AUTOMATED MATERIAL CLASSIFICATION BASED ON TEMPERATURE PROFILES

This article presents a method for automated classification of materials based on the analysis of their temperature profiles.The use of laser thermography in combination with deep learning algorithms allows contactless material type recognition with high Accuracy.This is especially relevant for robotic systems that perform manipulation operations, automated quality control, and technical inspection without physical contact with the object.A review of previous studies has shown that existing methods of material classification have certain limitations.Visual methods can be unreliable due to the dependence on lighting conditions and the similarity of the appearance of different materials.Contact methods provide high Accuracy, but they are challenging to implement and require physical interaction, which is not always appropriate in robotics.Therefore, the thermographic approach is promising, as it determines the type of material by its response to heating and cooling.This paper describes an experimental system that uses a laser beam to heat the surface of an object, after which a thermal imager records the temperature change over time.The resulting temperature profiles are analysed using machine learning methods.The process of heat transfer for four types of materials (wood, plastic, steel, and aluminum) was simulated in COMSOL Multiphysics, which allowed us to form a training dataset.Four neural network architectures were tested for temperature profile classification: Feedforward, LSTM, Bi-LSTM, and 1D-Convolutional.The Feedforward network demonstrated the best results, which achieved a correct answer rate of 76.6 %, although 1D-Convolutional showed better classification of certain materials.It was found that the temperature profiles of some materials have significant similarities, which complicates the classification, so further research should be aimed at expanding the training data set and optimising the model architecture.The proposed approach has a wide range of applications in industry and robotics, requiring rapid identification of materials without physical contact.An automated system for classifying materials based on temperature profiles can increase the efficiency of technological processes, improve production safety, and expand the capabilities of autonomous robotic systems.