The engineering use cases of AI are widespread but often misunderstood. Companies misjudge the capabilities of AI and because of this, fail to see suitable applications of it in their work. The reality is that AI is one of today’s fastest growing technologies and can be used by engineers to build and analyse data without the need for data science expertise. For engineers, AI is game-changing, providing a low-cost means to stay competitive by cutting time-to-market, improving quality and efficiency, making faster decisions, and upskilling highly valuable R&D talent so they can work more independently and effectively.
Artificial intelligence (AI) is quite a hot topic in the innovation space. Gartner predicts a whopping 21.3% growth in the AI market in 2022 compared to last year. The global AI software market is predicted to reach a total value of USD 22.6 billion in the coming years (Thormundsson, 2022). With the rise of AI, self-learning models are now gaining some well-deserved attention, particularly in the automotive, aerospace, and general industrial realms.
The distinction between the two is important. AI systems are able to perform tasks that would usually require human intervention such as speech recognition, visual perception or decision-making. Self-learning models are able to recognize patterns in large datasets and learn from the data trends over time.
By doing this, these models can draw meaningful conclusions from data when presented with different parameters or rare scenarios in engineering use cases. Self-learning AI can help design higher quality products in a fraction of the time, retain knowledge for future generations of designs, and involve no iterative guessing game resulting in higher ROI, better quality products, and product performance in half the time.
Using self-learning models can help handle repetitive jobs and predict trends instantaneously. By using this approach, the engineers of the BMW Group were able to use the wealth of their existing crash data using the Monolith AI platform to accurately predict the force on the tibia for a range of different crash types.
This gave highly accurate results and completely alleviated the need for very costly, time-consuming physical tests. This is an exciting prospect and companies have shown interest in implementing self-learning models into their organizations as quickly as possible to not lose touch with the digitalisation of the industry and stay ahead of their competitors.
Implementing AI where it is not needed, however, can cause companies to wander into some rather perilous territory. AI tech is frequently used incorrectly due to biases in data and teams misunderstanding the very nature of self-learning models.
It is often also seen as a means to replace skilled engineers and data scientists within a company, when it should rather be used by engineers as a tool to enhance and accelerate decision-making. After all, no one knows the data better than your engineers.
Another problem is that AI is being used by companies who have no business using it, and where its application does not have a solid business use case within the company. Horror stories from these cases cause other companies who would benefit from AI not to adopt it.
An article in the Harvard Business Review said that some companies are waiting for technology to mature and for expertise in AI to become more widely available before adopting it, which may lead to shortfalls and lack of a competitive advantage in implementing AI into their company systems.
Forrester (2016) summarised the reasons companies are skeptical to adopt AI in the graph below.
There are a few things engineering companies can do to increase the chances of success of AI implementation. This can help ensure that a company stays competitive by innovating and by still ensuring a positive ROI for the organisation.
It’s tempting for companies to adopt technology in ‘hot’ industries like AI or machine learning (ML) for the sake of being on the popularity bandwagon but are unaware of whether self-learning AI will directly lead to a positive ROI. Some of these organisations fall victim to overusing AI and train models in every sphere possible that will ultimately not benefit them.
This could lead to processes that work well overall, but that are of no value to the company. It is all well adopting AI to predict results, but not if those results are not useful or if the analytical model that AI replaced did not need improvement in the first place.
Companies should consider what parts of their operations are problematic or that are bottlenecks in their processes and think carefully about how AI could alleviate that pain. Managers should ask themselves if using AI could meaningfully increase efficiency or reduce cost in these areas. If not, they run the risk of unnecessary expense and loss of productivity.
For example, if physics-based methods such as Computational Fluid Dynamics (CFD) or Finite Element Analyses (FEA) can be run in a computationally light way, AI may not be needed to replace this process. However, in most cases where highly nonlinear, intractable physics is involved, simulations are a very time-consuming and costly road to pursue.
"For the development of a new gas metre, CFD models were not accurate enough to capture the complexity of the flow for varying temperature conditions and types of gases. Using Monolith, we were able to import our rich test stand data and apply machine learning models to conclude on different development options much faster."
- Dr Bas Kastelein
On the flip side, self-learning models are ideal when simulations are too inaccurate or too slow, and when physical testing is expensive and time-consuming. We have seen the effects of having sufficient data points but slow simulations in our work with BMW, for example, who have since benefited from using self-learning models in their design process.
Areas such as aerodynamics or material testing are ideal for AI models trained on test data to rapidly accelerate highly iterative design processes. Vehicle crash tests, smart meters, chemical manufacturing, and packaging are examples of this.
It is crucial for engineers to understand what AI can do and how to leverage its power. In our experience, people do not have a working knowledge of what machine learning is and, therefore, ask for unachievable outcomes. This wastes time and causes frustration.
AI is powerful, but it is not a magical all-encompassing solution to a company’s problems. Using the term “machine learning” reminds us that the AI algorithms must learn before they can think. Understanding the basics of AI or, at least, its basic concepts can make a huge difference.
A single afternoon workshop could be enough to fend off common fallacies and give insights into the sorts of problems that AI can help solve. Looking at case studies from other departments or companies can also highlight the realistic applications of AI in your industry.
Data is the lifeblood of AI algorithms. Many companies don’t know this, and often don’t understand their own data enough to know if AI will be useful. The accuracy of an AI model is proportional to the level of quality of the data and on how much data is available. The more complex the problem, the more data is needed. Some prospects have asked Monolith to predict results with an unrealistic accuracy given the amount of data available. The result: inaccurate predictions because of impossible accuracy targets.
The solution to this is to spend time understanding the data that is available. How many data points are available? In some cases, four points may be sufficient to give accurate results. In other cases, 4000 points may not be enough. Is the data highly complex or more categorically black-and-white? For example, determining if a component is manufacturable or not given certain parameters. Understanding the complexity of the problem is useful.
This can be measured according to the number of parameters needed or by the complexity of the relationship between inputs and outputs. Is it as simple as predicting the downforce on a vehicle given different chassis heights and body geometries? Or as difficult as using data from thousands of crash tests to predict the force on the tibia bone during multiple crash scenarios? This will give an understanding of the amount of data required to get accurate and reliable results from AI models.
In the end, AI cannot process what is not given to it, and it is not guaranteed to improve every problem that is thrown at it, but it can be an amazing tool to help predict trends in data and reach meaningful solutions - provided engineers understand the nature of their data and what is being fed into the model. No one knows your engineering data better than your engineers.
Instead of hiring more Python coders or data scientists and waiting months for meaningful insights, companies start to reap the benefits using Monolith by magnifying engineers’ expertise and the wealth of test data to develop better quality products in half the time.
Companies who are successful in their use of AI, and there are many, have managed to reduce their time-to-market, increase their operational efficiency, and have joined the ranks of some of the most technologically emergent companies on the planet.
To do this successfully, these companies understood that AI and ML are not the be-all-and-end-all solutions to their problems, what AI is, and where its capabilities lie in the context of their own problems. And by using Monolith’s self-learning model capabilities, they get better insights into the data that is available to them and how different algorithms can be used to build more performant and reliable products in a fraction of the time, knowing that AI can be a great solution for intractable physics problems.
The most valuable projects arise when your team of engineers understands their main pain points and the data they have gathered over the years. Educate them, introduce them to AI so that they understand its potential, and they will create realistic, successful, and valuable use cases to improve throughput, save time, stay competitive and grow their technological capabilities.