Scientific Artificial Intelligence

Levergaging the Best of Science and Artificial Intelligence

  1. Services /
  2. Scientific Artificial Intelligence

Intugent is a pioneer in combining material science and Artificial Intelligence. The scientific Artificial Intelligence (sAI) models are more accurate than the conventional Artificial Intelligence (AI) models and require less data for training the models.

Conventional AI has been around at least since 1955, but it has found commercial application only recently. These applications are in the areas of robotics, face and speech recognition, data mining, and banking software due to availability of large data sets and high-power computing (HPC) systems. Its applications in materials science applications are very limited if any due to lack of big data. In his article titled “AAAS: Machine learning 'causing science crisis' ”, P. Gosh, the science correspondent for the BBC news, reported that machine learning techniques used by thousands of scientists to analyze data are producing results that are misleading and often completely wrong. He quotes Dr. Allen of Rice University that the scientists need to improve their machine learning techniques.

In conventional AI, concentration of each material along with other process variables such as time-temperature profile are used as input. As shown in Figure 1, the predicted values of the material property are obtained as output of the AI model. Since the formulation components are chosen from a large pool of available raw materials, AI models require large data set for training.

The final material properties can also be predicted from math models based on material science. In these models, the formulation information is first converted to formulation descriptors (properties) using appropriate scientific correlations. Cohesive energy, glass transition temperature, crosslink density are a few examples of these descriptors. The material science math models are usually more accurate than AI models and can be used to predict properties when new materials are used. However, these models require a deep understanding of the underlying science and their development take long times.

The sAI models combine the best of AI models and scientific math models. Just like scientific math models, sAI models first convert the formulation information to formulation descriptors using scientific correlations. These descriptors are then used as input to sAI models. The sAI models employ scientific governing equations as a hidden layer. These hidden layers introduce a bias towards physically observed trends. As a results, sAI model predictions are more accurate than the AI models while requiring less data for training. A comparison of data needed for training and the efforts required to develop the model is shown in Figure 2. Development of sAI models do require more effort than AI models, but require less data for training, can handle new materials, and predict more accurate trends. However, commercially available AI software cannot handle a scientific governing equation as a hidden layer in the AI model.

Intugent is very proud to commercialize our Digitalized Innovation Process (DIP) software that has built in sAI capabilities. Each DIP software is custom developed for each partner and the material science industry. As more data is available, the sAI models can be retrained by our partners for improved accuracy.