CEOs in industrial sectors all state that AI is part of their top priority. But when it comes to real application, AI projects are not very attractive. Past demos and hiring data science teams often fail to produce the observable results of the highly anticipated digital transformation CEOs wish for.
No digital transformation without data
Today’s competitive environment, the visions drawn by the manufacturers and the excitement of artificial intelligence created by the speech industry can be deceptive in reaching the expected rapid gains. Projects made generally show limited improvement in a limited scope, which cannot be scaled.
These short projects aim to demonstrate the methodology for transforming limited operational environments. They often involve solving a problem with approaches to leveraging new technologies. These 100% digital technologies need fuel to work, and that fuel is data. So the first step is to make sure the data is available.
Firms hired to help create these quick wins will circulate between departments. They will collect datasets from different points until they have enough information to implement their solution.
This step can sometimes take time if the data is not well defined and distributed across the organization. However, it is a mandatory path to follow, as there will be no digital transformation without data.
No Artificial Intelligence without big data
After simple quick wins to kickstart the transformation, there comes a time when more ambitious projects are brought to the table and that’s when AI (Artificial Intelligence) comes into the conversation. The promises and illusion built around AI are so powerful that projects around AI and ML (Machine Learning) cannot be neglected.
The problem with current promises is that very few people really understand what artificial intelligence really means. For many, you are buying an AI the same way you buy a digital subscription, which is not true. The promise of AI is to bring in new, automated ways of using data to aid the decision process or even take over completely. This promise can only be fulfilled if the true AI, called a model, is actually trained on data from your own organization, and this requires once again the same digital transformation fuel data. The difference is that this time you need more.
Training a model really requires a lot of data covering various aspects of your business operations that you want the model to focus on, and it also takes a long time to model trend and seasonality.
This has various effects
- First, you can’t expect to train a model and efficiently incorporate AI into your operations until you’ve actually gathered enough meaningful data. And if your organization hasn’t already done so, you need to get started right away.
- The second impact is that this data collection process is not a one-time job. It doesn’t stop when you have enough data to train a model. It needs to continue, so you’ll continue to gather signals about how your business is doing to retrain your models if their performance starts to degrade in the future. This means that before your AI journey, you need to plan the collection, storage, and availability of big data to teams in your organization so teams can start looking at the data and imagine possible uses and patterns.
No industrial AI without Digital Twins
Among verticals, industrial organizations face the toughest data collection challenges. Sectors whose data is primarily related to users using their services are more fortunate. In the end, their data is not that big. Of course, we’ve all heard stories about banks or retailers accumulating piles of data. But we’re talking about a few thousand interactions per user per year. We’re talking about a few trillion events a year, even with a billion users that not many banks or retailers have.
Things are different in the industrial world, data-generating entities neither eat nor sleep. They work day and night and sometimes produce thousands of measurements per second.
The use of artificial intelligence in these verticals requires collecting and organizing this really big data. Since they are data on physical assets, it would be wise to use an approach that digitally mimics these assets, an approach called Digital Twins.
What are Digital Twins?
An entity’s Digital Twin is a set of signals from its sensors and actuators. These signals need to be monitored in a timely manner to capture the dynamics of asset operations. And the preferred technology to do this is a Time Series Database. Indeed, Digital Twins are nothing more than time series, some for sensors and some for actuators with their states.
Once you start collecting data from your assets in a Time Series Database, you can easily access the status of those assets at any time. More importantly, you can start extracting features to train models to detect anomalies and perform predictive maintenance.
Thanks to Artificial Intelligence techniques such as Machine Learning and Deep Learning, decision support applications are created for business and operational situations by generating insights from operational and business systems on top of reporting, alarm generation and historical analysis. Thus, operational and business situations that may occur in the future are predicted and it is possible to take the necessary measures or adapt the process in the best way.
With the creation of Quality Data and the necessary infrastructure to process this data can be provided both in the cloud and in-house environments, assets or processes owned by Machine Learning and Deep Learning are modeled and Empirical Digital Twins are created using the data.
Thanks to Cognitive Analytics, we can process information, draw conclusions about data, and even transfer our experiences about the process to the learning system, just like in human thinking processes.
With Digital Twins and Cognitive Analytics, the response of the system/process can be predicted in the future or in certain situations and maintenance or optimization steps can be applied accordingly.
Predictive maintenance can be done easily thanks to previously detected anomalies and root causes; By using the Digital Twin, the optimum operating value of the system or device can be found, its efficiency is increased or costs are reduced.
Artificial intelligence is on the agenda of every business, but the importance of data is often overlooked. When it comes to industrial AI, the first step towards successful implementation is the collection of all sensor data to create Digital Twins of the physical assets involved. This approach needs to leverage a Time Series Database, which is the kind of database we offer with the Cerrus Analytics Platform.
For more: firstname.lastname@example.org