life of an industrial asset, but Unsupervised Machine Learning is the most cost effective for end-of-life assets.
If you wish to learn more about Presenso Machine Learning for Asset Maintenance, please schedule a complimentary demo by clicking here.