Intelligent factories have existed since manufacturing’s historical inception, but intelligence—defined as the acquisition and application of manufacturing knowledge—resided only with the factory’s staff.
With the advent of numerical control (NC) and then computer numerical control (CNC) technologies, factory machines gained digital I/O capabilities but were still not smart. Digitally enabled machines, though increasingly productive, had no awareness of themselves, their environment, or the tasks being performed or to-be performed.
In spite of these limitations, centralized factory intelligence has been achieved at modest scales through a deterministic low-level set of digital commands and responses. An experiment in large-scale centralized factory intelligence was General Motor’s 1982 Manufacturing Automation Protocol (MAP), operating over token bus network protocol (IEE 802.4). The MAP-enabled factory intelligence experiment ended in 2004 as it was difficult to maintain operational reliability.
One of the most important reasons was a lack of system resiliency, a downside of required deterministic factory communication standards and protocols. Another reason was that the connected machines could not continue to operate at any level when instructions were not forthcoming from a central system.
An analogy might be made to the mainframe-to-terminal infrastructure that became obsolete in the 1990s with the development of the PC and distributed computing. Several significant changes have enabled the development of smart machines for the intelligent factory. The first is the extension of IT’s ubiquitous Ethernet LAN infrastructure to the shop floor, enabling rapid 3D downloads of model-based definition (MBD), and uploads of process and product data.
Secondly, today’s digital twins are smart in that they possess an awareness of not only their capabilities and operational status, but of work that can be performed on any particular MBD. In this manner, smart machines can bid on tasks, much like their human partners. A smart machine’s digital twin does not need deterministic low-level instructions, but instead responds to a submitted MBD, and, if selected, does real work with its physical counterpart.
STEP carries most, if not all, of design intent as a 3D semantic model in a non-proprietary format derived from a CAD system’s native master model. MTConnect enables computer-aided manufacturing equipment to consume and produce structured, contextualized process data using a non-proprietary format.
QIF enables computer-aided measuring equipment to consume and produce structured, contextualized data using a similar non-proprietary format. All three standards and their initial implementations are progressing in the metrics of manufacturing readiness.