System and method for maintenance recommendation in industrial networks

A non-transitory computer readable medium, storing instructions for executing process for managing a system comprising a plurality of apparatuses providing sensor data, the instructions comprising:

• determining a graph structure from the sensor data;
• detecting a plurality of communities in the graph structure;
• for each of the plurality of communities, conducting feature extraction to extract features regarding fault detection and isolation;
• and
• providing the extracted features to a machine learning process to generate a trained machine learning model configured to conduct fault detection and isolation for the system,
• wherein the conducting feature extraction comprises:
• conducting Graph Fourier Transform (GFT) on the graph structure to convert the graph structure to Fourier domain;
• extracting the features from the converted graph structure in the Fourier domain according to a filter;
• and
• converting the extracted features into time domain through applying an inverse GFT, and
• wherein the filter is a k-hop graph filter, and
• wherein the conducting feature extraction comprises conducting graph embedding to map the graph structure to a vector space domain for each of the plurality of communities, and training machine learning classifiers on the vector space domain each of the plurality of communities to detect and isolate faults in each of th plurality of communities.

A management apparatus configured to manage a system comprising a plurality of apparatuses providing sensor data, the management apparatus comprising:

• a processor, configured to:
• determine a graph structure from the sensor data;
• detect a plurality of communities in the graph structure;
• for each of the plurality of communities, conduct feature extraction to extract features regarding fault detection and isolation;
• and
• provide the extracted features to a machine learning process to generate a trained machine learning model configured to conduct fault detection and isolation for the system,
• wherein the processor is configured to conduct feature extraction by,
• conducting Graph Fourier Transform (GFT) on the graph structure to convert the graph structure to Fourier domain;
• extracting the features from the converted graph structure in the Fourier domain according to a filter;
• and
• converting the extracted features into time domain through applying an inverse GFT, and
• wherein the filter is a k-hop graph filter, and
• wherein the processor is configured to conduct feature extraction by:
• conducting graph embedding to map the graph structure to a vector space domain for each of the plurality of communities;
• and
• training machine learning classifiers on the vector space domain each of the plurality of communities to detect and isolate faults in each of the plurality of communities.