AI is more than a model: Four steps to success in engineering and scientific AI applications

Most AI projects focus on the AI model, and beginners usually spend a large percentage of their time learning to develop and fine-tune AI models.

However, in engineering and scientific applications, AI is usually only a small piece of a larger system. The AI model needs to work correctly in all scenarios with all other working parts of the product, including sensors, actuators, and traditional algorithms such as control, signal processing, and image processing.

AI as a workflow, not just model development

Engineers and scientists have an inherent knowledge about the problem they are trying to solve with their domain expertise, and with the right tools they can get started even if they’re not AI experts. Tools designed for engineers and scientists to answer technical and business problems can help them succeed. Ultimately, engineers are at their best when they can focus on what they do best and build on it with the right resources to help them bring AI into the picture.

In this scenario, engineers should see AI as a workflow and not as model development alone.

Step 1: Data Preparation

Data preparation is arguably the most important step in the AI workflow: it may be one of the most time-consuming steps.

If an engineer or scientist gives the model ‘bad’ data, he or she will not get insightful results—and will likely spend many hours trying to figure out why the model is not working.

So, the natural step is to begin with clean, labeled data, as much as you can gather. It is better to spend adequate time here rather than trying to tweak machine learning models when they do not work to expectations. This step often involves multiple iterations between interactive exploration and programmatic processing of data to get to the appropriate data (and labels) needed for training AI models.

For example, an optic fiber manufacturing company leveraged the superior data processing capabilities of MATLAB to optimize data coming in from various locations and in various formats and integrated with data from other specialized engineering simulation software to prepare a robust data processing pipeline that could be fed into a machine learning model and deployed for 24x7 use in production.

Step 2: AI Model Development

The goal of this stage is to create a robust, accurate model that can make intelligent decisions from the input data.

A key factor for success here involves being able to easily try out multiple approaches across traditional machine learning and modern deep learning techniques. Another important aspect is the ability to explain the model and interpret the results. This is critical in engineering and scientific applications to understand the range of reliable behavior, identify failure modes, and enable smoother integration with the broader system.

In addition to algorithms and prebuilt models, engineers and scientists can find the best approach for their specific problem by using examples. MATLAB provides hundreds of examples for building AI models across multiple domains and application areas.

Step 3: Simulation and Test

AI models exist within a larger system and must work with all other pieces in the system.

To build the required level of accuracy and robustness prior to deployment, engineers must ensure that the model will respond the way it is supposed to in various situations. Engineers and scientists can accelerate the testing phase by integrating the models early in the development process and using simulation to model the behavior of the environmental and physical aspects of the system to characterize the overall system performance.

Trust can be achieved once you have successfully simulated and tested various cases you expect the model to see and can verify that the model performs on target. By using tools like Simulink that enable multi-domain modelling and can integrate with a variety of specialized simulation tools, engineers can verify that the model works as desired for all the anticipated use cases, avoiding redesigns that are costly both in money and time.

Step 4: Deployment

Once you are ready to deploy, you need to ready the model in the designated environment. This step typically requires design engineers to share an implementation-ready model and the environment can range from desktop to the cloud to embedded devices such as MCUs and FPGAs.

It is useful to work with tools like MATLAB that generates deployable components and code automatically. These tools will help engineers deploy their model across a variety of environments without having to rewrite the original code.

Take the example of deploying a model directly to a GPU: Automatic code generation eliminates coding errors that could be introduced through manual translation and provides highly optimized CUDA code that will run efficiently on the GPU.

In Summary

• AI in engineering and science is not just about models.
• The right workflow to follow is – Data Preparation, AI Model Development, Simulation and Test, and Deployment
• Data Preparation – Focus on clean, labelled input data.
• AI Model Development – Look for prebuilt models and examples, as this is an iterative step. Simulation and Test – Verify the model for all the anticipated use cases using simulation, avoiding redesigns.
• Deployment – Leverage tools that automatically generate deployable components and code rather than spending time on rewriting the code.

Dr Amod Anandkumar, Application Engineering Manager at MathWorks leads a team of application engineers helping clients across industries successfully adopt and implement technologies like AI, automated Driving, wireless communications. You can reach him at aanandku@mathworks.com

Johanna Pingel is a Product Marketing Manager at MathWorks. She focuses on machine and deep learning applications and making AI practical, entertaining, and achievable. You can reach her at jpingel@mathworks.com