Data processing apparatus and data processing method

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing abnormality detection techniques face challenges in accurately updating inference models when operation data deviates from the learning stage conditions, leading to potential misclassification of normal or abnormal data, which can result in ineffective abnormality detection.

Innovation Solution

A data processing apparatus that acquires operation data, determines if it satisfies application conditions, stores unsuitable data as pending, and updates the inference model only with data deemed suitable for update, ensuring accurate model refinement.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If operation data is acquired in a short period of time to generate the regression model early, then the abnormality detection can start earlier, but the learning data has few variations and cannot appropriately predict the normal range

Engineering Contradiction:

Improvetime to start abnormality detectionVSAvoidaccuracy of normal range prediction

Core Design Contradiction:

Loss of timeVSReliability

Solution Approach 1:

The system performs preliminary learning with initial operation data to generate a regression model before full operational data is available. This preliminary model enables early abnormality detection while acknowledging it will be refined later as more diverse operational data accumulates, resolving the tension between early detection and prediction accuracy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The regression model is designed to be dynamically updated as new operation data becomes available. The model transitions from an initial state with limited learning data to an improved state with more varied operational data, allowing the system to adapt its prediction accuracy over time while maintaining continuous abnormality detection capability.

Inventive Principle:

Principle #15Dynamics

2Reliability

If operation data is acquired in a long period of time to generate the regression model, then the learning data has sufficient variations, but abnormal operation data may be treated as normal data and reflected in the inference value

Engineering Contradiction:

Improveaccuracy of normal range predictionVSAvoidcontamination of learning data with abnormal data

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The system uses feedback mechanisms to monitor operation data quality and detect abnormal patterns. When abnormal operation data is identified, it is excluded from the learning dataset, preventing contamination of the regression model while still allowing the system to learn from sufficient variations in normal operational data over time.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system extracts and separates abnormal operation data from the overall dataset before generating the regression model. By taking out contaminated data points that represent abnormal conditions, the learning process only incorporates genuine normal operational variations, ensuring the inference model accurately represents normal behavior without being skewed by anomalies.

Inventive Principle:

Principle #2Taking out (Extraction)

3Adaptability or versatility

If operation data that deviates from learning stage conditions is simply added to the learning data, then the application range of the inference model is widened, but abnormal operation data is incorrectly treated as learning data

Engineering Contradiction:

Improveapplication range of inference modelVSAvoidaccuracy of abnormality detection

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system applies different quality standards to different operation data based on its characteristics. Operation data that meets the application conditions is processed with standard procedures, while data that deviates from learning stage conditions undergoes additional validation and filtering. This local quality approach allows the model to adapt to diverse operational scenarios while maintaining detection precision by preventing abnormal data from being misclassified as learning data.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS20240044539A1Data processing apparatus and data processing method

Publication Date: 2024.02.08 MITSUBISHI ELECTRIC CORP

AI summary

An operation data acquisition unit acquires from an air conditioning system, operation data of the air conditioning system. A model application determination unit determines whether or not the operation data satisfies an application condition for applying the operation data to an inference model for inferring status of the air conditioning system. Further, the model application determination unit stores the operation data into an operation data holding unit as pending operation data, when the operation data does not satisfy the application condition. A model updating unit determines whether or not the pending operation data is operation data suitable for an update of the inference model, and updates the inference model using the pending operation data, when it is determined that the pending operation data is the operation data suitable for the update of the inference model.