The industrial revolution of the 21st century is the information revolution. The increase in the computing capacity of computers has given us the possibility of extracting highly useful knowledge to improve the efficiency of industry, service companies, and administrations.
In this context, supercomputing, due to the great capacity of its machines, but also due to the way they are programmed, becomes a key instrument for those countries, companies, and entities that want to be among the most developed and advanced. The information industrial revolution consists of computing capacity and algorithms.
SIMULATIONS
The most classic use of supercomputing for industry is the possibility of conducting experiments with the supercomputer instead of with physical means. This leads to enormous cost savings and acceleration in the product design process: experiments are faster and cheaper. And currently, remarkably precise.
What does it mean to be able to reproduce an experiment on a supercomputer? It means being able to simulate the physics and chemistry of natural phenomena on the computer, solving the mathematical models, the equations, that model these phenomena.
With the current computing capacity of computers, there are phenomena that twenty years ago could not be simulated with sufficient precision and today they can. For example, combustions. In combustion simulation, total precision is not yet achieved, but the level of precision is already high enough to be very useful in the design of different combustion devices, from a car engine to a gas turbine or an industrial combustor. Second example, the interactions between any type of mechanical device with fluids: we can simulate in detail the operation of a wind generator or a hydraulic pump. And third example, hydrocarbon exploration: currently we have a computing capacity that allows us to obtain images, and not only images, but to know the physical properties of the terrain with a precision that was impossible before petaflop machines. Petaflop machines are those capable of performing one thousand trillion operations per second (1015). And with exaflop machines (1018), planned for the beginning of the next decade, we will go even further.
PARALLEL SOFTWARE
At the Barcelona Supercomputing Center – National Supercomputing Center (BSC) we have the MareNostrum supercomputer, which ranks number 16 on the Top500 list of the fastest supercomputers in the world. MareNostrum 4 has a computing capacity of 13.7 petaflops, a main memory of 390 terabytes, and 14 terabytes of hard disk storage.
But, when talking about supercomputing for industry, it is important to think that it is not always necessary to have the fastest machine in the world. Supercomputing actually means parallel computing. There are industrial problems that need more modest computers than MareNostrum, but still parallel computers. Therefore, an important thing is to develop software capable of using these parallel machines efficiently. And one of the main contributions of the BSC to industry is to help develop this type of parallel software.
Today we can build gigantic neural networks that are already useful for addressing real problems. The procedure is to collect the problem's data and use it to train a neural network. This neural network will make predictions about the behaviors of that system and will act as an equation-based simulator.NEURAL NETWORKS
Sometimes there are industrial problems that cannot be simulated in a standard way, that is, by solving a series of equations that model the physics. This occurs for different reasons. Normally it is due to uncertainties, because there is something we do not know that prevents us from solving the equation. For example, in a waste incinerator we do not know what fuel we are burning, because we do not know the composition of the trash in detail. When these types of situations occur, we can model the phenomenon using simply data, without the need to use equations. Today, electronic development has reached a point where we can build gigantic neural networks that are already useful for addressing real problems.
The procedure is to collect the problem data and use it to train a neural network. This neural network will make predictions about the behaviors of that system and will act as an equation-based simulator. Neural networks have been known since the 60s, but twenty years ago obtaining useful results for real problems with them was not possible, simply because the circuits available did not allow creating networks of a sufficiently large size. Now, with the advancement of electronics, it is possible.
NATURAL LANGUAGE PROCESSING
Another new application of supercomputing is in the field of social sciences, for example, in the field of natural language processing, the language that humans use. Through new statistical algorithms, which use the great computing capacity of machines, we can process texts automatically to, for example, extract or find information from a series of databases or documents in which, without these tools, we could do nothing more than search for keywords. Who is this useful for? Well, it is useful, for example, for the processing of legal information, for the processing of all clinical information from hospitals, or for the processing of information from administrations, which currently have multiple dispersed databases and documents with different formats.
BIG DATA
In the field of social sciences, when we analyze large amounts of data, Big Data problems also appear. For example, when we face the management of citizen complaints in a large city or want to analyze different information from social networks, the volume of data is so large that we need supercomputers and special algorithms to process this information and extract useful conclusions.
In conclusion, computing capacity and algorithms can be key tools to improve the management of all types of processes that seek optimization and efficiency. In the 21st century, having them as allies will be key for all those companies, administrations, and entities that wish to play a prominent role in their respective areas.