Remote Infrastructure Defect Detection With Edge AI Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Monitoring infrastructure assets located in remote areas is challenging due to difficulties in accessing these locations, limited network bandwidth, and high power and computing resource constraints, which hinders effective detection and classification of defects such as corona discharge in electrical infrastructure.
Innovation Solution
Implementing edge computing with a scaled-down AI model, using an autoencoder to compress data on-site and transmit a binary representation to a centralized server for full classification, optimizing power and computing resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full AI classification models are deployed at remote edge devices, then defect detection accuracy is improved, but power consumption and computing resource demands increase beyond available constraints
Solution Approach 1:
The AI model is segmented into two parts: an encoder deployed at the remote edge device that extracts features and generates compressed binary representations, and a decoder/classifier deployed at the centralized server that performs full classification. This segmentation allows accurate defect detection while distributing computational load to match available resources at each location.
Solution Approach 2:
A binary feature representation acts as an intermediary between the edge device and centralized server. The encoder transforms audio signals into compressed binary features that retain essential defect information, enabling accurate classification at the server without requiring full-model deployment at the resource-constrained edge device.
2Measurement precision
If full AI classification models are deployed at remote edge devices, then defect detection accuracy is improved, but device complexity and resource requirements increase beyond available constraints
Solution Approach 1:
The AI model is segmented into two parts: an encoder deployed at the remote edge device that extracts features and generates compressed binary representations, and a decoder/classifier deployed at the centralized server that performs full classification. This segmentation allows accurate defect detection while distributing computational load to match available resources at each location.
Solution Approach 2:
The computationally intensive decoder and classifier components are extracted from the edge device and deployed only at the centralized server. The edge device retains only the lightweight encoder, significantly reducing device complexity and resource requirements while maintaining detection accuracy through server-based classification.
3Ease of operation
If audio signals are transmitted to centralized servers for processing, then network bandwidth requirements increase, but local processing capabilities are insufficient
Solution Approach 1:
The encoder at the edge device performs preliminary processing of audio signals by extracting features and generating compressed binary representations before transmission. This preliminary action reduces the data volume requiring network transmission while ensuring essential defect information is preserved for accurate server-side classification.
Solution Approach 2:
The audio signals are transformed from raw waveforms into compressed binary feature representations, changing the data format and reducing its volume. This parameter change enables efficient network transmission while maintaining the essential information needed for accurate defect detection at the centralized server.
Data Source
AI summary
Methods and systems for detecting conditions on remotely located infrastructure assets include: recording audio signals emitted by an asset using a sensor device proximate the asset; transforming the recorded audio signals into a spectrogram, the transformation performed by an edge computing device proximate the asset; encoding the spectrogram to generate a binary feature representation of the recorded audio signal, the encoding performed by the edge computing device using an encoder component of an autoencoder; transmitting the binary feature representation to a remote computing system located remotely from the asset; decoding the binary feature representation so as to regenerate the spectrogram, the decoding performed by the remote computing system using the decoder component of the autoencoder; and classifying a latent variable space of the regenerated spectrogram to detect whether the recorded audio signals indicate a condition of the asset, the classifying performed by a classifier hosted on the remote computing system.