← Back to Gallery

Deep Learning-based Approaches to Assess Waveform Quality for Engineering Applications

Due to rapid advancement in recording capabilities, the number of recorded waveforms has increased substantially, and this trend is projected to continue in the coming years. However, scaling conventional methods of waveform quality assessment to meet the demand of such datasets presents significant challenges, motivating the development of computationally efficient and scalable solutions. Deep learning (DL)-based approaches provide a promising solution to automate waveforms quality assessment over large databases by extracting the characteristics of high and low-quality waveforms. Despite their potential, limited work has explored the suitability of DL-based approaches for this application, where a significant knowledge gap exists on (1) algorithm selection, (2) efficient training workflows, (3) preparation of labeled data to support model training, and (4) the generalizability of trained model across different databases.

This study investigates the development and performance of various DL-based models for waveform quality classification. Using two waveform databases from New Zealand, the study evaluates the effectiveness and generalizability of convolutional and residual neural networks in this classification task. Additionally, different pre-processing and augmentation techniques, including over-sampling for data imbalance and record rotation for augmentation, were tested to understand the best practices for model training. Results show that a residual neural network-based architecture achieves accuracy and F1 scores exceeding 90%, where the same model retains its accuracy, to a lesser extent, when applied to a different waveform database from the same geographic region. Nevertheless, improving model generalizability and performance requires future efforts to establish a comprehensive training database supporting both DL-based and traditional algorithmic approaches.

Session: Modern Waveform Processing and Engineering Datasets - Accessibility, Quality Control, and Metadata - I

Type: Oral

Room: Key Ballroom 11

Date: 4/17/2025

Presentation Time: 05:30 PM (local time)

Presenting Author: Mohsen Zaker Esteghamati

Student Presenter: No

Invited Presentation: 

Poster Number:

Additional Authors