Shovel Abnormality Detection Using Maintenance-Filtered Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality determination models for working machines may use inappropriate normal operation data as training data, leading to inaccurate abnormality detection.
Innovation Solution
A shovel management system that collects and uses appropriate operation data ranges to determine the presence or absence of abnormalities by generating an abnormality determination model, incorporating input data from maintenance start and completion dates, and utilizing a training part to establish a relationship between operation data and shovel abnormalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If normal operation data is used as training data for abnormality determination models, then the model can be trained with readily available data, but the accuracy of abnormality detection deteriorates due to inclusion of inappropriate data
Solution Approach 1:
The system performs preliminary actions by collecting maintenance history data before training the abnormality determination model. The training data selection unit uses this pre-collected maintenance information to filter and select only appropriate training data periods, excluding periods when abnormalities were present. This preliminary data preparation ensures high-quality training data without requiring complex real-time analysis during model training.
Solution Approach 2:
The training data selection unit acts as an intermediary between the raw operation data and the abnormality determination model. It uses maintenance history as a mediator to identify and select only the appropriate portions of operation data for training, thereby filtering out inappropriate data while maintaining a relatively simple overall system architecture.
2Quantity of substance
If operation data from all periods is used as training data, then the quantity of training data increases, but the quality deteriorates due to inclusion of data from periods with abnormalities
Solution Approach 1:
The system performs preliminary filtering of training data by comparing operation data periods with maintenance history periods. The training data selection unit identifies and excludes operation data collected during periods when abnormalities were detected, ensuring that only high-quality training data from normal operation periods is used, thus maintaining both adequate quantity and high quality.
Solution Approach 2:
The system applies local quality control by selectively including or excluding specific time periods of operation data based on their quality. Instead of treating all operation data uniformly, the training data selection unit identifies high-quality periods (when no abnormalities were present) and uses only those for training, thereby ensuring local quality optimization in the training dataset.
Data Source
AI summary
A shovel management device includes a training part to determine a relationship between operation data of the shovel and an abnormality of the shovel by using, as training data, a data set that includes, among the operation data indicating the operation of the shovel, the operation data corresponding to a period for determining a presence or absence of the abnormality of the shovel as an input and information indicating the absence of the abnormality as an output.