Artificial intelligence has been causing quite a stir lately, including in the business world. However, it is worth noting that the term ‘artificial intelligence’ is very apt, as this technology has nothing to do with true wisdom. AI is only as intelligent as the database it has access to.
What do I mean by that? If we pay attention to social media and all the hype surrounding AI in the B2C model, we might get the impression that the same trends will work in B2B, particularly in manufacturing companies.
Nothing could be further from the truth! While some tools or procedures can be borrowed from the B2C sector — such as ChatGPT and techniques for generating videos and graphics — these constitute only a small part of the broader marketing support available. However, it is important to remember that these tools use public databases that are available on the internet.
So, how can we set the lever in motion in B2B?
If you want AI to become a business lever, you must first and foremost take care of your databases. And this is where the problems begin…
A typical company may store data in various locations. By ‘data’, we mean information that can be used by artificial intelligence for analysis and subsequent actions. What locations and data are we talking about?
Let’s start with financial data. Some companies have it better organised than others. It is stored in IT systems and sometimes in various spreadsheets. The scope of this data can also vary; for example, we can record every hour of an employee’s work or only record incoming and outgoing invoices.
A second potential source of information for AI is data from our know-how, such as knowledge of production technologies or management techniques. Imagine a scenario in which we build a model based on our documentation that works similarly to ChatGPT. How would it work? First, we place a set of data on a server in the cloud. Then, a language model such as ChatGPT learns from this information to help us identify problems within the company. To illustrate, we have a large amount of project documentation. Searching through it for specific information is time-consuming, so we ask the language model to find specific details for us (e.g. how we solved a particular problem in project X in 2022).
We act as teachers for such a language model, providing it with knowledge in the form of company data, which it assimilates in a similar way to GPT Chat. This is also relatively easy to achieve.
What challenges does AI face in a manufacturing company?
It is more difficult in manufacturing companies because much more data is collected, including financial and know-how-related information, as well as other types of data, such as information strictly related to the specifics of operations, the storage of certain items, and their categorisation. This includes information on where specific materials and tools can be found and how they should be stored to ensure they are issued in the correct order.
The most difficult challenge for AI is the manufacturing process. For artificial intelligence to be implemented at this stage, we must first automate the flow of data. In short, the basic information available to every company (e.g. the number of items produced on a given day, the number of products made on a given machine or by a given operator, shortages and the amount of raw materials used) is insufficient. Much higher resolution information is needed, ideally enriching the databases with a range of other information. This could include data directly related to the manufacturing process, e.g. quality or manufacturing efficiency. Therefore, to implement artificial intelligence successfully in manufacturing, we would need to focus on ensuring much greater data detail and frequency of collection than before. In addition, changes would need to be made to the way this information is stored.
Automation is an ally of digitalisation.
Let us take a look at the first of the above issues: the frequency of reporting. Currently, most manufacturing companies collect information at least every eight hours. This tells us how many units of a given product have been manufactured (including both defective and fully functional goods) and how much raw material of a given type has been used. However, this is not enough to justify the implementation of artificial intelligence.
Furthermore, even if we have large databases, if the process of collecting, processing and storing them is not automated, they may be subject to human error. If artificial intelligence has access to false data and the sample size is small, it will not be able to correct the data.
Automation can help here, eliminating the risk of human error and ensuring greater data resolution and correlation with other information.
How important is automatic production reporting?
To illustrate, consider a company that records the number of goods produced by a machine during a production shift. Dividing this number by the number of hours gives us the cycle time for that particular shift. Ideally, this cycle time should match the target time. But what if it doesn’t? In that case, we would need to ask ourselves, ‘Why?’ Did the machine work for a shorter period of time because of a breakdown, for example? Has the cycle time increased for some reason? Perhaps a tool has worn out? After all, if our vacuum cleaner is clogged up, it will take longer to operate, and eventually the device will stop working altogether. Similarly, if our knife is blunt, it will cut less efficiently than a well-sharpened knife.
This shows that increasing the resolution of information provides more detailed information that directly informs us about the cause of a problem and places the average cycle time in a broader context. The same is true of efficiency and any business component. The same is true for quality parameters. If we continuously monitor quality parameters (e.g. every half second) and take various variables into account, we can find many correlations between them.
What is the application of AI at the stage of production data analysis?
Of course, we might conclude that there is too much data to do anything with it and that we are unable to process it. This is where artificial intelligence comes into its own, as it requires a large amount of information to work effectively. Thanks to extensive databases and broad contexts, it can provide us with a lot of relevant information. First and foremost, it can suggest what we should change. Should we change the machine used to manufacture a specific product, for example? AI can tell us which machine to use to manufacture a given product. It works this out based on the success history of one machine compared to another.
Thanks to data and artificial intelligence, we can also implement predictive maintenance. This enables us to predict failures based on correlations between various phenomena that have contributed to failures in the past. For example, by analysing parameters relating to oil viscosity and temperature in relation to failures, we can predict which combinations of these indicators will result in machine damage. Therefore, by analysing the data, we can prevent failures. We can also look for correlations between other variables and machine failure rates.
Thanks to high data resolution and the automation of data collection, artificial intelligence enables us to analyse data immediately. There is no need to waste time interpreting it, a process which is labour-intensive and requires specific knowledge. We can ‘talk’ to a language model or use a system that, based on analytics and machine learning, will send us specific reports highlighting what we should pay attention to.
