2025 ECerS - FIRE Summer School - Ivan Kraljevski

 Cognitive Material Diagnostics

Ivan Kraljevski

Fraunhofer-Institut für Keramische Technologien und Systeme IKTS

Cognitive Material Diagnostics is an emerging interdisciplinary field that integrates materials science, non-destructive testing (NDT), and machine learning to revolutionize the assessment of material quality and structural integrity.

At Fraunhofer IKTS, the KogMatD group applies advanced data analysis and machine learning techniques to develop intelligent diagnostics for a broad range of industrial applications. Due to the scarcity and complexity of data in many real-world settings, it is not always feasible to deploy sophisticated deep learning models. High-dimensional, costly, or sparse datasets often limit the achievable accuracy. For instance, some industrial systems produce insufficient data over time, making it impractical to train conventional machine learning models.

We demonstrate how machine learning, in combination with signal processing of sensor data, enables fast, automated, and reliable quality control and predictive maintenance in industrial environments.

This work presents use cases involving diverse NDT methods and sensor modalities—including audio, vibration, ultrasonics, and imaging—paired with machine learning approaches tailored for small datasets. For example, Acoustic Resonance and Acoustic Emission testing were employed to detect defects in pharmaceutical glass bottles.

Sintered cogwheels were inspected using acoustic resonance combined with small-data machine learning algorithms. Acoustic emission was utilized for condition monitoring of microfluidic valves using piezoelectric sensors, while convolutional autoencoders estimated health indicators to support remaining useful life prediction. Additional use cases include ultrasonic testing of adhesive bond quality in composites, anomaly detection via sensor networks for aerospace materials, inline monitoring of battery electrode lamination processes, and even non-NDT applications such as assessing the softness of tissue paper via acoustic emission during specimen tearing.

Through these efforts, the KogMatD group is advancing material diagnostics toward adaptive, intelligent systems capable of continuous learning and integration into real-world production environments.

Back to the previous page