Caspar Herzberg |
A couple of years ago industrial internet of things (IIoT) predictive asset maintenance was merely on the radar of forward-looking executives. Today, the technology is an integral part of the digitisation roadmap for most manufacturers. The traditional industrial sector does not have deep expertise in the disciplines of artificial intelligence (AI) and machine learning, leaving a void filled by consultants and solution providers. The tight labour market means that it is important for manufacturers to compare total cost of ownership (TCO) for various solution alternatives. It is our belief that IIoT predictive asset maintenance is most effectively delivered through a cloud-based software-as-a-service (SaaS) platform.
There are, however, alternatives to IIoT predictive asset maintenance.
First, there is a new generation of hardware-based solutions that capture and interpret sensor data generated by industrial machinery. Although some of these provide valuable insights, they all have one thing in common: the need to use facility-level labour resources to deploy and maintain the systems. It is with some irony that, as traditional IT departments migrate from local hardware purchase to cloud-based solutions, some industrial plants are considering hardware-based solutions for machine learning.
The second alternative is the ‘digital twin’. For the past couple of years Gartner has included the digital twin as one of its top 10 strategic technology trends. There is no doubt that a virtual clone of a machine asset can provide valuable asset performance insights in real time. At the same time, to develop a virtual clone requires access to completely accurate physical blueprints and operation formulas of the machinery. A software model is then designed based on these blueprints.
The digital twin is a powerful technology with multiple operational applications, including predictive maintenance. However, industrial plant technicians are required to train the machine model on the underlying behaviour of the asset and that is time-intensive. The labour costs associated with this knowledge transfer are not always considered when estimating the project cost. If internal labour costs are ignored, TCO is distorted, thereby inflating the financial benefits.
Another option is a platform-based solution. More industrial plants are recognising the untapped value of operational data, which has opened the market for advanced statistical packages for the application of big data. Even if these solutions are priced reasonably, the bigger challenge is finding machine learning engineers with the expertise to select the optimal platform algorithm for various datasets.
There is a dearth of machine learning experts, and manufacturers are unable to compete with the compensation levels offered in the financial and high-tech sectors. A number of band-aid alternatives have been suggested, including training ‘citizen’ data scientists, or tapping into the student population of institutes of higher learning.
The reality is that these statistical packages require deep subject matter expertise that is both difficult to find and expensive. Even in cases where building a machine learning centre of excellence is achievable, it comes with a labour price tag that needs to be accounted for.
What is the best alternative? SaaS-delivered IIoT predictive maintenance solutions offer two major advantages. First, there is no need to install hardware within the manufacturing plant because big data captured by the existing sensors is analysed in the cloud. Second, instead of the manufacturing entity hiring experts in big data and machine learning, these functions are performed by the AI vendor that has domain expertise in these areas.
As IIoT predictive maintenance still in its nascency, confusion remains about the solution landscape. For manufacturers that focus on TCO, SaaS-based solutions provide a low-cost alternative to existing hardware and software offerings.
Deddy Lavid Ben Lulu is the CTO of Presenso