In the last couple of years, more executives are committing resources to deploying Maintenance 4.0 programs. There is widespread recognition that improvements in production uptime directly impact the bottom line. Once strategy is in place, the challenge for many industrial plants relates to the more pedestrian issues relating to infrastructure maturity, processes and tools.
This article explores whether there are short-terms ways to jumpstart Maintenance 4.0 implementation.
Maintenance 4.0 Defined
There is no uniform definition or Maintenance 4.0 given their nascency. At Presenso, we define Maintenance 4.0 as the application of industrial analytics and automation to existing O&M processes.
Maintenance 4.0 represents a shift away from expensive Reactive and Preventive Maintenance activities that result in unscheduled downtime. How does this work? With AI driven industrial analytics, Machine Learning is applied to the data that is generated from sensors embedded within plant machinery. From this information, plants are alerted to evolving failure prior to the occurrence of downtime, thereby providing sufficient time for parts to be ordered and repairs scheduled.
Factors Limiting Deployment
In research conducted by Emory University and Presenso on the Future of Maintenance 4.0, plant level O&M practitioners indicated that most organizations have the core infrastructure in place to deploy Maintenance 4.0. This makes sense – existing operational technologies (MES, CMMS etc.) can used for Maintenance 4.0 processes. That’s the good news.
At the same time, many consider the lack of Big Data experts as the most significant challenge faced by industrial plants deploying Industrial Analytics. If the core element of Maintenance 4.0 is comprised of advanced Machine Learning algorithms, then who will be responsible for the Data Science function within the industrial plant?
Workarounds for the Lack of Machine Learning Professionals
If the lack of Big Data Scientists is the most significant factor inhibiting the deployment of Industrial Analytics, are there shortcuts to scaling this competency? The answer is dependent on which of the following approaches that plants use to deploy Industrial Analytics:
#1 Manual Statistical Modelling: Industrial plants can use either custom or off-the-shelf software packages to perform statistical modelling on data generated by sensors. Interestingly, the most common package that is used is actually Microsoft Excel.
From a practical perspective, it is difficult to scale manual statistical modelling because it requires data to be extracted and for manual processed to be applied. Of equal significance is the challenge in hiring sufficient numbers of expertise in this field, given the skill-set shortage in the market.
The term “Citizen Data Science” has been used by Gartner to describe technicians who lack formal training to perform Data Scientist or Engineer tasks. Unfortunately, this workaround is not practical; manual statistical processes cannot (yet) be performed by novices.
#2 The Digital Twin: The Digital Twin is a virtual clone of industrial machinery that simulates its behavior in real time. The analytics provided by the Digital Twin is used to remediate potential asset degradation prior to breakdown.
Although robust machine learning algorithms are applied to maintain the Digital Twin, it is not a shortcut for industrial analytics. The process to build the Digital Twin for existing plant assets is time consuming, labor intensive and expensive. Because it must be completely accurate, blueprints of the physical equipment are needed. If deviations from the original blueprints were not recorded, it will not be possible to build an accurate virtual clone.
When the Digital Twin is bundled with new equipment it provides plant technicians with high quality Industrial Analytics. Unfortunately, the Digital Twin is not a practical option for most existing installed equipment.
#3 AutoML Machine Learning: Automated Machine Learning (AutoML) is when Machine Learning algorithms are applied to Machine Learning processes. Although this sounds confusing, the concept is relatively simple. When Machine Learning is used for Industrial Analytics there are a number of manual processes that the Data Scientist or Engineer must perform. These include data preprocessing, algorithm selection and dataset calibration. With AutoML, Machine Learning algorithms replace humans to perform these tedious tasks.
The result? Solutions that are built on AutoML do not require plant-level expertise.
Is AutoML a shortcut? The science behind AutoML is evolving rapidly and it would be remiss to classify this innovation as a workaround. However, from the perspective of the industrial plant seeking to adopt Industrial Analytics, solutions based on AutoML can be used to accelerate implementation.
Avoid Unrealistic Expectations for Industrial Analytics Deployment
There are many industrial plants that cannot implement Industrial Analytics because of inadequate data management and governance. For instance, there are scenarios where data is not stored in the Historian or where plant operational staff lack access. A few years ago, a study from McKinsey & Company gave an example of an offshore oil rig that was only able to use 1% of the data that it generated. Although we have seen improvement from this (low base), rudimentary steps to improve the quality and access of sensor generated data are an important first step for Industrial Analytics.
Summary and Conclusion
Industrial Analytics is a critical component of Maintenance 4.0. Although there are ways to accelerate its adoption (for instance, using AutoML), many industrial plants will first need to address basic constraints relating to the storage and access to sensor generated data.