Название: Synthetic dаta: Revolutionizing the Industrial Metaverse
Автор: Jimmy Nassif, Joe Tekli, Marc Kamradt
Издательство: Springer
Год: 2024
Страниц: 186
Язык: английский
Формат: pdf (true), epub
Размер: 47.2 MB

The book concentrates on the impact of digitalization and digital transformation technologies on the Industry 4.0 and smart factories, how the factory of tomorrow can be designed, built, and run virtually as a digital twin likeness of its real-world counterpart, before the physical structure is actually erected.

It highlights the main digitalization technologies that have stimulated the Industry 4.0, how these technologies work and integrate with each other, and how they are shaping the industry of the future.

It examines how multimedia data and digital images in particular are being leveraged to create fully virtualized worlds in the form of digital twin factories and fully virtualized industrial assets. It uses BMW Group’s latest SORDI dataset (Synthetic Object Recognition Dataset for Industry), i.e., the largest industrial images dataset to-date and its applications at BMW Group and Idealworks, as one of the main explanatory scenarios throughout the book.

For decades, computer scientists attempted and failed to develop computer vision algorithms that could reliably identify what is inside an image—telling us whether an image contains a cat or a dog. Most had given up hope that we would solve this problem in our lifetime. That was until the invention of AlexNet ignited the Big Bang of modern Artificial Intelligence. In 2012, Alex Krishevsky, Ilya Sutskever and Geoffrey Hinton had managed to take an idea that had its origin in the 1940s—Neural Networks—and apply it to the large amount of data available thanks to the Internet—the ImageNet database—with an extraordinary amount of compute readily available to any video gamer—NVIDIA’s GPUs. AlexNet smashed all previous records in the ImageNet challenge, and within a few years, neural network-based algorithms evolved to achieve superhuman abilities in classification and computer vision. The method by which we develop the most advanced algorithms and software had fundamentally changed forever. Up until that moment, developing advanced software simply required an intelligent human, a small computer, a text editor and a compiler. AlexNet showed us that algorithms that were out of reach to humans were now possible. We could now write software that can write software—algorithms we are incapable of writing directly. The catch is that we need large amounts of the right data combined with enormous amounts of compute. The admission price into the AI game is data and compute.

Creatures such as humans are born into the world without a true understanding of their new surroundings. Human babies learn how to see and perceive the world through life experience. Babies learn how to perceive shapes, depth, color, sound, scents and taste. They learn how to identify their parents and siblings using all of their senses over a period of time. They also learn the rules of our world—otherwise known as physics—by conducting specialized experiments. Babies test the world by throwing glasses and utensils off the dinner table, breaking their toys and spilling liquids. They do this repeatedly until they develop an intuitive understanding of the rules.

AIs learn in precisely the same way. We feed them life experience—another way of saying data—during the training process. We teach them how to see, how to perceive and how best to manipulate the world around them by giving them millions of experiences. Unfortunately, it’s impractical and in many cases unethical to have our AIs learn and gain these experiences in our world. We can’t afford to allow our self-driving cars or industrial robots to learn how to drive and operate heavy machinery in the real world. It will take too long for them to gain the experience they need on the job; and in the process, they can cause too much harm as student drivers and heavy machine operators.

The solution to this problem is simulation. If we can construct digital worlds that are indistinguishable from the real world—worlds that look, sound, feel and behave exactly like our real world—we can generate an unlimited amount of life experience for our AIs. The more compute we throw at the simulation, the more life experience we can generate in the same amount of wall-clock time in the real world. AIs are free to learn without any risk of harm inside these simulations. They can learn to drive cars in simulations where they experience children running into the middle of the street, millions of times in varied lighting and weather conditions, without any harm coming to children in the real world. The data we generate in these simulations come with perfect labeling—labels that are impossible to gather accurately from the real world.

It turns out that the computing technology that sparked modern AI were originally and primarily designed for simulating virtual worlds. This of course is the programmable GPUs initially developed for powering 3D computer graphics and rendering for interactive video games. Modern video games are in essence, simulations of fantastic virtual worlds. The most advanced video games approach the real world in complexity and physical accuracy.

There’s a beautiful duality to 3D computer graphics and computer vision. 3D computer graphics is a function that transforms a structured description of a 3D world into images over time—a simulation of what a camera sensor would experience in that world. Computer vision is the inverse of this function; transforming images over time into a structured representation of the 3D world. The AIs that will be the most impactful and valuable to human-kind will be the ones that can understand our real world and operate within it. To create these AIs, we must first model the real world and simulate it to generate the life experience for them. Once our AIs achieve proficiency in understanding and manipulating worlds, they will then assist us in designing efficient, sustainable and delightful virtual worlds—worlds that will act as the blueprints for what we choose to build in our real world.

This book examines how multimedia data and digital images in particular are inputs into the creation of fully virtualized worlds in the form of digital twin factories and fully digitalized industrial assets. It relies on practical use cases from the automotive and manufacturing industries and their digitalization technologies based on the SORDI dataset. With this book, you will understand the nature of data and its unique value for AI. You will learn how to capture, structure and generate data essential for AI in order to build the industrial metaverse.

Contents: