Newsroom

Join us on our exciting journey bringing AI and machine learning into the industrial day-to-day business

Type:
It has already been proven many times: In production environments across a wide range of industries, artificial intelligence can help prevent process disruptions. Among other things, this increases productivity and efficiency, reduces waste, and saves carbon – enough reasons to take a closer look at AI in industrial production.

Nevertheless, process engineers rarely turn to AI when it comes to optimizing processes. All the more they limit themselves to the manual review of process data and the general evaluation of data.   

AI could help to quickly and easily open up completely new insights into a problem, bringing new and sometimes unexpected results to the surface. This would enable process engineers to understand production processes even better and solve or prevent all sorts of problems, or at least detect them well in advance so that they can take countermeasures. 

However, corresponding AI systems are often difficult to exploit. Their use is complicated, and the systems must first be “fed” with comprehensive domain know-how before usable results can be obtained. And in the end, these results are often only useful for data scientists and are therefore of no direct help to process engineers. 

It is high time to take a closer look at what is important when selecting the right AI tool, which process engineers can use in their daily work to quickly and easily obtain a “second opinion” regarding any process malfunctions. The following 8 hints will help you to find the right system. 

1. Tolerance to varying data quality 

Again and again, AI tools promise fabulous insights and improvements. However, it quickly becomes clear that the required data quality is so high that it would take a very long time to prepare data accordingly. After all, industrial process data in particular is subject to wide swings in quality and availability. If a perfect data basis has to be created first, such a tool is unsuitable, at least for everyday use.  

2. Efficient handling of large amounts of data 

Especially in complex processes, a very large number of sensors are usually employed, since every relevant aspect of the process is measured, sometimes several times. If the clock rate of the signals is also high, large amounts of data are quickly generated – we are talking here about at least gigabytes, if not terabytes.   

After all, the data period to be examined must also be selected long enough so that any process disturbances occur several times in the data and can thus be better investigated by the AI. It is also often not clear which signals can be safely excluded from the root cause analysis because they are not relevant to the question. An AI tool should therefore be able to handle such huge amounts of data efficiently. 

3. Robust results even with small amounts of data 

Many AI approaches require large amounts of data, especially for training – and this rules out the use of AI for data-poor environments. The systems simply cannot align themselves to the issues at hand. So a key requirement is that the AI tool of choice is able to deliver resilient results even with small amounts of data. 

4. Handling of unsynchronized time series data 

While there may be a great amount of tabular data in production environments, unsynchronized time series data makes up the bulk of process data – finally, such data is provided by any sensor that measures a particular quantity.   

Occasionally, the intervals between the measurements of a single sensor fluctuate, but certainly the measuring intervals of the different sensors differ from each other. Some measure per second, others per minute, again others irregularly. A good AI must therefore simply be independent of the frequency of the different signals, as long as they refer to the same period of time. 

5. Independent distinction between relevant and irrelevant data 

An AI tool suitable for everyday use should work in a goal-oriented manner. This means that the users can concentrate on the question and do not have to differentiate which data is relevant or not for the respective question – the AI should be able to handle this. 

It must be able to filter out from the potentially huge amount of data which signal patterns or even relationships between signals are relevant with respect to the specific question and which are not.   

This capability also means that, once the data has been imported, a wide variety of questions can be answered without the need for time-consuming and cumbersome tailoring of the data to the exact question each time. 

6. Easy and intuitive to use 

Process engineers are not data scientists – and they usually don’t want to become one, just to be able to use AI beneficially. Therefore, the right AI tool must be flexible and intuitive to use when formulating the question.   

The question is the central element with which process engineers indicate what they expect from the AI. Everything else should happen independently in the background. In particular, no additional information or even decisions should be required that do not relate to the question, but rather to how the question should be solved.   

7. Minimal data preparation 

Data synchronization, data filtering, data cleaning, data preparation – these are all preparatory steps associated with the use of AI. In fact, in most cases they occur before AI can even be used to add value. The right AI tool should only rely on this to the smallest possible extent.   

In day-to-day work, data preparation should require no more than a simple conversion to the appropriate target format. Anything beyond that should be the task of the AI. Only if the effort for data preparation is minimal, the AI will be consulted on a regular basis. 

8. Interpretable results 

This is perhaps the most important point overall. AI is to be used to optimize industrial process flows – it is usually of very little use just to know what is happening without identifying the why.   

The right AI tool must therefore deliver results that offer full transparency about how a result was achieved – or more precisely, which signals or signal combinations are involved and how. This is the only way that process engineers can actually draw reliable conclusions that allow for a real improvement of the processes. 

Conclusion 

These 8 points are highly relevant when it comes to deciding for or against an AI tool. Process engineers are busy experts who should be spared time-consuming excursions into the world of too demanding or simply inappropriate AI tools.   

The right AI tool for the manufacturing industry must be tailored to the needs of process engineers:  

• simple in data import  

• flexible in question formulation  

• independent in all other requirements  

• transparent in the output of results.  

Such a tool has a real chance to find its way into the daily work of process engineers and to be included as a matter of course as a “generator of added value” in the solution of daily tasks and problems. 

Stay up to date!

Do you like what you see? Then stay up to date and sign up for latest news and blog posts!