TeraView contributes to new UK research demonstrating AI-powered terahertz metrology for pharmaceutical coatings

TeraView, the pioneer and leader in terahertz technology and solutions, has published UK research with collaborators demonstrating how physics-informed artificial intelligence can improve non-destructive measurement of pharmaceutical tablet coatings using terahertz technology.

The study, published in the journal Nature’s Scientific Reports¹, demonstrates that a deep learning model trained entirely on synthetic terahertz data can be applied directly to experimental reflection mode measurements, without experimentally labelled training data or post measurement calibration. The research involved collaborators from the Universities of Liverpool, Warwick, Cambridge and TeraView.

The approach combines electromagnetic modelling, domain randomisation and artificial intelligence to analyse complete terahertz waveforms and estimate coating properties, including film thickness and refractive index. This enables quantitative analysis of thin coatings where conventional peak finding methods can become unreliable because reflections from adjacent interfaces overlap.

The research used a commercial TeraView terahertz pulsed imaging system to inspect pharmaceutical tablets produced during production scale coating trials. The model delivered accurate predictions across the studied range and closely matched conventional analysis on independently measured tablet batches, while continuing to estimate coating thickness where peak based methods reach their practical limits.

This approach could also be applied across TeraView’s other core areas, including automotive coating inspection and battery production, where terahertz technology enables non-destructive measurement of critical coating properties such as thickness, density and conductivity.

Dr Phil Taday, Head of Research at TeraView, commented “This work shows the value of combining terahertz measurement science with ‘physics-informed’ AI. By training the model on realistic synthetic data, including the practical variation encountered in reflection measurements, we can transfer it directly to experimental data without the costly and often impractical requirement for extensive labelled datasets. That is an important step towards robust, deployable AI-assisted metrology.”

Dr Don Arnone, TeraView’s CEO and Co-Founder, commented “This publication demonstrates how TeraView is working with collaborators to build AI into measurement workflows in a way that remains grounded in the underlying physics. Our focus is on translating advanced sensing and data science into practical industrial tools that can provide richer information, improve confidence in decisions and help customers address measurement challenges that conventional approaches cannot easily solve.”