5 Challenges Of Adopting AI In Manufacturing

Artificial intelligence continues to be all the rage as its capabilities are rapidly increasing. Many businesses are capitalizing on its innovation early on, adopting AI into critical aspects of their business models or their products.

In the broader manufacturing industry, AI solutions in manufacturing can be a bit more tricky to implement for stable production. Some counter that the prospects of AI far outweigh the practical benefits on the factory floor.

Whether or not you are currently investing in AI solutions in your production systems, it’s prudent to be aware of the limitations of artificial intelligence. Here are 5 challenges to adopting AI in your manufacturing operations that you must consider before jumping on board the hype train.

1. Lack of Expertise

Manufacturing is currently dealing with the beginning of a critical labor shortage brought on in part by the mass retirements of baby boomers working in the field. This shortage is most widely felt in North American markets, including both the United States and Canada, where the demand for skilled manufacturing labor is experiencing a significant increase.

As we saw in our Insights on Trends for Canadian Manufacturers, government agencies all over are now attempting to mitigate the lack of workers by offering training programs designed to attract youth to the industry.

Manufacturing skills are in high demand, and experienced, skilled factory workers are few and far between. The labor shortage currently being felt is a big enough hurdle for manufacturers to jump already.

This doesn’t even touch on the difficulty of finding skilled workers to manage AI applications in a factory setting, which will be equally as difficult. While younger workers are entering fields like data science at a greater rate than trades and manufacturing, few are eager to take their AI experience to careers in the manufacturing industry.

There are many reasons for this – lower potential salaries, preference for research over physical production, lack of awareness in manufacturing applications for technology, etc.

Simply put, manufacturers who want to deeply invest in AI solutions on the factory floor need to be aware that these solutions will require many specialized workers to oversee their upkeep. At this current time, manufacturers must also be the ones to attract and retain talent, through compensation, growth opportunities, technological investment, and other methods.

There isn’t an easy workaround, such as upskilling current shop floor workers in AI, because it would be a costly and time-consuming investment, even if companies were able to convince their workers to shift their careers to an entirely different type of industrial work.

2. Integrations

This next challenge is well-known to most manufacturers, since they encountered it with the development of IoT devices: much legacy equipment exists in manufacturing settings, and integrating new technologies with existing ones can be frustrating and costly.

AI is developing for the most part outside of the manufacturing context, so scientists aren’t prioritizing making their AI inventions easily transferable to the shop floor setups. It is unclear if AI capabilities will be able to boost current production capabilities without costly upgrades to make everything work smoothly together.

In addition, this lack of interoperability within different systems would require even more workers to manage all the extra equipment and stop-gap solutions constructed just to coordinate actions. Then, given that the pace of AI’s development is at super-speed, even more upgrades will be needed as AI advances far beyond the capabilities of this legacy equipment, and so on (you can see what we’re getting at here).

Some manufacturing leaders might just forego all the trouble and stick with the tried and true machinery they already have.

3. Improper & Incomplete Data

One of the reasons why ChatGPT has managed to evolve so quickly is because it is built on massive, expensive large language models (LLMs). And then the AI needed to be tweaked and refined to limit hallucination and other errors or gaps in knowledge.

AI is only as good as the input data it is built upon. If the data sets it analyzes are sparse, output will be inexact or biased, as explored in our article on Automating Inequality. This can have a dramatic effect on real-world results when it comes to production quality and quantity. Inadequate conclusions driven by AI can add tons of waste into a lean production environment at best, and fail to meet customer expectations at worst.

Moreover, data pools need to be properly maintained at a continuous pace; it’s not enough to simply store all production metrics loosely in some central database. Someone (preferably multiple someones) needs to outline what data is collected, why it is necessary, and in what measurements and scenarios it must be collected, in order for it to be a useful exercise.

A good intermediate step is to implement a solid manufacturing execution system (MES) for sorting worker and product data like run times, defect rates, etc. At the same time, manufacturing experts already know that a background MES is not totally sufficient for insightful direction.

4. Unplanned Downtime

Factories experience difficulties such as sudden downtime, energy outages, and material shortages. Proactive factories have SOPs to prepare for reacting to these unexpected events when they do occur, but it’s impossible to completely avoid them.

Unplanned downtime in production occurs outside of AI’s purview. This means that relying on AI can lead to teams becoming complacent about anticipating and adapting to the unexpected.

Now, this challenge is outside of scenarios where automated systems can “predict” machine breakdowns. In these cases, automated systems keep tabs on the health and age of equipment, and flag workers for regularly-timed repairs when it’s reasonable to expect that parts are worn down and need to be replaced. This is different from AI because it largely relies upon IoT devices like sensors to transmit data about operation pace and maintenance needs.

AI is not good at expecting the unexpected, or “thinking outside of the box” – and this situational awareness is a type of skill that cannot be offloaded to automated or artificial intelligence systems. For your company’s safety and efficiency on the shop floor, it would be a mistake to overly rely on AI due to the threat of unexpected emergencies.

5. Quality Assurance & Industry Compliance

Finally, using AI in manufacturing is a potential misstep because it is unclear if the technology can promise the same quality standards that are currently expected within certain industries. In other words, don’t get too excited about prospects and put the cart before the horse.

For example, companies have promised fully autonomous, AI-driven personal vehicles for over a decade now, and the technology is still woefully far from what has been promised. It is certainly much better than it was when the idea was first propagated, but no one would say the actuality has met expectations.

Over-promising and under-delivering with AI can be disastrous in industries with dedicated compliance and quality standards, such as pharmaceuticals, automotives, and consumables like food and beverages.

Perhaps a solution would be to use AI in the manufacturing process but then double-check its work with an extra quality assurance screening at the end of the production process – but adding an extra step in the production process feels just like adding more waste to the manufacturing cycle.

This brings us back to the most efficient method for ensuring quality: building into the process in the first place with work instruction software. It’s too much work – and not nearly precise enough – to try to add quality in as an extra step in the production cycle. It’s the better option for manufacturers to instead make sure that quality is embedded into the product build, by having workers follow precise, interactive guidebooks that guarantee actions are performed the right way every time.

AI is no match for the insight brought by experienced workers who know the ins and outs of production and have standardized their tribal knowledge through work instructions.

Overall, it’s important that manufacturing explores and evolves with technological innovations like AI, particularly as we move into Industry 5.0 which centers the collaboration of worker and machine. However, those who aim to adopt AI solutions in their factories should be aware of the many potential pitfalls as well as the necessary resources to properly implement their use.