Artificial Intelligence is the ability for a machine to learn and think like a human either for a specific task or in general. Though the concept of AI has existed since the 1950s, recent advancements in simulation, high-performance computing, sensors, and the Internet of Things have brought AI for engineering from the realm of possible to reality. The brain of AI comes from recent developments in machine learning. Machine learning enables a program to expand its understanding of a topic beyond what it is strictly programmed to do.
Today, variations of AI can be found in almost anything from diagnosing problems in manufacturing and developing autonomous systems to predicting the next catastrophic weather event.
Performing analytics with AI to address engineering problems faces several formidable challenges across the whole product life cycle.
A design engineer can train an AI system to assist in finding optimal designs. For example, the AI learns from the simulation models that a design engineer provides. However, in many cases engineering models are deterministic, whereas the underlying problems they are modeling are subject to uncertainties and therefore stochastic in nature. If the engineer only provides the AI with rules in the form of models that present a deterministic world view, the AI will be constrained to this as the nature of reality. The AI may produce an optimal solution, but one that corresponds to the unrealistic scenario where uncertainty does not exist, rather than the desired solution incorporating real world uncertainty. To achieve the true aim, the AI must be taught the stochastic nature of the outcomes of interest by incorporating uncertainty into its decision rules.
Modern manufacturing lines are equipped with a large variety of sensors and an ability to gather high volume and high velocity data on all aspects of the manufacturing process from equipment health to product quality. Sensor data will always have some level of imprecision or noise associated with it; the combined uncertainty due to so much data from so many disparate sources can make it difficult for AI algorithms to identify signals of interest amongst the noise.
Maintenance and servicing can present the opposite challenge of manufacturing, a lack of data. This can be an issue when the goal is to train an AI system to predict patterns or anomalies. Predicting anomalous or rare events can be of high importance if the consequences of the event are great. However, because of their rare nature, too little data can be a hinderance to the AI system learning to predict such events, i.e. as for humans it is more difficult to predict events you have never experienced than those you have.
In general, AI is dependent on the quality of the data it receives. If the data contain inaccuracies, they will be reflected in the AI results. As AI and machine learning algorithms are complex systems and blackbox functions themselves, the application of Verification, Validation, and Uncertainty Quantification (VVUQ) is important to ensuring the results are as accurate to reality as possible. For example, engineering data will follow physics-based relationships which may include invariants such as momentum or the total energy of an isolated system. Some outputs of a system could also have known monotonic relationships to one or more inputs. It is important to perform VVUQ on AI systems to ensure these significant relationships are preserved.
For many engineering applications, analysis and performance checks done at the system level are critical. This involves combining information from lots of subsystems each with potentially many inputs. For example, the system level analysis of a vehicle’s health requires data on subsystems including the braking system, engine, transmission, and chassis. Being able to account for the total number of parameters across all these subsystems and how they combine to impact various outcomes of interest like vehicle safety and comfort is a large-scale and difficult problem.
SmartUQ addresses these challenges using a variety of analytics tools:
Analyzing AI data successfully can be challenging, but with the appropriate tools and support, engineering and other industries can find new ways to automate tasks and better understand complex phenomena.
To learn more about analytics for AI applications, check out SmartUQ white papers and webinars.
