Injection Molding State Determination Using Filtered Learning Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning methods for diagnosing injection molding machine abnormalities face challenges due to divergence in measured values and varying conditions, leading to inaccurate diagnoses and high costs in preparing diverse learning conditions, with unsuitable learning data often being used, resulting in incorrect operating state assessments.
Innovation Solution
A state determination device and method that acquires and processes data from injection molding machines, excluding unsuitable learning data through specific extraction conditions to improve machine learning accuracy, using a data acquisition unit, extraction condition storage, learning data extraction unit, and machine learning device for supervised, unsupervised, and reinforcement learning, generating a learning model for precise abnormality estimation and operational state determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning is performed using time-series data from injection molding machines with different specifications, resins, and incidental facilities, then the adaptability of the diagnosis system is improved, but the cost and time required for data collection and model preparation increase significantly
Solution Approach 1:
The patent extracts and excludes unsuitable learning data from time-series data based on predetermined conditions (alarm states, startup/shutdown periods, maintenance periods). This extraction of inappropriate data allows the system to use a broader range of machine configurations without being contaminated by data that would degrade model performance, thus improving adaptability without proportionally increasing data collection time.
Solution Approach 2:
The patent changes the parameter of data suitability by introducing extraction conditions that dynamically filter time-series data based on machine state parameters (alarm flags, operational phase indicators). This allows the same data collection process to yield high-quality learning data across diverse machine configurations without requiring separate data collection campaigns for each configuration.
2Quantity of substance
If all time-series data including data during alarm states and operational transitions is used for machine learning, then the quantity of learning data is increased, but the reliability of the diagnosis decreases due to inclusion of unsuitable data
Solution Approach 1:
The patent extracts and removes unsuitable data segments from time-series data based on extraction conditions such as alarm states, startup/shutdown periods, and maintenance periods. This allows the system to maintain a large overall data quantity while systematically excluding portions that would compromise diagnosis reliability, achieving both high data volume and high reliability.
Solution Approach 2:
The patent converts potentially harmful data (data collected during alarm states or transitional periods) into beneficial information by using the presence of these states as extraction conditions. The alarm flags and operational state indicators that initially mark problematic data are repurposed as useful filters to automatically identify and exclude unsuitable learning data, improving reliability without reducing overall data quantity.
3Manufacturing precision
If machine learning models are prepared for each specific machine configuration and resin type, then the manufacturing precision of the diagnosis is improved, but the device complexity and preparation cost increase
Solution Approach 1:
The patent creates a universal learning model that can handle multiple machine configurations, resin types, and incidental facilities simultaneously. By using extraction conditions to filter out configuration-specific anomalies rather than creating separate models for each configuration, the system achieves high diagnosis precision across diverse scenarios with a single multi-functional model, reducing device complexity while maintaining precision.
4Productivity
If data collection is performed without excluding periods of operational change and alarm states, then the productivity of data acquisition is improved, but the quality of learning data deteriorates
Solution Approach 1:
The patent implements self-service data filtering where the system automatically identifies and excludes unsuitable learning data segments using extraction conditions based on alarm flags and operational state indicators embedded in the time-series data itself. This automated self-filtering maintains high data acquisition productivity while ensuring learning data quality, as the system serves itself to identify and remove problematic data without manual intervention or separate verification processes.
Data Source
AI summary
A state determination device acquires data on an industrial machine, extracts data used for processing related to machine learning from the acquired data, out of the acquired data, according to an extraction condition for extracting the data, and executes the processing related to the machine learning using the extracted data.