Sound Anomaly Detection Using Non-Compression CNN Reconstruction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing anomaly detection systems face challenges in detecting anomalies in older products without the necessary detection functions, requiring new product purchases to incorporate these capabilities.

Innovation Solution

A sound anomaly detection apparatus and method using a non-compression convolutional neural network that generates a restored value from an input value without dimension reduction or expansion, enabling anomaly detection across various devices by improving time-series information processing and computational efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a new product with anomaly detection function is purchased, then anomaly detection capability is provided, but cost increases and old products cannot be upgraded

Engineering Contradiction:

Improveanomaly detection capabilityVSAvoidapplicability to old products

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The anomaly detection apparatus is designed to be universally applicable across different product types and generations. The system uses a general-purpose audio unit and data processing unit that can be installed on various devices, making the anomaly detection function adaptable to both new and old products without requiring product-specific customization

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Productivity

If dimension reduction is applied in the neural network, then computational efficiency improves, but time-series information processing capability deteriorates

Engineering Contradiction:

Improvecomputational efficiencyVSAvoidtime-series information

Core Design Contradiction:

ProductivityVSLoss of information

Solution Approach 1:

Instead of reducing dimensions, the patent uses dimensionality expansion by applying convolution operations across the time-series dimension. The 1D convolutional layers process the temporal sequence while maintaining the original feature dimension, and the 2D convolutional layers operate on the time-frequency representation, preserving time-series information while achieving computational efficiency through localized feature extraction

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS12535354B2Apparatus and method for sound anomaly detection based on non-compression convolutional neural network

Publication Date: 2026.01.27 SK PLANET CO LTD

AI summary

According to the present invention, a sound anomaly detection method includes acquiring, by an audio unit, a noise from an inspection target; generating, by a data processing unit, an input value that is a feature vector matrix including a plurality of feature vectors from the noise; generating, by a detection unit, a restored value imitating the input value through a detection neural network that is a deep neural network learned for the input value; determining, by the detection unit, whether a restoration error indicating a difference between the input value and the restored value is greater than or equal to a calculated reference value; and determining, by the detection unit, that there is an anomaly in the inspection target when determining that the restoration error is greater than or equal to the reference value.