Injection Molding State Diagnosis Using Normalized Time-Series Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current state determination methods for injection molding machines require significant cost and time to collect learning data due to differences in machine specifications and auxiliary equipment, leading to inaccurate diagnoses and high operational costs.
Innovation Solution
A state determination device and method that uses numeric conversion of time-series physical quantities through normalization or differentiation to estimate abnormality degrees, allowing machine learning to absorb differences in equipment and machine types, enabling accurate diagnosis without extensive data collection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning is performed using measured values from injection molding machines with different specifications, then diagnostic capability can be improved, but divergence between measured values and learning data becomes too large to perform correct diagnosis
Solution Approach 1:
The patent transforms the learning data by changing its parameters through normalization processing. Specifically, it converts actual measurement values into normalized values that represent rates of change or tendencies, thereby removing the influence of absolute magnitude differences caused by varying machine specifications. This allows the learning model to focus on pattern recognition rather than absolute value comparison.
Solution Approach 2:
The patent introduces normalized values as an intermediary between the raw measured values and the learning model. These normalized values serve as a common language that bridges different machine specifications, allowing measurements from machines of various sizes and configurations to be meaningfully compared and learned from a single model.
2Measurement precision
If learning data is collected from various injection molding machines to improve diagnostic accuracy, then machine learning performance can be enhanced, but cost and time for data collection increase significantly
Solution Approach 1:
Instead of collecting actual measurement data from multiple different machines, the patent creates normalized copies of the data that preserve the essential diagnostic patterns while removing machine-specific characteristics. This allows a single machine's data to be transformed into learning data that can represent multiple machine types without requiring physical data collection from each.
Solution Approach 2:
The normalization processing makes the learning data universal, enabling a single learning model to be applied across different injection molding machine specifications. This eliminates the need to collect separate learning data from each machine type, as the normalized representation captures universal diagnostic patterns applicable to all machines.
3Measurement precision
If learning data is collected from various injection molding machines to improve diagnostic accuracy, then machine learning performance can be enhanced, but cost for data collection increases significantly
Solution Approach 1:
The patent creates normalized copies of measurement data that can serve as learning data without requiring expensive data collection campaigns from multiple machines. The normalization transformation is a computational process that generates universally applicable learning data from existing measurements, eliminating the need for costly field data collection.
Solution Approach 2:
The system performs self-service by using its own measurement data, after normalization processing, as the learning data. This eliminates the need to externally collect data from other machines, reducing both cost and complexity. The single machine's data, when normalized, serves the dual purpose of operational data and training data.
Data Source
AI summary
A state determination device is capable of assisting maintenance for various injection molding machines. The state determination device acquires data related to an injection molding machine, performs numeric conversion for extracting a feature in a temporal direction or an amplitude direction, with respect to time-series data of physical quantity in the acquired data, and performs machine learning using the data obtained through numeric conversion so as to generate a learning model.