Machine Failure Prediction Using Sensor Transitions and Control Commands
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for predicting machine tool failures are not accurate enough, as they rely solely on sensor information that may change synchronously with control commands, making it difficult to distinguish between normal and abnormal variations.
Innovation Solution
A training program that converts time-series numerical values from sensor data into character strings representing time-series transitions, and trains a machine learning model using both control command data and failure indication labels to predict machine tool failures with high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning is performed using only sensor information, then the system is simple to implement, but the prediction accuracy is insufficient because sensor values change synchronously with control commands making it difficult to distinguish normal from abnormal variations
Solution Approach 1:
The patent segments the input data into two distinct components: control command data and sensor information. By separating these data sources and processing them independently before integration, the system can identify correlations between control commands and sensor variations, thereby improving failure prediction accuracy while maintaining manageable processing complexity through modular data handling
Solution Approach 2:
The patent adds a new dimension to the data analysis by incorporating control command information as an additional input layer alongside sensor data. This dimensional expansion allows the machine learning model to distinguish between sensor variations caused by normal control operations versus those indicating actual failures, thereby improving prediction accuracy without proportionally increasing processing complexity
2Measurement precision
If only sensor data is used for training, then the data collection process is simple, but the model cannot distinguish between normal operational variations and abnormal failure signs
Solution Approach 1:
The patent applies preliminary action by collecting and storing control command data in advance alongside sensor information. This pre-captured contextual information about what control commands were issued during each measurement period enables the model to later distinguish between sensor variations that are normal responses to control commands versus variations indicating actual failures, thereby improving detection accuracy without losing critical context
Data Source
AI summary
An information processing apparatus converts a plurality of time-series numerical values included in sensing information of a sensor set in a machine or around the machine into a character string representing time-series transition. The information processing apparatus trains a machine learning model using training data that includes input data, which is generated based on a control command that controls the machine and information regarding the character string representing the time-series transition, and a label indicating whether or not abnormality is occurring in the machine.