In our first post —Breaking Down the Concept of Artificial Intelligence and Dispelling False Myths!— we defined what Artificial Intelligence (AI) is and clarified other terms that tend to generate confusion. To complement it, in this post we will delve a little deeper into the definition of AI, and its connection with the optimisation processes. We will also describe the different types of optimisation, as well as their advantages and disadvantages.
So, What is an Optimisation Process?
It is a mathematical procedure that focuses on finding the best result according to some criteria and conditions. In addition, this procedure combined with AI techniques can achieve optimal results in a more efficient way, by exploring only a small percentage of all possible solutions.
There are several areas of application, ranging from the resolution of complex engineering problems, to AI training, and even economic and transport problems, among others.
Different Types of Optimisation
In order to obtain an optimised result, multi-objective optimisation methods have been studied and designed over time. These methods can be divided into two categories depending on whether they guarantee the best result or “global optimum” (exact method), or not (approximate method).
Although it is always expected to obtain the global optimum, a mathematical definition of the problem is required to apply the exact methods. This definition becomes complicated as the complexity degree of the problem to be optimised increases, it is therefore necessary to use methods that do not guarantee the global optimum.
Among the approximate methods, which do not ensure the global optimum, there are two subcategories—heuristics and metaheuristics. Even though both are focused on trying to get the best results according to a specific problem and its conditions, there is a key difference.
What is that Key Difference?
The techniques used for metaheuristic optimisation are characterised for being more complex and refined, involving therefore a greater diversity of problems, which implies a performance penalty. This loss of performance is the result of not approaching the problem particularities, but as a main advantage, this type of optimisation is ready to be implemented.
Heuristic optimisation, on the other hand, is an ad-hoc designed process for a specific type of problem, allowing high performance. However, implementing these particularities in the problem will lead to cost increases due to the time required to carry out a customisation.
So, What is Better?
There are no better or worse methods, it will depend on the problem and the objectives that we want to achieve. If the results obtained through the metaheuristic optimisation process are “sufficiently” good, there is no need to use a heuristic process. Otherwise, it would be necessary to include the particularities of the problem, thus becoming a heuristic optimisation process.
Moreover, it is worth mentioning that if a problem is sufficiently complex and costly to simulate, it will entail a number of constraints, which in turn may make the process more time-consuming or even unworkable. In order to overcome or mitigate these inconveniences, surrogate algorithms come into play. Using AI techniques, they are responsible for replacing the costly simulation with an approximation of it, which in this case is instantaneous.
At Lurtis, in order to determine if this ad hoc adaptation of the problem is worth it, we recommend to our customers to analyse how far they want to go and what the benefits and costs of this implementation will be.
Despite the fact that Artificial Intelligence has been researched extensively for decades, it had not been driven in the business world until recent years. This has been possible thanks to computing advances, breaking down the barriers that were holding back the implementation of AI among companies.
At Lurtis’ R&D department, we want companies to be able to implement this technology quickly and easily. To do this, we are developing Lurtis EoE (Engineering Optimisation Engine), an optimisation platform that integrates an intelligent combination mechanism that allows companies to integrate multiple heuristics and metaheuristics methods, as well as the option to implement surrogate algorithms in their processes, – we will explain more about this topic in one of our next posts.
Here we mention some of the projects that are currently being developed at Lurtis which integrate optimisation processes:
Metamaterials Research Project
This project is about the automated design of metamaterials through the optimisation of its structure by using metaheuristic methods and surrogated algorithms to perform it.
The aim of this research is to be able to model the behaviour of materials to certain stimuli of the force or displacement type. To do this, the finite element simulation Abaqus is interconnected with the Lurtis optimisation library. As a result, there was an improvement in the response of the metamaterial to a 30% stimulus in relation to the state of the art proposal.
At the AEC sector development department, different applications are being developed aiming to offer solutions in the preliminary stages of building design. One of these applications is focused on reducing the environmental impact with a view to optimise building envelope materials in relation to several purposes, such as energy consumption, carbon footprint and material costs. To achieve this, the evaluation of energy simulation and the calculation of carbon footprint, among others, are used as a function of quality to drive a heuristic search for optimisation, which applies surrogated models based on Machine Learning.
At this point we conclude the introduction of the optimisation concept. In one of our next posts we will explain more about the surrogate algorithms, as well as their advantages, disadvantages and the projects in which Lurtis applies them.
Pablo Sánchez Naharro – Lead Researcher & Project Manager