ML Inspection Pipeline With Selective Decoding for Abnormal Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for inspection targets require high processing loads due to all data being subjected to decoding processing, leading to inefficiencies.
Innovation Solution
Employing separate encoder and decoder models where the encoder outputs an abnormality degree and the decoder restores data only when necessary, reducing overall processing load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all data is input to the encoder model and subjected to decoding processing using the decoder model, then comprehensive inspection coverage is achieved, but processing load becomes excessively large
Solution Approach 1:
The inspection process is segmented into two distinct stages: first, the encoder model processes all inspection target data to extract features and determine abnormality degrees; second, only data flagged as abnormal by the encoder is subjected to decoding processing by the decoder model. This segmentation eliminates unnecessary decoding operations on normal data, significantly reducing processing load while maintaining comprehensive inspection coverage through the encoder's preliminary screening.
Solution Approach 2:
Instead of applying decoding processing to all data (excessive action), the system applies decoding only to the subset of data identified as abnormal by the encoder (partial action). This partial application of the decoder ensures that inspection resources are concentrated on cases that truly require detailed analysis, reducing overall processing load while maintaining sufficient inspection reliability.
2Manufacturing precision
If the decoder model processes all input data, then complete data restoration is achieved, but computational resources are wasted on normal data
Solution Approach 1:
The encoder model performs preliminary processing on all inspection target data to extract features and assess abnormality degrees before the decoder processes any data. This preliminary action filters out normal data, ensuring that the decoder only processes data that truly requires restoration, thereby avoiding waste of computational resources on normal data while maintaining restoration quality for abnormal data.
Solution Approach 2:
The encoder extracts and identifies abnormal data from the complete dataset, separating it from normal data. Only this extracted abnormal subset is then passed to the decoder for restoration processing. This extraction mechanism ensures that computational resources are allocated only to data that needs restoration, reducing energy consumption while preserving restoration quality for the relevant subset.
Data Source
AI summary
An encoder model and a decoder model, which are machine learning models, are used for inspection of the inspection target. The encoder model is a model in which inspection target data is input and an abnormality degree of the inspection target is output. The decoder model is a model in which an OK/NG value, which is a value indicating whether the inspection target is normal or abnormal, and a feature amount of data of the inspection target are input, and when the input OK/NG value indicates an abnormality, restored data of the inspection target based on the input feature amount is output.