In our short series of challenges facing the forward-looking manufacturing professional ready to adopt AI, we have so far talked about data and change management challenges.
Now it’s time to talk about technology challenges our customers have encountered and how to address them. Here are the four most relevant ones:
Technology Challenge 1: Integration with Other Systems
We said it before but it bears repeating: finding a good AI model as a starting point is not the main problem, good models are readily available in “model zoos” from where they can be downloaded, adjusted, and then trained. Once your model is ready in a Jupyter Notebook, connected through an API, or similar there is more hard work ahead of you: you need to make sure your model integrates with your other systems, like your ERP and MES applications, quality systems, Cloud services, databases, and – critically – your PLCs.
This is a difficult task that can be very time-consuming. Let’s take the case of integration with PLCs: you might use PLCs from different manufacturers so you don’t just need to make sure your models talk to your Siemens and Rockwell but also your GE and Honeywell PLCs.
If you are a manufacturer with high-speed requirements that makes products in a matter of milliseconds you face another thorny issue: making sure you are able to synchronize your AI with that level of performance.
Developing these integrations requires domain expertise in both AI and manufacturing, something that at least currently is hard to find and a major reason why buying a solution purpose-built by experts is a better option than trying to go the “do-it-yourself” route.
Technology Challenge 2: Capacity of Your Network Infrastructure
Data, especially images generated at high speed, add up really quickly and transmission of these data even within the plant might become the limiting factor. You might have to upgrade your 100 MB line to 1 GB or more to ensure a sufficiently fast transfer of the data between the sensors, algorithms, and the PLC that might take corrective action, such as blowing a product that was identified as defective off the production line.
We recommend a careful analysis of the existing infrastructure and future infrastructure needs driven by AI adoption to avoid delays due to necessary infrastructure updates. An experienced partner can help you estimate the requirements and recommend solutions that support not just an initial AI application but the adoption of many models on the shopfloor.
Technology Challenge 3: Cloud vs. On-Premise Data Storage
Another big issue is data storage, especially if vast numbers of images are generated in high-speed manufacturing or large, high-resolution images are required, e.g. to see very small defects. Even terabyte-size storage can fill up quickly and you might need petabyte storage capabilities sooner than expected.
There is a second, important decision that you need to make once large amounts of data are accumulating: where to store that data. If you have kept your data in the cloud so far, the monthly bill from your cloud provider will be frighteningly high once you store all these images and an on-premise data storage solution will likely be the most cost-effective way to go. Alternatively, you can carefully select the images you want to keep, e.g. you can opt to keep only images showing defects while only keeping a small sample of good products (the overwhelming majority of images).
Our experience has shown that given the data storage needs of typical manufacturers of automotive parts, consumer packaged goods, or other products manufactured in high quantities the investment in an on-premise sever can be recouped in about six months.
Adding in-house storage requires planning as delivery of large servers can take weeks or even months.
Technology Challenge 4: Security of Data and Systems
Keeping your systems and data secure is another technology challenge you need to plan for. While many manufacturers have switched to cloud storage over the last few years and with that step opened up their systems somewhat to the world outside, the preferred way of the Infosec groups is still working in their own, isolated environments.
AI adoption will force an additional degree of openness on manufacturers and Infosec will have to figure out how they can balance data and system security with the need to give external partners access via VPN so they can perform tasks like model training or ensuring that the sensors/cameras work as they should. This is as much a change management as it is a technology challenge and therefore needs attention from the very start so Infosec has a reasonable lead time to implement solutions.
Adopt AI Now, or Be at a Disadvantage Soon
The phrase “paradigm shift” is horribly mis- and overused but in the case of AI it is a fitting description of the sea change that is happening. The value AI adds to companies of all types, including manufacturers, is simply too large to ignore. Quality control, maintenance, analytics, and many other applications will be powered by AI in just a few years and companies who adopt now – despite initial technology challenges – will have a competitive advantage for some time to come. In a few years, manufacturers who haven’t adopted AI will face a serious disadvantage: they will no longer be able to keep up with their AI-enabled competitors in terms of quality and cost.
The data, change management, and technology challenges associated with AI implementation are real but solvable, the value AI brings to manufacturers is clear and easily provable through a proof-of-concept study. Investing now and gaining an advantage versus playing catch-up later is the smart strategic decision.
If you want to get started with your AI implementation and need help and expertise please remember: we are only an email away.