Unprecedented technological transformation worldwide, and the increase in demand is prompting industries to reimagine their production processes. Manual operations are being integrated with automated solutions, but the next stage of development in industrial AI remains stagnant due to several reasons. However, the focus remains fixated on reducing operating times, enhancing production capabilities, and cutting downtimes.
While theoretically, industrial AI not only addresses these issues but also enhances the overall manufacturing capabilities, its development is being stalled owing to a number of reasons.
Let us first delve at a high level into the question of why industrial AI is lagging as a whole compared to other fields.
Lack of industrial understanding
The key reason is that people who run industries are not that knowledgeable about AI and people who know AI have very little understanding of industrial processes. Industries are extremely diverse and complex. Understanding them requires in-depth understanding of several complex fields such as thermodynamics, kinetics, vibration, heat transfer, fluid mechanics, structural dynamics, electromagnetics, materials science and more.
The number of people with knowledge of these is limited and acquiring this is often difficult and time-consuming. This knowledge gap makes it difficult for industrial engineers and AI professionals to work together to make significant progress in the Industrial AI space. Further, the risk of AI systems going rogue or hallucinating can be catastrophic, with possible loss of human life as well.
Data accessibility
Let us now look into the data accessibility question. As we all know, the availability of clean data is fundamental for the training and development of AI models. Let us start by identifying where industrial data is resident. Most industrial data is generated by instruments and sensors like pressure transmitters, flow meters, contact switches, relays, etc. This data is generally read by a controller like a PLC that is housed within a machine and controls its operations.
So, in theory, all this data is available with the machine controller/PLC. There are several manufacturers of these PLCs like Siemens, Fanuc, Rockwell, Schneider, ABB etc. Data from these PLCs can be accessed only by authorised users using communication protocols supported by their OEM. Most suppliers of machines will not provide access to their PLC, let alone take the data out. In many cases the PLC would not even have a communication interface like RS232/RS485/Profibus.
The reality
Let us assume that we cross the hurdle of access to the controller and the controller having a communication port and the network protocol being open. The next question is going to be how to wire this network port to a computer since most of these machines are operating round the clock and plant owners may not risk anyone touching their machines for the risk of something going wrong.
Next let us consider a factory that is fully networked and has fully blown SCADA software implemented with a historian. This would be an ideal situation, as data is already flowing into a central database. However, even in this ideal situation, you would be required to go through a lot of paperwork and procedures to gain access to the data and significant effort would be needed to properly identify data for specific machines/process parameters.
Another requirement around data beyond access is the ability to tag data to represent various status conditions. Data may be available but it is hard to identify the status of the asset that it represents.
Finally, machines generate real-time data so the volume of data that is being generated is time-series data that is high volume and not very appropriate for AI algorithms. As an example, a vibration sensor generated 25 thousand samples per second, and you need to perform an FFT on this time series data to take it into the frequency domain before it can be processed.
Conclusion
In conclusion, access to industrial data is limited by machine OEMs’ perceived risk of exposing data, lack of communication hardware, proprietary protocols, data security, the real-time nature of data and lack of wireless networks requiring a hardwired approach that is difficult in retrofit situations. All this limits the ability for AI to deliver its true value.
Source :- We are proud to have been featured by Manufacturing Today India The original article is republished below with permission.