In situ investigation and detection of opto-mechanical properties of polymeric fibres from their digital distorted microinterferograms using machine learning algorithms

•We established an experimental dataset containing 400 microinterferograms.•This dataset includes the four classes for the dynamic investigation of the fibres.•To automate the classification for the classes, a fine-tuned pre-trained CNN is used.•We suggested a combination between AlexNet and the SVM and the KNN.•Cold drawing and creep experiments are carried out for iPP fibres using KNN network. During the dynamic investigation of opto-mechanical properties of the fibres, hundreds of microinterferograms that carry the fibres information may be produced. These microinterferograms may be classified into four main classes due to focus plane of CCD camera and the inclination of the fibre relative to the fringes. Each class of them has a specific analysis method according to its class. The manual classification for these microinterferograms is a complex process and needs a professionally expert in the field. So, the automatic classification and investigation for these classes are main goal of the present study. Hence, an experimental dataset, including 400 microinterferograms for the four classes is established. To perform this task, Mach–Zehnder interferometer with a stretched device is used. To automate the classification for the classes in our dataset, we fine-tuned a pre-trained AlexNet neural network. After training our dataset using this network, we obtained an 89.46% validation accuracy (VA). To increase the accuracy for the constructed network, two methods are proposed. The first method contains a combination between the AlexNet and the support vector machine and this network achieved an 94.11% VA. The second method includes a combination between the AlexNet and the K- Nearest Neighbor and an VA 96.64% is yielded. After the classification process, each microinterferogram is analyzed according to its class and the 2D refractive index of their fibres are investigated. Dynamic cold drawing and creep experiments are carried out for iPP fibres by considering AlexNet and the K- Nearest Neighbor network to give the relation between its refractive indices and strains.