
Industries embrace AI with cautious investment. According to Digital Commerce 360, over 70% of manufacturers consider data quality, contextualization, and validation significant barriers to AI implementation.
Not to end in the same pitfall, the desalination mega-plants should be designed with the AI strategy to gradually enhance the mega-plant "brain" - the control system.
The control system processes thousands of signals in a simple way - by using strict instructions. They don't provide much room for optimization of the plant performance (as well as for the alarm rationalization). Some AI-driven optimization tasks are described in the article.
To create the AI-based applications fine-tuning the control system, signals shall be converted into high-quality data to be used for AI model training. (Like a function, the AI model takes some input and produces some output.) It is a prerequisite # 1 for effective AI adoption.
What makes data high-quality is the contextualization and validation mentioned above. Contextualization deals with metadata of the data collected from the megaplant. Metadata is data about the data.
For instance, to correctly decode the pump vibration data stream, metadata shall include the vibration transmitter characteristics, the pump type, rotation speed, capacity, impeller design, and many others. In other words, metadata is the data source technical specification.
Modern control systems are not designed to store metadata. How to overcome this challenge entirely blocking AI proliferation?
The challenge does not exist for crenger.com as it stores all the metadata for all the plants in the same database.
Remarkably, any instrument or equipment piece is automatically assigned a global unique identifier (GUI). GUI is a key to obtaining metadata and the data source location. (Lack of GUI is a challenge #1 for all industrial IoT applications worldwide.)
The measurement values continually sent by the plant control system to data storage (in the cloud) have a simple structure.
{value: double, status: integer, timestamp: long, GUI: long}
It contains neither the value unit name nor the value description. Both are securely retrieved from the value metadata. GUI makes data meaningful. Besides, tearing metadata off the value substantially decreases the data storage volume (by over 90%). Importantly, it makes the process immune to the control system upgrade.
The image below explains how data is collected and retrieved by the AI model training application.
To retrieve metadata, the AI data consumer sends a request to the crenger.com API - Application Programming Interface. After authentication, the request returns a JSON response including metadata and possible errors. An example of the API request is given below.
https://www.crenger.com/api/v1/device?id=896
The data validation (DV) task is to screen good data from outliers. Normally, the DV rules are part of the SCADA program running on PLC. Most of them set the operational and controllable ranges of measurement (statically or dynamically).
The challenge of how to broadcast the DV rules from PLC to the AI data consumers does not exist for crenger.com as it auto-generates the SCADA program (together with the internal PLC address of a signal) from the same metadata.