BACKGROUND

In contemporary industrial applications, expert operators still have a key role to play operating machines, evaluating a controller’s performance, performing system diagnosis, tuning controllers, etc. Examples are operators

• that tune controllers during commissioning based on on-site trial-and-error experiments to see how the system responds to different settings;
• that adapt heuristic or sometimes model-based setpoints during operation,
• that evaluate (subjectively) system performance;
• that determine root causes and change design parameters that caused mediocre performance, etc. Some humans excel in this and their expertise contributes to company IP as much as other key technologies and innovations do.

PROJECT TARGET

This project aims to capture and digitize operator expertise and knowhow. When successful the project will allow to fully support the (re)tuning procedure or assist junior operators so that the supported tuning cycle is equivalent or superior to that of an expert operator.

PROJECT DETAILS

We want to support expert as well as non-expert operators to ensure tuning or adjusting is done a lot quicker, reducing lead times and shortening design cycles and testing times, but also more consistently than operators can currently do. We moreover want to ensure this is done efficiently and in a comfortable and intuitive manner, for which we want to extract and use the knowhow that expert operators exhibit, yet without them having to mathematically write their expertise down. We will develop methods to automatically capture and digitize operator expertise and knowhow. To achieve these goals the project will develop methods that analyse data of the operator interacting with the machine or process to capture the knowhow the operator specifies implicitly (since in this mode the operator interacts with the machine/process as they would normally do), as well as methods to explicitly let the operator provide information but in an intuitive and comfortable manner, such as accepting or rejecting suggestions for settings to try, ranking suggestions or evaluating the resulting performance (this is explicitly provided information, but the expert does not need to write down their entire reasoning mathematically). We aim to do the above based on historical datasets using data mining techniques and based on interactive tuning procedures wherein multiple experiments are performed which have been chosen during the procedure itself, first to better understand the system’s response and second to fine tune and optimize the performance. All of this in a manner that allows cooperation with an operator.