Machine Fault Prediction Using Sensor and Control Software Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fault prediction methods for industrial machines are inadequate in complexity and accuracy, especially as machines become more sophisticated, leading to a need for a more advanced method to detect and predict faults accurately.
Innovation Solution
A machine learning device that observes state variables from sensors and internal control software data, learns fault conditions using a training dataset, and predicts faults by determining the occurrence and degree of faults, with the ability to update based on time-weighted determination data and share learning results across multiple machines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional fault prediction methods are used, then the system is simple and easy to implement, but the accuracy and reliability of fault prediction deteriorates as machine complexity increases
Solution Approach 1:
The patent transforms fault prediction from a static threshold-based approach to a dynamic machine learning approach. The system collects multiple state variables (vibration, temperature, pressure, etc.) and uses neural networks to learn optimal prediction parameters dynamically, allowing accurate fault detection even as machine complexity increases. The learning unit continuously adapts parameters based on training data from normal and faulty states.
Solution Approach 2:
The patent introduces a machine learning device as an intermediary between the complex industrial machine and the fault prediction output. This intermediary processes multiple sensor inputs, internal control data, and operational parameters through learned models to produce accurate fault predictions, simplifying the relationship between complex machine states and fault detection while maintaining high accuracy.
2Reliability
If conventional threshold-based methods are used, then the implementation is simple, but the ability to detect faults in complex circumstances deteriorates
Solution Approach 1:
The patent segments the fault prediction system into distinct functional modules: state observation unit for data collection, learning unit for model training, and prediction unit for fault detection. Each module handles specific tasks, allowing the system to achieve high reliability through specialized processing while managing overall system complexity through modular architecture.
Solution Approach 2:
The patent implements preliminary learning during normal operation phases, where the learning unit trains prediction models using data from normal and faulty states before actual fault prediction is needed. This preliminary action prepares the system with pre-trained parameters and patterns, enabling reliable fault detection when actually needed without requiring complex real-time analysis during critical moments.
3Measurement precision
If predetermined procedures are used for fault prediction, then the process is straightforward, but the adaptability to actual circumstances and accuracy deteriorates
Solution Approach 1:
The patent implements continuous feedback loops where prediction results and actual fault outcomes are fed back to the learning unit. When faults are detected or prevented, the system learns from these outcomes and updates its prediction models accordingly. This feedback mechanism enables the system to continuously improve accuracy and adapt to actual operating circumstances, moving beyond static predetermined procedures.
Solution Approach 2:
The patent transforms the fault prediction system from a static predetermined procedure to a dynamic adaptive system. The learning unit continuously updates prediction models based on new data, changing parameters and thresholds dynamically rather than relying on fixed predetermined values. This dynamic approach allows the system to adapt to varying operating conditions and improve accuracy over time.
Data Source
AI summary
An anomality prediction system, which predicts an anomality of a machine, includes: one or more memories; and one or more processors configured to: obtain a state variable including at least one of output data from at least one sensor that detects a state of at least one of the machine or a surrounding environment, internal data of control software controlling the machine, or computational data obtained based on at least one of the output data or the internal data; generate, by inputting the obtained state variable into a machine learning model, a degree of anomality of the machine based on output from the machine learning model; and notify information based on the generated degree of anomality, wherein the notified information includes at least one of the generated degree of anomality at one or more time points, or one or more levels of anomality based on the generated degree of anomality.