Glossary

Liveline Technologies employs advanced concepts from controls engineering and artificial intelligence. Terminology from one of those fields may not always be clear to practitioners in the other. Here we define some common terms you may encounter when working with our products.

Common terms

Advanced Process Control

APC

An industry term for high-level “supervisory” controls over an entire production process. APC solutions control multiple pieces of equipment and attempt to optimize overall production outcomes while respecting various constraints. The term is often used interchangeably with Model Predictive Control.

artificial intelligence

AI

We feel a bit guilty using this term, because many people assume “AI” refers to a artificial general intelligence, or AGI, which does not yet exist, and may not ever be possible to create. A true AGI would be able to understand or learn any task that a human can perform, and could possibly attain self-awareness or consciousness. Liveline Technologies employs advanced machine learning techniques that are often described as AI in casual conversation.

controller performance

Controls engineers usually mean something very specific when they talk about evaluating the performance of a controller. There is a well-developed body of theoretical and applied knowledge around this topic.

Performance may consider the time response of the controlled system as measured by rise time, peak overshoot, settling time, etc. It may also consider robustness in the face of uncertain or noisy measurements. Such metrics are common when implementing, for example, a classic PID controller.

In the case of controllers based on trained machine learning models, we may be interested in additional performance metrics, such as level of activity (how frequently the controllers want to change equipment parameters) and drift of controller performance over various timescales.

We evaluate controller performance using a Liveline Test Harness that is configured according to your performance requirements.

If This Then That

ITTT

A basic type of control scheme consisting of simple triggers and actions. Often used for automation of tasks in small IOT and web-based environments. Example: If temperature exceeds 28C, then turn on the air conditioner.

Liveline Controls Package

LCP

An encrypted software object that contains a trained model to control equipment in a factory system. Typically we train one LCP for each combination of SKU and line or cell within your plant. Training is highly automated, and LCPs are “refreshed” as conditions change within the plant.

Liveline Data Quality Score

LDQS

A proprietary measure of how useful a particular chunk of factory data will be for training new models. DQS is continuously calculated while your plant runs normal production. We automatically select the most useful data for training our models.

Liveline Digital Controls Platform

LDCP

This is the basic digitization platform that allows us to implement factory controls. The technology stack is proprietary but built entirely on open-source technology.

Liveline Physics Package

LPP

An encrypted software object that contains a trained model of factory physics, along with the feature engineering pipeline necessary to process raw data in batch or online modes. Typically we train one LPP for each combination of SKU and line or cell within your plant. Training is highly automated, and LPPs are “refreshed” as conditions change within the plant.

Liveline Test Harness

LTH

A standardized suite of metrics to assess controller performance. Metrics will be adjusted based on the requirements of a particular production environment. The metrics may include typical items from classical controls theory, along with items that are unique to implementing controls via machine learning models.

machine learning

ML

Unlike AGI, machine learning (ML) is very real. ML uses statistical models to learn from data, rather than being explicitly programmed to perform a specific task. Models can be extremely powerful, and their ability to learn subtle thing can sometimes feel like magic. In reality, it’s just a form of very advanced statistics.

Liveline Technologies makes extensive use of two of the three major fields within ML: supervised learning and reinforcement learning. The third field is unsupervised learning, which is less commonly used within the Liveline technology stack.

Manufacturing Execution System

MES

A computerized system used in manufacturing to track and document the transformation of raw materials to finished goods. MES captures real-time information from the operations to enable reporting, documentation, and optimization.

MES typically operate across multiple functional areas, such as resource scheduling, order execution and dispatch, production analysis and downtime management for overall equipment effectiveness (OEE), product quality, or materials track and trace.

MES systems sit “higher” in the ANSI/ISA-95 definition of industrial computing and automation levels. Specifically, MES is Level 3, while Liveline’s solutions are Level 2. Liveline’s solutions can work with any MES system.

Model Predictive Control

MPC

An industry term that is often used interchangeably with Advanced Process Control. In MPC, some sort of process model is used to predict the future evolution of the plant. Based on these predictions, actions can be chosen to optimize future outcomes while respecting various constraints. Note that the way we perform optimization is fundamentally different from traditional MPC techniques. Also, our software automatically generates and updates models for even the most complex processes, so users don’t need any capabilities for modeling.

PID

A proportional–integral–derivative (PID) controller is a control loop mechanism employing feedback that is widely used in industrial control systems. A PID controller continuously calculates an error value as the difference between a desired setpoint and a measured process variable and applies a correction based on proportional, integral, and derivative terms (denoted P, I, and D respectively). See Wikipedia for more details.

Programmable Logic Controller

PLC

An industrial computer that has been ruggedized and adapted for the control of manufacturing processes. Typically contains a large number of signal input and output (IO) points. Modern PLCs are usually networkable via Ethernet and can communicate with equipment using standard protocols such as OPC/UA.

reinforcement learning

RL

An area of machine learning concerned with how intelligent agents ought to take actions in an environment in order to maximize some notion of cumulative performance.

At Liveline Technologies, our controllers are actually RL-based agents that are trained to “drive” a factory system. As part of our automated machine learning process, we create a model of your factory system, and our RL agents learn how to manipulate controllable variables to achieve desirable outcomes while avoiding negative things like exceeding maximum settings on machines.

Typically, we create a factory model and controller agent for each combination of SKU and line or cell within a plant.

Factory models are encrypted and stored in Liveline Physics Packages (LPP), and controller are encrypted and stored in Liveline Controls Packages (LCP).

A sonnet

We asked ChatGPT to define machine learning in the form of a Shakespearean sonnet. This it what it returned:

Oh noble art of Machine Learning, so grand,

A gift of modern science to the hand,

Of engineers and coders, who impart

Its mysteries to machines, that they might stand

In excellence before all of mankind,

And do what once was thought beyond their might.

With algorithms and models, they imbue

The power to learn, to make their own right

Decisions, with no need for human hand.

Thus, they evolve, with accuracy in sight,

And bring new solutions to our world’s plight.

Oh, how they sift through mountains of data,

With grace and speed, to find what we need.

Thus, do they make our lives so much brighter.

And so, I sing the praises of this breed,

Of machines that learn, that we might succeed.

— ChatGPT, 9 February 2023