Artificial intelligence (AI) is quickly changing everything around us, and the manufacturing industry is no exception. Just like with big data years ago, there's plenty of headlines and solutions promising to revolutionise everything from production lines to supply chains.

The potential for innovation is certainly there, but the reality is a bit more nuanced. AI can't do everything (yet), especially when it's not implemented correctly. So how do you tackle these kinds of projects, and what should you look out for?

In this blog, we've gathered some insights and practical information based on our recent AI panel conversation with industry experts. We'll help you go beyond the bold headlines and promises to find out what really matters – and how you can get there.

‍

‍Find the right use cases

In the back of your mind, you just know that AI has the potential to change your manufacturing plant, but where do you even begin? The key is to start with the right use cases: those that align with your specific goals and challenges.

Before you jump into any AI project, ask yourself:

• What are your biggest pain points?
  Are you struggling with quality control issues, production bottlenecks, high operational costs, or knowledge gaps?
• What do you want to achieve with AI?
  Are you aiming for increased efficiency, improved product quality, reduced waste, or better decision-making?

When you have a clear understanding of your objectives, you can explore the many AI use cases available for manufacturers. Here are a few practical examples that might spark some ideas:

• Predictive maintenance: analyse sensor data to predict when machines are likely to fail, proactively schedule maintenance, and prevent costly downtime.
• Quality assurance: detect defects and anomalies in real time with computer vision systems to keep product quality consistent and reduce scrap rates.
• Process optimisation: analyse production data for adjustments to process parameters that will increase efficiency and reduce waste.
• Anomaly detection: identify unusual patterns that indicate equipment malfunctions or security breaches and address problems before they escalate.

‍

Find the right data

Once you know what you want to do, it's time to talk about the fuel that powers every AI engine: data.  

If there's one thing that we and our integrations partners like to stress, it's that AI algorithms are only as good as the data they're trained on. In our experience, the quality and availability of data is unfortunately lacking in most projects.  

The data needs to be reliable (calibration and context are crucial here) and consistent. For example: the accuracy of a sensor is important, but if there are conversion errors or other errors in the data flow towards your ERP system or BI report, just improving the sensors won't be enough.

Also, don't just collect the data - make sure you're also capturing the necessary metadata to provide context for your AI algorithms. This could include information about machine settings, operating conditions, or external factors that might influence your data. Finally, you need a certain level of data maturity in your organisation, for example with a data centre of excellence.  

It's hard to give general advice here, since what is the ‘right’ data will depend on your use case. However, you can start by going through your data historian and ask yourself the following questions around data quality and data accessibility:

• Is your data accurate and reliable?
• Do you have sufficient historical data to train your models effectively?
• Are you capturing the right data points and metadata to provide context for your AI algorithms?
• Can your team easily access the data they need?
• Is your data siloed across various (decentralised) systems?
• Do you have processes in place to ensure data security and integrity?

‍

Find the right team

With your AI use cases ready to go and your data prepared, it's time to bring in the right people to turn those artificial intelligence visions into reality.

Building a successful AI team often means bridging the gap between IT and OT. We're seeing this distance being closed in large multinationals, but many organisations still have two different teams that have separate goals, concerns, and terminology.  

To overcome this, you'll have to assemble a multidisciplinary team that not only possesses the right technical know-how but can also communicate effectively and understand each other's perspectives. Here's who should be included:

• Data scientists can analyse and build models that uncover hidden patterns and insights.
• Process engineers bring in-depth knowledge of your production lines and processes.
• IT and OT specialists work together to bridge the gap between your legacy systems and modern AI technologies – but more on that later.
• Project managers keep everything running smoothly by coordinating tasks, managing timelines and budgets, and keeping communication effective.

Once you've assembled the right team, you'll also have to get everyone to work together. Based on our experience, here are some things that can help.

• Get everyone to speak the same language.
  ‍You can take that quite literally: in some projects, we've had to draft terminology lists for key terms and concepts.
• Where possible, take the opportunity to break down silos so representatives from different departments learn to communicate with each other.
• Organise workshops or training sessions to promote mutual understanding.
  Make sure that everyone knows what the others want to achieve.
• Adapt and evolve your technology and team.
  AI is constantly changing, and so should you. Don't be afraid to add new roles, adjust responsibilities or get help from an external expert as your scope and needs evolve.

‍

Bridge the gap

Of course, there's one important aspect we haven't really touched on so far: dealing with legacy. In an ideal world, all projects would start from a brand-new shopfloor with the latest sensors and equipment, but we all know that brownfield projects are far more common.  

Many plants still rely on isolated networks with legacy devices that exchange data locally, and connecting those with AI technologies can be tricky. You'll probably face data silos, security concerns, and compatibility issues, but it's far from impossible. Here are some things that may help:

• Don't try to overhaul your entire system overnight but work in phases.
  Identify the most critical data sources for your AI use cases and focus on integrating those first. Reaching a goal will feel good and motivate your team to keep innovating.
• Bringing some of the AI processing power closer to your machines with edge computing can help you overcome data transfer limitations and latency issues.
• Consider a hybrid cloud approach to benefit from flexibility, scalability, and cost-effectiveness for your applications.  
• Make sure to keep an eye on cybersecurity when connecting legacy devices to modern networks to protect your data and prevent unauthorised access.
• Not to toot our own horn, but an experienced partner can make all the difference.
  Don't hesitate to reach out to the experts and ask for advice or guidance.

‍

Make AI work for you

AI has the power to transform your manufacturing operations, but only if you approach it strategically. By identifying the right use cases, ensuring high-quality data, building a collaborative team, and carefully integrating AI with your legacy systems, you can move beyond the hype and unlock the real value of AI for your business.