PSI - Issue 64

Nikhil Holsamudrkar et al. / Procedia Structural Integrity 64 (2024) 580–587 Holsamudrkar Nikhil et al./ Structural Integrity Procedia 00 (2019) 000 – 000

581

2

* Corresponding author. Tel.: +91-22-25764343 ; fax: +91-22-25767302. E-mail address: nikhil.nnh@gmail.com 1. Introduction

Fiber Reinforced Polymers (FRPs) represent a class of composite materials comprising high-strength fibers, such as carbon, glass, or aramid, embedded within a polymer matrix, typically epoxy or polyester resin. But FRPs have limitations like non-compatibility with substrate, non-moisture breathability, high costs, etc. Conversely, Fiber Reinforced Cementitious Matrix (FRCM) systems represent a distinct paradigm wherein carbon, glass, or steel fabrics are incorporated within a cement-based matrix to confer enhanced mechanical properties and durability. Over a decade of research, FRCM has proven effective in various strengthening applications (Holsamudrkar et al., 2023b; Marcinczak et al., 2019). Other methods, such as pre-impregnating FRCM fabric with epoxy to mitigate fiber-fiber slip, have converted conventional FRCM into FRP mesh embedded within the cementitious matrix (Holsamudrkar et al., 2024). Non-Destructive Testing (NDT) methods are indispensable for ensuring the integrity of various structural members without damaging them. However, NDT requires frequent visits to the site, and it is highly unlikely that actual damage progression will be captured at the time of NDT. Therefore, Structural health monitoring (SHM) comes into the picture, wherein sensors are placed permanently on the structure under scrutiny. Acoustic emission (AE) is a popular passive SHM technique that utilizes the detection and analysis of stress waves emitted by active defects, providing real-time monitoring capabilities crucial for early anomaly detection. This technique offers insights into material behavior under varying loads, facilitating informed maintenance and design improvement decisions. In past research, AE-based health monitoring has been implemented for FRP-strengthened structural elements (Degala et al., 2009) and even for FRCM-strengthened members (Pohoryles et al., 2017). A study by Reboul et al., 2021, focused on the different failure types within the FRCM system and was able to correlate the same to the frequency content. AE-based SHM generates huge amounts of waveform data for every micro-structural damage known as hits. Generally, features extracted from these waveforms, such as hits, counts, amplitude, absolute energy, average frequency, partial power, etc., are utilized to analyze and interpret results. Therefore, many researchers use ML to analyze and interpret complex results. Other applications of ML, including data-driven models to predict the FRCM material properties, were explored extensively by some authors (Holsamudrkar et al., 2023a). The application of ML to AE data for characterizing various frequency contents corresponding to different damage mechanisms was implemented by Mandal et al., 2022. This method is common in a variety of literature and implementable in practical applications. However, the interaction of damage mechanisms in multi-part composites, such as retrofitting or strengthening applications, creates complexity in SHM data. ML-based models can be very handy in such cases. A deep convolutional network (CNN) model is developed in the present study and trained using waveform data of individual damage mechanisms. The data is collected with separate experiments facilitating particular failure modes. The waveform data is transformed into images using discrete wavelet transform (DWT). This model is validated with exclusively separate data. Further work includes implementing this model for full-scale strengthened members using FRCM. 2. Test Setup, Data Collection and Processing The experimental study involves FRCM mesh and steel rebar under direct tension, lap-shear tests on FRCM mesh attached to concrete surfaces, and concrete/mortar beams subjected to four-point bending. The experimental setup is illustrated in Fig. 1, while details about the tests can be found in Table 1. Only two specimens are considered for concrete and mortar flexural beams, as they generate huge data due to micro-cracking, crack coalescence, and macro cracking activity. AE-based monitoring was conducted using the 8-channel PCI-based Micro II Express 8 system by the Mistras group. The study employed sensors with a 20-1000 kHz working frequency range. The signals were pre-amplified with a 40dB gain, and a sampling rate of 2MSPS was selected as per the Nyquist – Shannon sampling theorem (Shannon et al., 1949). A substantial number of samples are tested to gather a comprehensive dataset of waveforms associated with a specific type of failure.