Automation and Control: Implementation of Artificial Intelligence for Advanced Production Control Systems - DIAL Newsletter Summer 2019

DIAL has investigated the use of Artificial Intelligence for advanced production control systems in many of its industrial automation and control projects over the last 20 years. These control systems have enabled enhanced manufacturing capabilities such as customised manufacturing, product traceability and improved production resilience when catering for production equipment failures and/or product quality issues. Core to the implementation of these control systems has been the use of holonic or agent based architectures. The group has not only investigated the algorithmic side of the AI systems, but also explored different interfacing mechanisms that will allow these systems to be deployed in a real environment and operate reliably. 

 

The challenge in developing these enhanced control systems is being able to provide effective links between robust deterministic control system components run on a Programmable Logic Controller (PLC) at the production line; and non-deterministic high-level software engines running in programming environments such as JAVA. Depending on the remit of the control system being developed and the source of data being used, it can significantly affect the design and communication technologies used within the system.

 

As part of the Auto-ID project for Gillette in 2003-8, the group developed a consumer goods packing line. The line was implemented using JACK intelligent agents by the Agent Oriented Software Group. The production  line interfaced to web resources for customer orders and RFID product tracking information using SOAP messaging. The agent environment communicated to Omron production PLC’s using a visual basic interface. Agents were used to represent both orders and production resources and Believe-Desire-Intention (BDI) models were used to deliver intelligent behaviours for resource scheduling and resource failures.

 

As part of a resilience project for Boeing in 2014-16, the group developed a resilient gearbox production line. The line was implemented using Java Agent Development Environment (JADE) framework. The agent environment communicated to Omron production PLC’s using an OPC UA server. Agents were used to represent parts, products, orders, order manager, inventory manager and production cell managers. In this implementation, the BDI models were used to discover resilience strategies for re-ordering of parts, managing the location and size of part buffers to minimise order completion time.

 

From the above implementations, a number of considerations were critical:

    a) what data is required to allow learning or control system decisions to be made?

    b) Is the timeframe of the data, learning and decision suitable for the application?

    c) What protocols or interfaces will be available and suitable?

    d) Would this data / interface put my production system at a security risk?

DIAL’s automation and control team continues to explore these questions.

Date published

25 July 2019

For further information please contact:

Alan Thorne

E: ajt28@cam.ac.uk

 
Share This