Temporal Convolution-based Hybrid Model Approach with Representation Learning for Real-Time Acoustic Anomaly Detection

TL;DR

This research introduces an innovative approach to Real-Time Acoustic Anomaly Detection that combines semi-supervised temporal convolution with representation learning and a hybrid model strategy with Temporal Convolutional Networks (TCN) to handle various intricate anomaly patterns found in acoustic data effectively.

Abstract

The early detection of potential failures in industrial machinery components is paramount for ensuring the reliability and safety of operations, thereby preserving Machine Condition Monitoring (MCM). This research addresses this imperative by introducing an innovative approach to Real-Time Acoustic Anomaly Detection. Our method combines semi-supervised temporal convolution with representation learning and a hybrid model strategy with Temporal Convolutional Networks (TCN) to handle various intricate anomaly patterns found in acoustic data effectively. The proposed model demonstrates superior performance compared to established research in the field, underscoring the effectiveness of this approach. Not only do we present quantitative evidence of its superiority, but we also employ visual representations, such as t-SNE plots, to further substantiate the model's efficacy.

Figures (10)

• Figure 1: Defining high-level intuition of anomaly detection into a formalized model using probability theory concepts mentioned in Equation 1.1 with support of two aspects which are "concept of normality" and "deviations or exception of the data ."(1) Defines the hypothetical data space which contains both normal and anomaly instance. Outside the context(cloud) depicted the area which is irrelevant to the considered domain. (2) Defines the identified normal data distribution and the rest of space inherently becomes the region of unknown/not accepted with normal(anomaly) data region which is denoted in red on (3). The threshold level(size of the blue region) is highly domain-specific and needs delicate work to determine it.
• Figure 2: Layer breakdown of data in each design stage of Real-Time Acoustic Anomaly Detection problem using the proposed Deep Hybrid Model
• Figure 3: Implemented Real-Time Inference user interface mentioned in the below diagram. It creates a Mel-Spectrogram in real-time with a sliding windowing technique and feeds into the Python backend.
• Figure 4: TCN Architectures Overview: it is never the less a dilated hierarchical convolution layer network which provides recurrent neural capability with lesser parameters
• Figure 5: Anomaly Score Calculation with respect to the Hypothesis of Modelling Normal of a Data Distribution