October 3, 2023 How Manufacturers Are Gearing Up for Machine Learning

Machine learning will transform industrial environments. But it is not a plug-and-play solution. Deployment requires thorough preparation.

By Mariusz Stolarski, Global Head of Technology at Mobica 

Interest in AI and Machine Learning (ML) has soared this year in the wake of the emergence of highly impressive Generative AI (GAI) solutions such as ChatGPT.  

ML adoption had already doubled over the last five years, but that growth is expected to accelerate. Gartner found that almost half of executive leaders have increased investments in AI since the launch of ChatGPT, while McKinsey has reported that almost two thirds of organizations will do so by 2025.

In the manufacturing sector we have already seen these solutions being integrated into working environments. For example, it’s no longer unusual to see ML-enabled Automated Guided Vehicles (AGVs) operating autonomously in warehouses. New applications for the technology are also emerging. ML is enabling Computer Vision solutions that can detect faults on production lines or potential dangers in industrial settings. It is also helping to optimize production parameters to increase yields and simulate more efficient workflows using digital twins.  

As this technology continues to be rolled out, it’s important that all manufacturers are gearing up for this change. It’s hard not to agree with Kai-Fu Lee, the former president of Google China, who said: “Artificial intelligence is like a speeding train. We can see it coming, but by the time we realize it’s here, it will already be ahead of us.”

If you’re considering AI as an option for your business, we’ve identified a few key pointers based specifically on our experience in the manufacturing sector.

Start With the Data

For ML/AI solutions, your results are entirely dependent on your data. Your first step with any AI implementation must be to establish a robust data collection and management system. The quality of the output is only as good as the quality of the input. If the data is biased, the AI will be as well. This will mean taking into account everything from data type, quality and robustness through to security – particularly when information is in transit.

Data collection is likely to include the installation of an array of sensors, monitoring factors such as vibrations, temperature, humidity and so on, as well as streamed video and audio. Our experience from a number of projects we are involved with, clearly shows how important it is to adjust the sensors’ capabilities for the specific use cases. This may require adaptation on the production of dedicated sensors (OEMs), customisation of sensors’ embedded applications or edge computing of the collected data. 

Transferring and managing this type of data often requires investment in a broad mix of technologies. You will need to choose the right technology for your set-up from a wide range of options including wireless sensor networks, 5G private infrastructures, Industrial IoT platforms and potentially a combination of Edge and Cloud computing.

In addition, think about how this data will be shared with your teams. New dashboards and interfaces will be required, possibly using the power of the latest augmented and virtual reality solutions.

Bring in the Experts to Train the Tech – Continuously

To make any ML solution effective, expert data analysts will be required to assess the quality of the data, and to train the model.

Firstly, this means understanding the environment in which the solution will be operating to identify potential anomalies. For example, seasonal workloads or heat could affect the readings reported by sensors. A ‘normal’ reading in winter when the weather is cold might be considered abnormal in the summer months.

Additionally, you’ll need your analysts to negotiate the potential for ‘data drift’ and the risk of bias in the algorithm on an ongoing basis – to check the validity of the information being supplied and the results being produced. Initial models have been known to be over-simplistic or to include a degree of inadvertent bias. Continuous monitoring, including retraining where necessary, is essential to keep this under control.

Take Your Team Along With You

Acceptance by your workforce will be key to the success of your project. It is highly likely that an ML deployment will fundamentally change company processes and affect job roles, so I strongly recommend running a change management program in parallel.

This will involve educating and training existing employees on how to work with their new environment. It may be that the introduction of solutions such as computer vision will help to remove mundane and unwanted tasks, but for the people who currently perform these jobs, a new technology may be perceived as a direct threat to their livelihood.

These employees often hold the know-how that will be vital for training the models, so make sure that you communicate any change sensitively.

Take the Time to Think it Through

ML is a new technology and there are still unknowns that need to be given proper consideration. Use cases, benefits and ROI all require careful analysis, and, crucially, so do the potential risks involved.

Mistakes can happen and ML isn’t always able to explain why. Not enough companies are making sure that they have an answer to an ML- based blunder. According to IBM’s research, 61% of organizations have not been taking steps to ensure they can explain ML-based decisions.

ML is not a plug-and-play technology. Beyond the solution itself, a very broad set of skills will be needed to make this a reality. Specialist expertise will be required to install the technological infrastructure and then for the ongoing data analysis.

Understanding all the risks, as well as the potential, that ML can bring will greatly increase the likelihood of a successful implementation.

Start Thinking About It Now!

The potential benefits for businesses are substantial, from improved productivity, production quality and safety, to the efficiencies delivered through better processes, supply chain management and energy usage.  

There is a lot to consider, but the sooner you begin the process the better. While it is never constructive to devote time to worrying about the threat of competitors or new market entrants stealing a march on you, ML undoubtedly offers an opportunity that could be truly transformative. You just need to be ready to take advantage of it.

Mariusz Stolarski is the Global Head of Technology at Mobica, a Cognizant company. He works closely with some of the world’s largest industrial OEMs on innovative areas of tech development, including Machine Learning (ML).

Subscribe to Industry Today