Digital Transformation

Neural network models, also known as artificial neural networks (ANNs or just NNs), are a type of machine learning model that mimics the neurons and synapses of organic brains. By training an NN on historic patterns and outcomes (supervised learning), NNs can learn to match new data to trained patterns and classify them or predict future events.  Several classes of industrial AI software use neural network programming including predictive quality, predictive equipment reliability, and machine vision applications.  NNs can also be trained via unsupervised learning in some cases where the application automatically compares predicted outcomes to observed outcomes and self-corrects.  Many industrial NNs are only three layers deep and a dozen or so neurons in size, far simpler than applications like ChatGPT or even an insect brain.  Despite appearances, most industrial NNs do not understand the process they are examining.  They are only looking at statistical correlations between input patterns and output patterns.

Multi-Variate Analysis applications are a wide range of software applications that use statistical methods to analyze multiple input parameters at once. They are particularly well suited to the “big data” approaches often used in Industry 4.0 concepts where large volumes of data are fed into applications that will decide which parameters are important and which can be ignored to achieve a desired outcome.  Many statistical tools fall into this category including neural networks, and Monte Carlo simulations.

Digitalization refers to the generation and use of data in a digital native form, often from multiple sources. It is a core component of the Industry 4.0 framework.

Digitization is the conversion of data into digital form. This could be the scanning of paper documents, manual data entry of log sheet recordings, etc.  It is different from digitalization which involves native digital data processing.  Digitization unlocks value from paper records but can be time consuming and error prone to convert into digital data.

A digital twin is a digital model of a physical asset or process. The ultimate goal in industry would be to have a digital replica of an entire plant that represents all of the installed equipment and is capable of reproducing all of the behaviors of that plant.  This would allow for experimentation and optimization of the digital twin simulation that would provide insights into possible improvements to the physical plant.  In today’s reality, digital twins are more focused and specialized, with multiple digital twins replicating certain features and processes.  Engineering digital twins attempt to inventory and organize all the assets in a plant and share that data among all disciplines; however, they rarely have the ability to replicate process behavior.  Process digital twins are built to replicate the chemical and physical behaviors of plants to determine design details but are simplified representations of the equipment.  Other digital twins may include 3D design, finite analysis, CFD (Computational Fluid Dynamics), or multi-variate predictive models for individual assets.

Industry 4.0 refers to the fourth Industrial Revolution, sometimes abbreviate 4IR. The term is commonly attributed to Klaus Schwab, the founder of the World Economic Forum around 2015.  Industry 4.0 is an umbrella term referring to many software and hardware technologies but are generally characterized by increased data connectivity between systems, the application of AI/ML, cloud computing, IoT, 5G communication, 3D printing, augmented reality, etc.  Other associated terms are Smart Manufacturing, Smart Industry, Smart Factory, and Factory of the Future.

In chronological order, the generally recognized industrial revolutions were:

1^st Industrial Revolution / Industry 1.0 – 1780s: Mechanization, water and steam power, interchangeable parts

2^nd Industrial Revolution / Industry 2.0 – 1870s: Electrification, decoupling of factories from power source, assembly lines, motors, relays

3^rd Industrial Revolution / Industry 3.0 – 1970s: Automation, electronics, computerization, semiconductors

4th Industrial Revolution / Industry 4.0 – 2010s: Artificial Intelligence, self-learning systems, bridging of physical and digital worlds

The process historian is usually at the center of any Industry 4.0 initiative and shouldn’t be taken for granted. An industrial process historian, sometimes called a PIMS (Process Information Management System) is a specialized database designed for the efficient collection and storage of time series data.  General purpose databases like SQL and Oracle are designed to store many data formats, often in many interconnected tables.  Process historians are designed to store relatively simple data formats, but often use compression techniques to efficiently store the data.  How a process historian collects the data is also important.  Process historians are designed to communicate using numerous standard and proprietary protocols to collect data from both common and obscure PLC and DCS platforms.  They often also offer redundant and/or buffered communication options.  It is not uncommon for process historian packages to come bundled with a more traditional database application like SQL to store other plant and equipment information outside of process data, further demonstrating that the two types of databases are not interchangeable.  Lastly, process historians nearly always come bundled with a variety of trend viewing and analysis packages.  For many clients, these trend viewing packages are the face of the product and can substantially influence buying decisions.

Usually not, but possibly. It really depends on what Industry 4.0 technologies are being deployed.  Many technologies are already fully productized and require little to no “programming”, just configuration work in the software package.  Some platforms like AWS are more component based, and some coding is required to move data from one component to another and do basic data manipulations.  Some companies decide they need custom web applications to manage and view their data.  While there are some low code/no code products available in this area, this is where most Industry 4.0 software development has been done.  Programming languages common to web applications like HTML/XML, Python, and JavaScript are common here.

AI refers to Artificial Intelligence, mimicking the natural intelligence of humans or animals in computerized systems. ML refers to machine learning, mimicking the way organic brains learn from their surroundings in a computer.  ML is a subset of AI, but since many AI software applications utilize ML, the terms are sometimes used in combination or interchangeably (despite the distinction).  ML algorithms often use neural network programming to statistically match input patterns with outcomes to either classify outputs or predict future events without the need for complex application specific programming using traditional If/Then logic that requires the programmer to think of all scenarios that may be encountered in the future.