When discussing machine learning in the engineering industry, people have a few misconceptions of the benefits or uses of machine learning or AI in this field. A lot of science fiction imagery is introduced by today’s movies, media, and expansion of technology. This area however, isn’t only concerned with self sufficient robots as we had thought. A lot of machine learning in this industry is concerned with efficiency in the decision making process, and making well informed decisions with the help of machine learning to make these decisions. The application of these machine learning systems can be beneficial, as they can allow you to input lots of data into the system, that could not only help in decision making, but also solution implementation. This can be done, when machine intelligence units are able to learn from the data they are given, and evaluate it through a series of algorithms that could eventually produce a model or framework that an engineer can use as a supplement to his knowledge, to make the most informed decision.
In order to demonstrate the potentiality of what machine learning can be capable of in engineering, one must understand the goal or function of said resource. The term Parento-optimality is a concept of efficiency, which can be defined as where “at least one individual becomes better off without anyone becoming worse off”(JIN). This concept of equilibrium is important in the machine learning decision making process, as there should not be disparities and inappropriate biases toward certain situations or applications if not beneficial to the current operations of the machine learning unit. In the article, they explain how in aerodynamic design they use machine learning when designing turbine blades. In this process, energy efficiency is stressed, and variables like pressure distribution and pressure loss should be minimized. Usually, computational fluid dynamics analysis is used for these calculations. However, machine learning techniques like meta modeling can also be employed. This process would then use parento-optimality to balance the pressure distribution and pressure loss variables, and make sure that the efficiencies in these areas are met.
In different situations, different criteria need to be met as well. With the onset of computer simulation, a lot of engineering analysis tools have been introduced and developed, as they can “accurately predict the consequences of design decisions”. This eliminates the need to spend money prototyping and drafting a concept as a first step, as the same testing can be done virtually by way of machine learning in some instances. The Adaptive and interactive modeling system ( AIMS ) was created to collect data in a simulation environment, then use this data in different examples, evaluate different parameters, create different models, then incorporate different learning and evaluation strategies to aid in autonomous decision making. Machine learning can also be used to generate different scenarios based on input given by an engineer. These can be different decision parameters that are inputted into a machine learning, decision making system, then converted into a model. AIMS provides accurate models, and various optimization algorithms to develop strategy to aid an engineer in their decision making process. AIMS in the example, is used in designing a diesel engine that outputs power within a certain kilowatt range, consuming the least amount of fuel. In this case AIMS is used to “support users in implementing their strategies” as they had used it in conjunction to a diesel engine simulator.
Machine learning can also be used to reduce the risks of failure. The consequences of these failures can vary, but can also be very devastating and lead to the loss of lives. Lessons from previous failures can be utilized when making future decisions. The availability of a machine learning decision making resource could potentially be used to save lives with it’s ability to retain a great deal of information, and use that information to provide some suggestion in what path might be best for an engineer to take with a certain problem. Developing a dataset that can encapsulate all of the data needed for decision making is necessary, including case histories from past failures. Certainty factors and other probability distributions must be acknowledging and implemented in order to make the most informed decision. The desired purpose of machine learning in civil engineering for this case, is as a “management tool for the control of structural safety” (Stone). It is to be used as an advice system rather than an expert system which makes decisions for you. Machine learning uses similarity between it’s data, and the current task or project at hand to make these recommendations.
The potentiality for machine learning in engineering is enormous and ever growing and evolving. Although the software must be perfected in order to apply machine learning to hardware, I believe that in the near future, a lot of things that we wouldn’t expect to be autonomous, will be autonomous. The increased intelligence that comes with machine learning can be beneficial in numerous amounts of ways, from helping save lives, to making informed and highly efficient business decisions, machine learning can be used to inform engineers of things that they would have been unaware of otherwise. This allows for more informed processes, and decision making, without prototyping or having to learn from failure in order to gain said information. The implications of machine learning for the future are enormous, and I hope that advancements in this area could eventually lead to a better world.
JIN, Y. (2017, January 22). Pareto-optimality is Everywhere: From Engineering Design, Machine Learning, to Biological Systems. Lecture presented at 3rd International Workshop on Genetic and Evolving Fuzzy Systems in Germany, Witten-Bommerholz.
Stone, J. R., Blockley, D. I., & Pilsworth, B. W. (2002, August 6). Managing risk in civil engineering by machine learning from failures – IEEE Xplore Document. Retrieved January 22, 2017, from http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=151259
Yerramareddy, S., Tcheng, D., Lu, S., & Assanis, D. (1992). Creating and using models for engineering design: a machine-learning approach. IEEE Expert, 7(3), 52-59. doi:10.1109/64.143239