The Smart Factory is moving away from being a boardroom strategic initiative to a factory floor implementation. At every industry and customer event that I have recently attended, there is a receptiveness to new technologies that support digitization and Industry 4.0. Industry analysts are predicting astronomical growth rates and companies such as GE and Siemens are re-inventing themselves to capture a share of the Smart Factory category.
Now for the bad news. Let’s put some of the vendor-induced exuberance aside. When I speak to C-level executives there is a shift in tone. Although there is a recognition that the Smart Factory is inevitable, there are some real concerns that the costs may not be justified and that smaller and medium size factories may not accrue the same benefit as larger industrial plants with more resources. (Ironically, some of the larger factories with significant sunk capital are concerned about how to adjust to digitization within a reasonable time frame and budget.) Apart from the generic concept that extracting insights from big data and applying them to factory production and machine maintenance, there is almost no common definition of the Smart Factory or a road map for how to develop one.
Based on some of my more recent conversations, I would like to share some of the concerns that are often missed by the vendor-sponsored industry publications:
The Cyber Security Threat is Perceptible
Executives are worried. It seems like every day another article appears in the media about state sponsored cyber warfare or a data breach. The most recent data from the Department of Homeland Security’s (DHS) Industrial Control Systems Cyber security Emergency Response Team (ICS-CERT)is alarming for industrial plants. In fiscal 2015, ICS-CERT responded to a total of 295 cyber incidents, up 20 percent from the previous year. In the critical manufacturing category (including makers of machinery, electrical equipment and transportation), the DHS investigated 97 cyber security incidents.
After the cloud concept was first introduced, the first wave of objections to it was based on security considerations: when the organization’s internal staff maintains its data, it controls access to unauthorized third parties. The assumption underlying this argument is that enhanced control results in better security.
Because the cloud is a critical component of the Smart Factory, the debate about on premise versus cloud has been re-ignited. In some ways, openness to the Digital Twin (see section on Machine Learning) can be attributed to the perceived advantage of storing factory data within the on premise virtual clone of the machine.
For industrial plants considering using a cloud-based solution for predictive asset maintenance, it is prudent to inquire about how a company transfers, stores and accesses a customer’s data. For instance, in the case of Presenso, our security model is the same as used in the online banking industry. Furthermore, we do not record machine asset information: our advanced algorithm searches for patterns of abnormal sensor data and is an unlikely target for state-sponsored or criminal cyber-attack.
Conversely, one should not simply assume that an on-premise solution is more secure. Cyber security is a function of data policies and protocols, employee training and various other factors. In my experience, there are many industrial plants that lag behind other industries and therefore are more vulnerable to attack.
Betting on a Technology Vendor is Risky
Change is the new normal. The Smart Factory requires a significant investment in technology ranging from automation tools to business intelligence. The issue is that while a Smart Factory strategy requires flexibility and rapid adoption of new processes, some of the underlying technologies require long term commitments.
Specifically, the list of vendors that now offer the Digital Twin is growing. The recent addition of GE Predix and Siemens MindSphere to the mix has created a great deal of confusion. Why? Because these industry behemoths are asking factories to make a long- term investment in their platform without sharing or committing to a technology roadmap.
I am not suggesting that GE, PTC, Siemens, SAP etc., will disappear. At the same time, it is not clear which, if any, entity will dominate the category. As GE rushes to build an entire vendor ecosystem around its platform, plant owners are concerned that will be locked into a Digital Platform that will limit their own flexibility long-term.
The Digital Factory is not always Economically Viable
Over the years, I have heard vendors argue that if you don’t purchase their solution, your business may suffer the consequence. Although I shy away from this type of hyperbole, there is some validity to the concern that companies that do not embrace digitalization and the Smart Factory could lose their competitive advantage.
Ironically, there is a more acute (and contradictory) concern that over-investing in the Smart Factory could hurt a factory’s bottom line. In the traditional factory environment, one buys machinery and can amortize the cost over an extended period. But with the Smart Factory, virtual equipment may not last that long and can become redundant relatively quickly.
Therefore, there is a risk of over-investing in technologies with a short- term shelf life. Miscalculating the lifetime of an asset could lead to distortions in ROI.
Presenso is cloud based and does not require investments in hardware and other infrastructure but many other Smart Factory solutions can be cost prohibitive, especially for smaller plants.
Organization Change is Hard to Implement
The Smart Factory is not a technology solution or suite. The Smart Factory is a new mindset. The ability to find new ways to monetize big data is not simple and there is often a lag between the promise of digitalization and the organization infrastructure that needs to operationalize it.
If change management was easy, there would be less job security for overpaid Booz Allen and McKinsey consultants.
In the age of the Smart Factory new skills are required that are both technical and business oriented. The good news is that industrial plants will have access to massive amounts of valuable data. At the same time, if frontline employees are not trained and empowered to act on the insights in close to real time, the facility will not realize the value from its Smart Factory investments.
Finally, there is often a temptation to allow technology vendors to support corporate employee trainee initiatives. The Smart Factory is your core strategy and allowing a vendor to create momentum or stickiness for their solutions within the organization is not wise for the long term.
Misunderstanding Big Data Machine Learning
Gartner has stated that Machine Learning is one of the top 10 technology trends for 2017. For asset maintenance, plants can analyze sensor data in real time and identify abnormal behavior patterns. The learning algorithm can use big data to predict machine failure before it occurs, thereby preventing machine or factory downtime.
There are two models for Machine Learning: Supervised and Unsupervised. In order to use Supervised machine learning, the factory needs to create a virtual replica of the physical machine. To build this Digital Twin, the physical machine needs to be recreated via 3D modelling using the machine blueprints. Any variances between the machine blueprints and the physical machine could result in accuracies in the model.
Undoubtedly the Supervised Machine Learning solution offered by the Digital Twin is compelling and sensors provide an accurate real time virtual depiction of the actual machine asset.
The concern on the part of factory owners is that they simply cannot dedicate the resources that are required for deployment. In order to use the Supervised model, the algorithm needs to “learn” how the physical machine works. This requires labor intensive input from both big data engineers and facility staff.
The alternative to Supervised Machine Learning is the Unsupervised Big Data Machine Learning solution provided by Presenso. How does it work? Presenso streams all sensor data from the Historian database to the cloud and advanced algorithms are trained to detect abnormal data patterns. This approach is significantly less labor and cost intensive than Supervised Machine Learning because the model is agnostic with respect to data or sensor type, machine age or class.
Rather than learning to understand the actual physical asset, the Unsupervised model can detect potential failure by using advanced artificial intelligence algorithms.
Conclusion: change should be incremental
If there is one thing that I have learned from my experience in IT management, it’s that change does not need to happen overnight. Another term for an early adopter is a beta customer. The Smart Factory can be adopted carefully and incrementally. A new and flexible mindset is needed for the Smart Factory, but the same level of prudence that is required before physical capital expenditures should be extended to Smart Factory investments.