Apparatus and method for generating learning data for artificial intelligence model

1. A learning data generation apparatus comprising:

a collection unit for collecting at least one among structured data and unstructured data; and

a processing unit for generating a learning data set for learning of an artificial intelligence model that predicts and diagnoses anomalies of a plant facility using at least one among the structured data and the unstructured data, wherein

the unstructured data is an external image of the plant facility acquired by an image sensor, and

when the learning data set is generated from the structured data, the processing unit extracts feature information of anomaly data by classifying the structured data into normal data corresponding to a normal pattern and anomaly data corresponding to an anomaly pattern by using a time domain analysis, extracts feature information of the anomaly data from the structured data through a frequency domain analysis in order to grasp a feature pattern that is unknown through the time domain analysis, analyzes a correlation between the anomaly data among the structured data by analyzing magnitude and frequency of occurrence by frequency for the extracted feature information of the anomaly data in consideration of a time domain and a frequency domain, and generates the learning data, after generating a plurality of matrix maps on the basis of the correlation between the analyzed anomaly data, by grouping the matrix maps by anomaly type, and

when the learning data set is generated from the unstructured data, the processing unit sets a region of interest in the external image, extracts features from the region of interest and stores the features in combination with location information of an object in the region of interest, classifies the extracted features into normal features and abnormal features, generates a labeling data set by performing labeling on the region of interest or the external image containing the anomaly features, and generates the learning data set on the basis of the labeling data set, and

wherein the processing unit generates the learning data set by fusing a learning data set based on the structured data and a learning data set based on the unstructured data.

2. The apparatus according to claim 1 , wherein the structured data includes at least one among temperature, vibration, pressure, voltage, current, and an image measured by a plurality of sensors installed in the plant facility during an operation of the plant facility.

3. A learning data generation method of a learning data generation apparatus, the method comprising the steps of:

collecting at least one among structured data and unstructured data, by the learning data generation apparatus; and

generating a learning data set for learning of an artificial intelligence model that predicts and diagnoses anomalies of a plant facility using at least one among the structured data and the unstructured data, by the learning data generation apparatus, wherein

the unstructured data is an external image of the plant facility acquired by an image sensor, and

when the learning data set is generated from the structured data, the step of generating a learning data set includes the steps of:

extracting feature information of anomaly data by classifying the structured data into normal data corresponding to a normal pattern and anomaly data corresponding to an anomaly pattern by using a time domain analysis;

extracting feature information of the anomaly data from the structured data through a frequency domain analysis in order to grasp a feature pattern that is unknown through the time domain analysis;

analyzing a correlation between the anomaly data among the structured data by analyzing magnitude and frequency of occurrence by frequency for the extracted feature information of the anomaly data in consideration of a time domain and a frequency domain; and

generating the learning data, after generating a plurality of matrix maps on the basis of the correlation between the analyzed anomaly data, by grouping the matrix maps by anomaly type, and

when the learning data set is generated from the unstructured data, the step of generating a learning data set includes the steps of:

setting a region of interest in the external image;

extracting features from the region of interest and storing the features in combination with location information of an object in the region of interest;

classifying the extracted features into normal features and abnormal features;

generating a labeling data set by performing labeling on the region of interest or the external image containing the anomaly features; and

generating the learning data set on the basis of the labeling data set, and

wherein the step of generating the learning data set includes a step of fusing a learning data set based on the structured data and a learning data set based on the unstructured data.

4. The method according to claim 3 , wherein the structured data includes at least one among temperature, vibration, pressure, voltage, current, and an image measured by a plurality of sensors installed in the plant facility during an operation of the plant facility.